{"meta":{"query_hash":"e941b4fc72db","filters":{"venue":"Statistics in Medicine"},"cohort_total":708,"direct_labels_cover":5,"predictions_cover":708,"exported":708,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/e941b4fc72db","api":"https://metacan.xera.ac/api/v1/cohort?venue=Statistics+in+Medicine"},"results":[{"id":"W1541729297","doi":"10.1002/sim.4334","title":"Variance estimation and confidence intervals for the standardized mortality ratio with application to the assessment of a cancer screening program","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Colorectal Cancer Screening and Detection","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Centre hospitalier universitaire de Québec; Institut National de Santé Publique du Québec; Université Laval","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Confidence interval; Statistics; Standardized mortality ratio; Estimator; Statistic; Variance (accounting); Standard error; Medicine; Mathematics","score_opus":0.050999920586490166,"score_gpt":0.42048881365164414,"score_spread":0.369488893065154,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1541729297","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010851958,0.0018822401,0.9848328,0.0003175733,0.00009982677,0.0001243175,0.000178741,0.00038548306,0.0013270659],"genre_scores_gemma":[0.39607227,0.002205649,0.5970425,0.00029814267,0.00053720135,0.0013360406,0.0011566449,0.00032037232,0.0010312283],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9236067,0.058759872,0.0026303714,0.0043975483,0.009731405,0.00087409874],"domain_scores_gemma":[0.5281592,0.42395052,0.014304435,0.01775327,0.0150737725,0.000758797],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09324035,0.0015152646,0.0026072655,0.007486032,0.00075337116,0.0029764827,0.0036797535,0.0030545478,0.002605732],"category_scores_gemma":[0.4155309,0.0009470017,0.0026058448,0.0050684023,0.0029920123,0.0038222098,0.0029785496,0.0040279953,0.00041234816],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005854961,0.00015922386,0.031769864,0.0009008709,0.0018243819,0.0005574386,0.0009787906,0.23325585,0.0016545226,0.51161134,0.004560348,0.2121419],"study_design_scores_gemma":[0.0002109647,0.0003578813,0.012405424,0.00071152224,0.0004486622,0.0005582153,0.000269112,0.7798397,0.0020686535,0.196165,0.0067413584,0.000223537],"about_ca_topic_score_codex":0.0057135685,"about_ca_topic_score_gemma":0.0024870478,"teacher_disagreement_score":0.09324035,"about_ca_system_score_codex":0.0020507856,"about_ca_system_score_gemma":0.0017051721,"threshold_uncertainty_score":0.49310797},"labels":[],"label_agreement":null},{"id":"W1633467982","doi":"10.1002/sim.4326","title":"An informed reference prior for between‐study heterogeneity in meta‐analyses of binary outcomes","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Economic and Environmental Valuation","field":"Economics, Econometrics and Finance","cited_by":70,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; St. Joseph’s Healthcare Hamilton","funders":"","keywords":"Prior probability; Variance (accounting); Vagueness; Bayesian probability; Computer science; Inference; Bayesian inference; Econometrics; Meta-analysis; Statistics; Mathematics; Artificial intelligence; Fuzzy logic; Medicine","score_opus":0.5384232542332547,"score_gpt":0.40781236794008474,"score_spread":0.13061088629316991,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1633467982","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0024205,0.0021944044,0.9913958,0.0015391416,0.00019946325,0.00025751558,0.0003588079,0.00033964933,0.0012947313],"genre_scores_gemma":[0.14747545,0.0032669913,0.83949596,0.002503072,0.0006140044,0.0032322882,0.0016554411,0.00030187325,0.0014549111],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.83250713,0.13639058,0.008552004,0.008919804,0.012648205,0.0009823217],"domain_scores_gemma":[0.63328713,0.30555427,0.015618606,0.033237,0.01144107,0.0008619576],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.19546038,0.0023485504,0.005057243,0.007432294,0.0014960378,0.0063466663,0.0072253854,0.010536938,0.004670464],"category_scores_gemma":[0.5114664,0.0029571184,0.006401758,0.0068921703,0.004175882,0.009209388,0.0041602817,0.012551065,0.0017246954],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013471921,0.00022825402,0.0044676373,0.003908802,0.0041414285,0.0009918406,0.0024900062,0.18017253,0.0031519926,0.58055604,0.013529099,0.20501526],"study_design_scores_gemma":[0.0006019367,0.00045351023,0.0045219506,0.002573121,0.003017458,0.00077131524,0.00017453541,0.14710936,0.0030461878,0.8179579,0.019457083,0.00031568945],"about_ca_topic_score_codex":0.002067662,"about_ca_topic_score_gemma":0.00219718,"teacher_disagreement_score":0.8045396,"about_ca_system_score_codex":0.0031795127,"about_ca_system_score_gemma":0.0026583222,"threshold_uncertainty_score":0.9921406},"labels":[],"label_agreement":null},{"id":"W1748097147","doi":"10.1002/sim.6584","title":"Joint estimation of multiple disease‐specific sensitivities and specificities via crossed random effects models for correlated reader‐based diagnostic data: application of data cloning","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta; University of Calgary","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Random effects model; Computer science; Maximum likelihood; Bayesian probability; Maximization; Expectation–maximization algorithm; Statistics; Artificial intelligence; Marginal likelihood; Machine learning; Mathematics; Medicine; Mathematical optimization; Pathology","score_opus":0.19629008733233633,"score_gpt":0.402939802217801,"score_spread":0.20664971488546469,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1748097147","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.021836529,0.00020504947,0.97703326,0.00023735849,0.000022077105,0.00012115585,0.00017096975,0.00011202059,0.00026163645],"genre_scores_gemma":[0.4990153,0.0005949009,0.49533394,0.00039461444,0.00009516676,0.0011298645,0.0008168613,0.00013046578,0.0024888841],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.96702975,0.025848554,0.0011196956,0.0034392865,0.0018945694,0.00066809024],"domain_scores_gemma":[0.8228527,0.1555048,0.008732988,0.009005289,0.0031636718,0.0007405856],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07077909,0.0019197544,0.0025810723,0.0022593034,0.000646727,0.0030540598,0.0047420077,0.00279212,0.0021830301],"category_scores_gemma":[0.114474736,0.0018135753,0.0044269385,0.0027945389,0.0035425425,0.0031986847,0.0038352513,0.0033020834,0.00041052574],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006885303,0.00021493714,0.01800047,0.00022538929,0.0010379334,0.00079685426,0.0008867321,0.73478234,0.0021084473,0.18589266,0.0007118891,0.05465382],"study_design_scores_gemma":[0.0000832067,0.0001543025,0.0021106286,0.000041983527,0.00022243433,0.0001753767,0.000034617966,0.92875427,0.0007999427,0.06669501,0.00084230606,0.00008590874],"about_ca_topic_score_codex":0.008606748,"about_ca_topic_score_gemma":0.004946235,"teacher_disagreement_score":0.07077909,"about_ca_system_score_codex":0.002532372,"about_ca_system_score_gemma":0.0022044175,"threshold_uncertainty_score":0.3743201},"labels":[],"label_agreement":null},{"id":"W1873375728","doi":"10.1002/sim.6314","title":"EM for regularized zero‐inflated regression models with applications to postoperative morbidity after cardiac surgery in children","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre; Montreal Children's Hospital","funders":"National Center for Advancing Translational Sciences; National Institute of Environmental Health Sciences; National Heart, Lung, and Blood Institute; National Institute of Diabetes and Digestive and Kidney Diseases; National Cancer Institute; National Institutes of Health; Charles H. Hood Foundation","keywords":"Poisson regression; Count data; Poisson distribution; Expectation–maximization algorithm; Regression; Medicine; Likelihood function; Statistics; Overdispersion; Regression analysis; Zero-inflated model; Cardiac surgery; Model selection; Computer science; Econometrics; Mathematics; Maximum likelihood; Surgery; Population","score_opus":0.05210787311159458,"score_gpt":0.3740036816381332,"score_spread":0.3218958085265386,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1873375728","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0029491878,0.00021117364,0.99627703,0.00021191729,0.000017172693,0.000021713411,0.000046548634,0.000106219326,0.0001588997],"genre_scores_gemma":[0.17194909,0.0010915316,0.8214486,0.00044376773,0.0002475972,0.0007043567,0.00081978115,0.00029255004,0.003002661],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9962919,0.00285522,0.00015241612,0.00038564793,0.0002002706,0.00011464524],"domain_scores_gemma":[0.981215,0.016269935,0.0009793604,0.00056972273,0.00078088423,0.00018509106],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011086915,0.0012005589,0.001883834,0.0012821815,0.0005411872,0.0009037141,0.0030524617,0.0017060433,0.002115655],"category_scores_gemma":[0.032843444,0.0010457644,0.0017539031,0.0014980193,0.0014516708,0.0017685655,0.0019016683,0.0025839305,0.00054675626],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001264235,0.000048409893,0.002365843,0.00018457705,0.00016299481,0.00014717238,0.00012572063,0.8929557,0.00042201177,0.06306124,0.0018656437,0.038534287],"study_design_scores_gemma":[0.000014720822,0.000017916735,0.00015423312,0.000013967786,0.000010170875,0.000025730928,0.000010000364,0.9808611,0.00012713006,0.018119866,0.0006355397,0.000009508493],"about_ca_topic_score_codex":0.0045593022,"about_ca_topic_score_gemma":0.0033585187,"teacher_disagreement_score":0.011086915,"about_ca_system_score_codex":0.0011869724,"about_ca_system_score_gemma":0.0018188588,"threshold_uncertainty_score":0.058633864},"labels":[],"label_agreement":null},{"id":"W1909579684","doi":"10.1002/sim.6582","title":"A joint model for interval‐censored functional decline trajectories under informative observation","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ontario Institute for Cancer Research; University of Victoria","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Inference; Bivariate analysis; Computer science; Bayesian inference; Disease; Econometrics; Joint probability distribution; Statistics; Interval (graph theory); Bayesian probability; Process (computing); Random effects model; Bayes' theorem; Machine learning; Artificial intelligence; Medicine; Mathematics; Meta-analysis","score_opus":0.1438789696419815,"score_gpt":0.3567442275512428,"score_spread":0.2128652579092613,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1909579684","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.044292152,0.00061501435,0.95060325,0.0013980465,0.0000862014,0.00018233928,0.0010793189,0.0003615386,0.0013821264],"genre_scores_gemma":[0.7910633,0.0020866967,0.1837273,0.0006374282,0.000371508,0.0018628817,0.0030811713,0.0001944545,0.01697513],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9945311,0.002509549,0.00027660525,0.0014474094,0.0007688038,0.00046652238],"domain_scores_gemma":[0.97538245,0.019036157,0.0023277805,0.0014889543,0.0012496924,0.0005149338],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.015876794,0.0013226564,0.0032821982,0.003141099,0.0009199318,0.0026623341,0.0049572745,0.0037637472,0.0044823783],"category_scores_gemma":[0.03480616,0.0015744675,0.0027075242,0.00321932,0.002904689,0.0046345266,0.0022308761,0.004480415,0.0010658746],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00050638785,0.00024302214,0.015335558,0.00030633237,0.00045242562,0.0005487381,0.001282993,0.47923735,0.0017053331,0.45476934,0.0030193254,0.04259314],"study_design_scores_gemma":[0.000112836395,0.00014940114,0.003839761,0.000091453556,0.00018293824,0.00020877882,0.00007135171,0.866644,0.00027792703,0.12643045,0.0019087162,0.000082359904],"about_ca_topic_score_codex":0.015085921,"about_ca_topic_score_gemma":0.010890651,"teacher_disagreement_score":0.015876794,"about_ca_system_score_codex":0.0022843112,"about_ca_system_score_gemma":0.0025224343,"threshold_uncertainty_score":0.08396554},"labels":[],"label_agreement":null},{"id":"W1910761353","doi":"10.1002/sim.6529","title":"Bias in progression‐free survival analysis due to intermittent assessment of progression","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Survival analysis; Clinical endpoint; Progression-free survival; Estimator; Medicine; Kaplan–Meier estimator; Confidence interval; Endpoint Determination; Imputation (statistics); Tumor progression; Clinical trial; Statistics; Oncology; Econometrics; Overall survival; Internal medicine; Cancer; Mathematics; Missing data","score_opus":0.6145995339197039,"score_gpt":0.6510660572099151,"score_spread":0.03646652329021116,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1910761353","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04453982,0.012816948,0.9337927,0.0043983115,0.0004414364,0.00031305006,0.00026828476,0.00042372313,0.0030057405],"genre_scores_gemma":[0.7675542,0.0066271913,0.2164493,0.0034672408,0.001405571,0.0011186644,0.0004297857,0.00032954634,0.0026184216],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.85792536,0.12188594,0.0045199585,0.0058850236,0.008593058,0.0011906944],"domain_scores_gemma":[0.362814,0.5977965,0.01678603,0.017809028,0.0041357493,0.00065869483],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.23091921,0.0009172422,0.00269035,0.0023998818,0.0006375843,0.0034039258,0.0022824295,0.0027721124,0.0019398569],"category_scores_gemma":[0.55146486,0.0007140563,0.002173306,0.002730402,0.0061867745,0.0043289773,0.0038693137,0.003652051,0.00048898446],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002799051,0.0001785211,0.07480471,0.0025593084,0.0030289849,0.0011627552,0.0028182282,0.0678876,0.001925764,0.4758714,0.006392261,0.36057138],"study_design_scores_gemma":[0.0002972869,0.0006766806,0.015564115,0.000684162,0.00084547285,0.0012586444,0.0002469418,0.16065109,0.0021186753,0.81159955,0.0059368596,0.00012051418],"about_ca_topic_score_codex":0.001085589,"about_ca_topic_score_gemma":0.0007679804,"teacher_disagreement_score":0.76908076,"about_ca_system_score_codex":0.0017201622,"about_ca_system_score_gemma":0.0025247063,"threshold_uncertainty_score":0.94841355},"labels":[],"label_agreement":null},{"id":"W1918133462","doi":"10.1002/sim.4320","title":"Modeling continuous diagnostic test data using approximate Dirichlet process distributions","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Royal Victoria Hospital; McGill University Health Centre; Royal Victoria Regional Health Centre; McGill University","funders":"","keywords":"Nonparametric statistics; Dirichlet process; Computer science; Context (archaeology); Hierarchical Dirichlet process; Latent Dirichlet allocation; Parametric statistics; Identifiability; Cluster analysis; Bayesian probability; Data mining; Econometrics; Mathematics; Statistics; Machine learning; Artificial intelligence; Topic model","score_opus":0.09951210076774852,"score_gpt":0.36345671379794736,"score_spread":0.26394461303019884,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1918133462","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.025340153,0.00055534963,0.9713254,0.001082216,0.000067128676,0.00013821069,0.00026267258,0.00022688882,0.0010019023],"genre_scores_gemma":[0.599576,0.0013850187,0.39219996,0.000544342,0.00026118584,0.0012708986,0.00095282856,0.00014427093,0.003665557],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9851444,0.010722381,0.0004823576,0.0016228409,0.00155157,0.00047647388],"domain_scores_gemma":[0.91947675,0.072177626,0.0030948534,0.0029355884,0.0017905558,0.00052461645],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.024845505,0.0013198766,0.0032795381,0.00323189,0.0009912981,0.0039353734,0.004506814,0.00405528,0.0030937337],"category_scores_gemma":[0.09393426,0.0013243676,0.0023921963,0.003184183,0.00415928,0.004624662,0.0026285131,0.0043679317,0.0007961828],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00033014966,0.00012826416,0.008900165,0.00029107733,0.00022959452,0.00048789717,0.0011817449,0.593878,0.0007378647,0.33828568,0.001790631,0.053758953],"study_design_scores_gemma":[0.000061032104,0.000038571216,0.00077270175,0.000045227403,0.00002850118,0.0001429461,0.00008070217,0.8028238,0.00023826314,0.19482726,0.0009041721,0.00003680224],"about_ca_topic_score_codex":0.006151764,"about_ca_topic_score_gemma":0.004489252,"teacher_disagreement_score":0.024845505,"about_ca_system_score_codex":0.002483779,"about_ca_system_score_gemma":0.00148763,"threshold_uncertainty_score":0.13139719},"labels":[],"label_agreement":null},{"id":"W1922077132","doi":"10.1002/sim.6972","title":"Joint modelling of longitudinal and multi‐state processes: application to clinical progressions in prostate cancer","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":48,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Cancer Institute","keywords":"Prostate cancer; Covariate; Longitudinal data; Random effects model; Proportional hazards model; Longitudinal study; Joint (building); Baseline (sea); Prostate-specific antigen","score_opus":0.29581018521558533,"score_gpt":0.5043951133576747,"score_spread":0.2085849281420894,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1922077132","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.025081256,0.0016610159,0.9707267,0.00071424624,0.000081157064,0.00008813264,0.00028711662,0.00046625367,0.0008940488],"genre_scores_gemma":[0.7089305,0.0033312866,0.27543348,0.0004096931,0.00037697802,0.0011304409,0.0013683463,0.00043602582,0.008583177],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9953762,0.0034899842,0.00015872104,0.0004948262,0.0002871352,0.00019306046],"domain_scores_gemma":[0.9811203,0.016259579,0.0010292009,0.0006440089,0.000585595,0.00036129708],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012213016,0.0011186214,0.0019964448,0.0012009094,0.00067664456,0.0020157509,0.0016581939,0.0020983615,0.0035529954],"category_scores_gemma":[0.024257084,0.0008778634,0.0037472448,0.0015989541,0.001412267,0.0013245872,0.0033265422,0.0030417626,0.00061888],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001724883,0.00011688207,0.0074063228,0.00019719421,0.00041777734,0.00030973487,0.0003910154,0.9046508,0.00044624283,0.045498725,0.0009267209,0.039466098],"study_design_scores_gemma":[0.000015290818,0.000037043945,0.00078582583,0.000016740876,0.000032447868,0.000035609155,0.000020887357,0.97725505,0.00005269934,0.021103447,0.0006278844,0.00001706115],"about_ca_topic_score_codex":0.024063608,"about_ca_topic_score_gemma":0.013078667,"teacher_disagreement_score":0.024063608,"about_ca_system_score_codex":0.0013551101,"about_ca_system_score_gemma":0.002931698,"threshold_uncertainty_score":0.06458938},"labels":[],"label_agreement":null},{"id":"W1940783033","doi":"10.1002/sim.6110","title":"Shift-invariant target in allocation problems","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Invariant (physics); Scale invariance; Computer science; Mathematical optimization; Optimal allocation; Mathematics; Statistics","score_opus":0.10073763129737792,"score_gpt":0.4457221730689747,"score_spread":0.34498454177159676,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1940783033","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011091317,0.000120248405,0.9974004,0.00029370096,0.00004150386,0.00015526151,0.00004767303,0.00007209673,0.00076000456],"genre_scores_gemma":[0.0898921,0.0004061281,0.9026642,0.0006023157,0.00025831157,0.0030222312,0.00018675097,0.00014779983,0.002820146],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.961642,0.03143743,0.0010154865,0.0029084787,0.0023295793,0.0006670873],"domain_scores_gemma":[0.9531956,0.039547313,0.002036444,0.0030924233,0.0017218717,0.00040643825],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.042802267,0.002103382,0.00395098,0.0022391065,0.0008440615,0.0022888745,0.0034025852,0.003765183,0.008736488],"category_scores_gemma":[0.089203194,0.0016091167,0.0026267886,0.002553645,0.004157081,0.0036603857,0.0036486606,0.005468263,0.0022020799],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00061190926,0.00024633945,0.0010134309,0.0007813387,0.0003406187,0.00017660756,0.00039282054,0.18200694,0.0016867262,0.6879432,0.004052243,0.120747715],"study_design_scores_gemma":[0.00030184613,0.00037118717,0.00026590165,0.00008452855,0.00006450364,0.00008773674,0.000035167854,0.34512195,0.0012442602,0.6470432,0.0053334194,0.00004637227],"about_ca_topic_score_codex":0.0006944378,"about_ca_topic_score_gemma":0.00048822415,"teacher_disagreement_score":0.042802267,"about_ca_system_score_codex":0.0024437471,"about_ca_system_score_gemma":0.0024309938,"threshold_uncertainty_score":0.2263627},"labels":[],"label_agreement":null},{"id":"W1953290087","doi":"10.1002/sim.6733","title":"Development of a diagnostic test based on multiple continuous biomarkers with an imperfect reference test","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Canadian Institutes of Health Research; National Institute on Aging; National Institutes of Health; U.S. Department of Defense","keywords":"Test (biology); Computer science; Imperfect; Statistics; Diagnostic test; Medicine; Mathematics; Biology; Pediatrics","score_opus":0.07780357091141932,"score_gpt":0.3776192654494842,"score_spread":0.2998156945380649,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1953290087","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015027293,0.00056147715,0.9830125,0.00034691326,0.000029766796,0.00007251867,0.000105683226,0.0003249943,0.0005188617],"genre_scores_gemma":[0.34632105,0.00046160055,0.65125895,0.0004021792,0.00011722558,0.00031613684,0.00039877667,0.00007749364,0.0006466395],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98968047,0.004880838,0.00064625824,0.0021309976,0.0023498968,0.0003114558],"domain_scores_gemma":[0.9780733,0.014872531,0.0022869166,0.0020605866,0.0022558756,0.00045074357],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016929168,0.0012126693,0.0027382257,0.0043888036,0.00059182453,0.0024727958,0.0029863664,0.0025740846,0.0010292253],"category_scores_gemma":[0.050360657,0.00090396724,0.0016270957,0.0022549082,0.0019020776,0.002854177,0.0025002297,0.002203613,0.0007058921],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013267999,0.0004426787,0.08417581,0.0009909883,0.0012533552,0.0010794076,0.00043017598,0.26203358,0.021615988,0.052537724,0.0039211325,0.5701924],"study_design_scores_gemma":[0.00013944757,0.0006314946,0.010511816,0.00016323717,0.00043863145,0.0015321806,0.000077628254,0.9307631,0.014250122,0.037290588,0.0040385835,0.00016308998],"about_ca_topic_score_codex":0.0018818796,"about_ca_topic_score_gemma":0.0015864085,"teacher_disagreement_score":0.016929168,"about_ca_system_score_codex":0.0012300783,"about_ca_system_score_gemma":0.002311122,"threshold_uncertainty_score":0.089531064},"labels":[],"label_agreement":null},{"id":"W1960011844","doi":"10.1002/sim.6123","title":"Extending the Bayesian Adjustment for Confounding algorithm to binary treatment covariates to estimate the effect of smoking on carotid intima-media thickness: the Multi-Ethnic Study of Atherosclerosis","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université du Québec à Montréal","funders":"National Center for Advancing Translational Sciences; National Heart, Lung, and Blood Institute; Fonds de Recherche du Québec - Santé; National Center for Research Resources; Natural Sciences and Engineering Research Council of Canada","keywords":"Confounding; Covariate; Statistics; Bayesian probability; Sample size determination; Medicine; Mathematics; Weighting; Algorithm; Econometrics; Radiology","score_opus":0.10143956355485871,"score_gpt":0.46787909410228273,"score_spread":0.366439530547424,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1960011844","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.013673043,0.00041888354,0.98404145,0.00044086404,0.000051369178,0.00013612652,0.0001379683,0.0004307993,0.0006694375],"genre_scores_gemma":[0.1400332,0.0005096341,0.85659933,0.00043300126,0.00008703658,0.0005950288,0.000499282,0.00020704027,0.0010364233],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9869653,0.010400602,0.0003646612,0.0008179556,0.0012308395,0.00022074366],"domain_scores_gemma":[0.9822396,0.012897566,0.0012460104,0.0015800121,0.0017838948,0.00025300597],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.026323326,0.00083219993,0.0012037124,0.0015806358,0.0007761768,0.0013700514,0.0017709597,0.0010850132,0.0023542754],"category_scores_gemma":[0.07904331,0.000693425,0.0017968067,0.0014995729,0.0008197742,0.00088472053,0.0024013105,0.0018593018,0.0006069891],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009771651,0.00038507953,0.07168107,0.0008155641,0.0033042538,0.0003280052,0.0016185357,0.13801885,0.00703021,0.0823541,0.007885609,0.6856016],"study_design_scores_gemma":[0.0006653594,0.000557649,0.024705261,0.0003585913,0.00085668726,0.0005622055,0.00023452578,0.785659,0.0047260914,0.15625975,0.025211131,0.0002037428],"about_ca_topic_score_codex":0.011952209,"about_ca_topic_score_gemma":0.010908587,"teacher_disagreement_score":0.9736767,"about_ca_system_score_codex":0.00060708134,"about_ca_system_score_gemma":0.0033734834,"threshold_uncertainty_score":0.13921273},"labels":[],"label_agreement":null},{"id":"W1964213907","doi":"10.1002/sim.1026","title":"An improved CML estimation procedure for the Rasch model with item response data","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Natural Sciences and Engineering Research Council of Canada; Alberta Heritage Foundation for Medical Research","keywords":"Rasch model; Estimator; Item response theory; Polytomous Rasch model; Computer science; Ordinal data; Statistics; Conditional independence; Independence (probability theory); Goodness of fit; Local independence; Econometrics; Mathematics; Psychometrics","score_opus":0.5951936693828095,"score_gpt":0.5402817360493364,"score_spread":0.054911933333473106,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1964213907","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.001416286,0.00005683773,0.9975049,0.000086384156,0.000027922335,0.00010849248,0.000112538495,0.00029023353,0.00039648663],"genre_scores_gemma":[0.056704935,0.00015216321,0.9396273,0.00014788867,0.00010675025,0.000854589,0.00046397332,0.00020551821,0.0017369175],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9901502,0.0066963146,0.00039626402,0.0010340541,0.0015194772,0.00020367833],"domain_scores_gemma":[0.9830598,0.010778115,0.0008378565,0.002271892,0.0028991895,0.00015312772],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011632325,0.000941878,0.0014660074,0.0018694673,0.0008541429,0.0011510322,0.0016527467,0.0011876345,0.008320136],"category_scores_gemma":[0.053019393,0.0006803228,0.0011561925,0.0033418664,0.0005754777,0.0024880713,0.0021766506,0.0021829803,0.0018068934],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00036333365,0.00019537022,0.0057276585,0.0004655643,0.00032004877,0.00019833542,0.0005051116,0.12650354,0.0029864772,0.14788234,0.011937682,0.70291454],"study_design_scores_gemma":[0.00009944816,0.00015555914,0.0030258412,0.00006461829,0.000076443,0.00020170285,0.00008093109,0.91631246,0.0018969879,0.058542687,0.019461282,0.000082073806],"about_ca_topic_score_codex":0.005943593,"about_ca_topic_score_gemma":0.005699008,"teacher_disagreement_score":0.011632325,"about_ca_system_score_codex":0.0006933266,"about_ca_system_score_gemma":0.0033324554,"threshold_uncertainty_score":0.06151831},"labels":[],"label_agreement":null},{"id":"W1964475341","doi":"10.1002/sim.3150","title":"A critical appraisal of propensity‐score matching in the medical literature between 1996 and 2003","year":2007,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1250,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Propensity score matching; Covariate; Wilcoxon signed-rank test; Statistics; Observational study; Matching (statistics); Selection bias; Logistic regression; Sample size determination; Statistical significance; Medicine; Mathematics; Mann–Whitney U test","score_opus":0.2610013184506083,"score_gpt":0.5385518179316583,"score_spread":0.27755049948105,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1964475341","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00045947757,0.9956713,0.00040804213,0.00229579,0.0005183068,0.00018478419,0.00007661323,0.00000605359,0.0003796209],"genre_scores_gemma":[0.006608132,0.98858225,0.0015709675,0.0018563104,0.00067722035,0.00038617576,0.00014450811,0.000005954185,0.00016851345],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.9621062,0.015441939,0.012662008,0.0013482538,0.008068171,0.00037344985],"domain_scores_gemma":[0.7730316,0.13998975,0.039433226,0.0032070626,0.043422252,0.00091617316],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.058335125,0.0013294391,0.004062122,0.025843741,0.00076119305,0.003283175,0.001759189,0.0022280144,0.0022598226],"category_scores_gemma":[0.22393522,0.0014739647,0.0033697658,0.025962556,0.0019595884,0.004014051,0.001824863,0.0014567912,0.0005482028],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007162528,0.0000383889,0.002134661,0.27778885,0.0030601502,0.00038370836,0.0006293024,0.00044973314,0.0003200697,0.0042567453,0.028776998,0.6814451],"study_design_scores_gemma":[0.00046920058,0.0005288425,0.017088953,0.5946848,0.016222928,0.001766398,0.0008723705,0.000550853,0.0011958124,0.0068741157,0.3596042,0.00014154024],"about_ca_topic_score_codex":0.003820794,"about_ca_topic_score_gemma":0.0065435944,"teacher_disagreement_score":0.9416649,"about_ca_system_score_codex":0.008626286,"about_ca_system_score_gemma":0.014943245,"threshold_uncertainty_score":0.3085093},"labels":[],"label_agreement":null},{"id":"W1964786254","doi":"10.1002/sim.4005","title":"Flexible modeling of competing risks in survival analysis","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":39,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Canadian Institutes of Health Research","keywords":"Covariate; Proportional hazards model; Estimator; Statistics; Econometrics; Hazard; Event (particle physics); Regression analysis; Hazard ratio; Survival analysis; Regression; Accelerated failure time model; Computer science; Mathematics; Confidence interval","score_opus":0.20323835890471878,"score_gpt":0.4819807042041127,"score_spread":0.2787423452993939,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1964786254","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0056638033,0.0003535896,0.99292105,0.00042629536,0.00004030561,0.00008329468,0.000106698106,0.00012219448,0.0002827864],"genre_scores_gemma":[0.44963256,0.0014192022,0.54208493,0.00055628776,0.000285716,0.0018258595,0.00070650544,0.00019998189,0.0032890334],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.971675,0.02300575,0.0007800084,0.0019336661,0.0018631619,0.0007424708],"domain_scores_gemma":[0.93217236,0.05892479,0.003550877,0.0035658698,0.0011519707,0.0006340448],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.039161388,0.002012949,0.00235498,0.0021461535,0.0009435788,0.0024987534,0.0052124257,0.0028179192,0.0023322871],"category_scores_gemma":[0.07806874,0.0013803896,0.0032657029,0.0027769357,0.0035470678,0.002806014,0.0038971934,0.0048742127,0.00048968603],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00020069427,0.00008330331,0.0046531786,0.00023605922,0.00034805242,0.0004455277,0.000608049,0.67213464,0.0006196687,0.28873798,0.0012727326,0.030660074],"study_design_scores_gemma":[0.00007651029,0.000096102696,0.00064314064,0.000044895754,0.00006423119,0.00010495012,0.00003341993,0.7629354,0.00020519042,0.23438066,0.0013685474,0.0000470016],"about_ca_topic_score_codex":0.0058617694,"about_ca_topic_score_gemma":0.0044410094,"teacher_disagreement_score":0.039161388,"about_ca_system_score_codex":0.0021130333,"about_ca_system_score_gemma":0.002307459,"threshold_uncertainty_score":0.20710766},"labels":[],"label_agreement":null},{"id":"W1966338615","doi":"10.1002/sim.1877","title":"Two‐stage adaptive strategy for superiority and non‐inferiority hypotheses in active controlled clinical trials","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of New Brunswick","funders":"National Cancer Institute","keywords":"Sample size determination; Interim; Type I and type II errors; Flexibility (engineering); Computer science; Margin (machine learning); Adaptive design; Stage (stratigraphy); Early stopping; Statistical hypothesis testing; Statistics; Econometrics; Clinical trial; Mathematics; Medicine; Artificial intelligence; Machine learning","score_opus":0.7120274947519263,"score_gpt":0.6475429193750869,"score_spread":0.06448457537683938,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1966338615","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0032410694,0.00032633427,0.98886126,0.00041134292,0.0002587844,0.005742593,0.0000788258,0.0003485169,0.00073126756],"genre_scores_gemma":[0.055440616,0.0001953914,0.9196137,0.00074452156,0.00022549975,0.022771044,0.00011760409,0.000106771695,0.0007848203],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6310433,0.32800594,0.012193136,0.012979037,0.014497359,0.0012812358],"domain_scores_gemma":[0.5516124,0.4018446,0.011027307,0.023434358,0.010545632,0.0015357843],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.3213359,0.003570085,0.0070306924,0.0055334535,0.0014388264,0.0033495352,0.006363658,0.0068570795,0.011789792],"category_scores_gemma":[0.43326578,0.0023542834,0.0054821777,0.0034814305,0.006750605,0.006002059,0.0040611,0.010697204,0.0015522091],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.024926677,0.0019106616,0.0056348424,0.006031189,0.0059323725,0.0010436276,0.0031263104,0.041147184,0.0076842937,0.3247109,0.008442226,0.5694098],"study_design_scores_gemma":[0.015192973,0.010381302,0.0038051025,0.0011262862,0.0021505605,0.00061478716,0.00019193145,0.5104181,0.012080627,0.42551497,0.018051933,0.0004714969],"about_ca_topic_score_codex":0.00048366995,"about_ca_topic_score_gemma":0.00054123695,"teacher_disagreement_score":0.3213359,"about_ca_system_score_codex":0.0023427554,"about_ca_system_score_gemma":0.005232409,"threshold_uncertainty_score":0.83691365},"labels":[],"label_agreement":null},{"id":"W1967330157","doi":"10.1002/sim.1015","title":"Identification of significant host factors for HIV dynamics modelled by non‐linear mixed‐effects models","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Institute of Allergy and Infectious Diseases","keywords":"Covariate; Missing data; Bayes' theorem; Statistics; Linear model; Computer science; Bayes factor; Imputation (statistics); Model selection; Generalized linear mixed model; Data set; Econometrics; Mathematics; Bayesian probability","score_opus":0.022087407292849283,"score_gpt":0.29134701606028024,"score_spread":0.26925960876743094,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1967330157","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.10870742,0.00093937095,0.8869177,0.00094835786,0.00009428555,0.00036939044,0.0007088215,0.00042966826,0.0008849915],"genre_scores_gemma":[0.68822384,0.0016075152,0.3008419,0.000362165,0.00017387542,0.0018707601,0.0013619625,0.00015593407,0.0054020947],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9886199,0.009333362,0.00029403646,0.00083718833,0.0005174728,0.0003980146],"domain_scores_gemma":[0.945872,0.048899498,0.0026345553,0.001321808,0.00089537416,0.00037669978],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020542318,0.0017132448,0.0030939027,0.0021066684,0.0007122997,0.0021223107,0.0027287991,0.0024376044,0.003247269],"category_scores_gemma":[0.048231333,0.0012252994,0.003378434,0.0024051836,0.0013208564,0.0025144431,0.0017209994,0.0028532515,0.0005846404],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0021217382,0.00043796087,0.03400729,0.0008369366,0.0022065425,0.0009653728,0.0012820592,0.749404,0.0019628876,0.13602817,0.00251794,0.06822911],"study_design_scores_gemma":[0.00021636634,0.0001930932,0.0030440364,0.00004960784,0.00033871952,0.00009290717,0.00006323591,0.9493253,0.0004240981,0.045049496,0.0011480171,0.0000551291],"about_ca_topic_score_codex":0.006868594,"about_ca_topic_score_gemma":0.008398654,"teacher_disagreement_score":0.020542318,"about_ca_system_score_codex":0.0014072125,"about_ca_system_score_gemma":0.0018310734,"threshold_uncertainty_score":0.10863948},"labels":[],"label_agreement":null},{"id":"W1968677777","doi":"10.1002/sim.3829","title":"Bayesian adjustment for exposure misclassification in case–control studies","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; BC Cancer Agency; McMaster University","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Bayesian probability; Statistics; Context (archaeology); Observational study; Econometrics; Computer science; Sample size determination; Observational error; Explanatory power; Mathematics","score_opus":0.08384288068993881,"score_gpt":0.43549343458039363,"score_spread":0.35165055389045485,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1968677777","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0063569667,0.0019016489,0.98958886,0.0008021235,0.0002471426,0.00017369479,0.00007848713,0.00021043625,0.00064066146],"genre_scores_gemma":[0.32166874,0.0032461474,0.66931444,0.0011136064,0.00064371416,0.0013551677,0.00047152702,0.00018641794,0.0020002197],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9150171,0.072323374,0.002651238,0.004008048,0.0053519467,0.0006483008],"domain_scores_gemma":[0.82649577,0.14639124,0.009556498,0.010872258,0.006124981,0.0005592096],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08449205,0.001313949,0.0030450413,0.0029100329,0.0011880265,0.002061464,0.004291045,0.0028866671,0.0020089545],"category_scores_gemma":[0.32934555,0.0011218229,0.002497229,0.0033326978,0.0019639034,0.002031281,0.0027232298,0.0037951118,0.0005253472],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001421824,0.00027224462,0.030407105,0.0020982658,0.004146634,0.00094586884,0.0016946169,0.1903606,0.0026201096,0.24130638,0.01215325,0.51257306],"study_design_scores_gemma":[0.0005158264,0.00047874,0.016593028,0.00076148094,0.0012597474,0.0007212091,0.00018466826,0.49739167,0.00207427,0.46314314,0.016670994,0.0002051616],"about_ca_topic_score_codex":0.0060395338,"about_ca_topic_score_gemma":0.0047775414,"teacher_disagreement_score":0.915508,"about_ca_system_score_codex":0.0015850035,"about_ca_system_score_gemma":0.0026780677,"threshold_uncertainty_score":0.446842},"labels":[],"label_agreement":null},{"id":"W1969898458","doi":"10.1002/sim.5980","title":"Dynamic prediction of risk of death using history of cancer recurrences in joint frailty models","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":49,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Institute of Cancer Research; Institut National Du Cancer; Cancer Research UK","keywords":"Breast cancer; Event (particle physics); Cancer; Medicine; Cancer recurrence; Clinical endpoint; Disease; Time point; Random effects model; Time horizon; Computer science; Statistics; Internal medicine; Clinical trial; Mathematics","score_opus":0.19265737789735804,"score_gpt":0.4183232547021625,"score_spread":0.22566587680480443,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1969898458","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.43295994,0.0009493197,0.56313396,0.0010205827,0.00006009147,0.0000806319,0.00060769403,0.000273986,0.00091379933],"genre_scores_gemma":[0.9726447,0.0003313262,0.025215141,0.00007772441,0.00005833627,0.0000890701,0.0005128241,0.000022252496,0.001048588],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9983936,0.0007539011,0.00009142954,0.00040727475,0.00013660405,0.00021715292],"domain_scores_gemma":[0.9855356,0.011546569,0.0014302232,0.0006923634,0.00038767888,0.0004075333],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009076001,0.00087217597,0.0012688877,0.0014140226,0.00039502667,0.0011271331,0.0015954114,0.0010979265,0.0011128236],"category_scores_gemma":[0.02435507,0.0004572194,0.0012832163,0.0008217323,0.0008998193,0.0013268983,0.0015749325,0.001594218,0.00017340983],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00032529942,0.0001164912,0.05254426,0.00006999695,0.00026580226,0.0002842728,0.00020709902,0.89080864,0.00051847467,0.018066691,0.00063997007,0.03615295],"study_design_scores_gemma":[0.00001889927,0.00005841149,0.0038055275,0.000018476967,0.00005095188,0.000046094396,0.000016887609,0.98479074,0.00015506867,0.010860461,0.00015870896,0.00001976102],"about_ca_topic_score_codex":0.014857994,"about_ca_topic_score_gemma":0.00925205,"teacher_disagreement_score":0.014857994,"about_ca_system_score_codex":0.0010852752,"about_ca_system_score_gemma":0.0012195441,"threshold_uncertainty_score":0},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"observational","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W1970654520","doi":"10.1002/sim.3701","title":"Flexible modeling of the cumulative effects of time‐dependent exposures on the hazard","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":162,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Canadian Institutes of Health Research; Faculty of Medicine, McGill University; McGill University","keywords":"Covariate; Proportional hazards model; Statistics; Econometrics; Hazard; Regression analysis; Medicine; Computer science; Mathematics","score_opus":0.2270755262068856,"score_gpt":0.43366854871840455,"score_spread":0.20659302251151895,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1970654520","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.05596634,0.0006939931,0.9376453,0.0014325912,0.00014870639,0.000155844,0.0011902645,0.00033193777,0.0024350183],"genre_scores_gemma":[0.8941261,0.0012574882,0.079784766,0.0003756485,0.00025122773,0.0007009439,0.0011000512,0.0001828962,0.022220878],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99532795,0.002565495,0.00017425319,0.0008372876,0.00046815822,0.0006267706],"domain_scores_gemma":[0.98229605,0.013202224,0.0016860162,0.0014866518,0.0008146168,0.0005144114],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013961894,0.0014971002,0.0020398477,0.0018217727,0.00072917313,0.0024485318,0.005424262,0.003134058,0.0054133763],"category_scores_gemma":[0.02834576,0.0011895479,0.0029127337,0.0022683428,0.002572824,0.0025830555,0.0024073722,0.003824927,0.00074868015],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012929917,0.000060004182,0.004327074,0.000056670724,0.00016555382,0.00032359705,0.00021939242,0.90292585,0.0004731421,0.07841695,0.0009841643,0.01191837],"study_design_scores_gemma":[0.000050779505,0.000063771506,0.0018992712,0.000025008414,0.00008410385,0.000091900794,0.00004145297,0.94175696,0.0001431333,0.054453023,0.001348403,0.00004208352],"about_ca_topic_score_codex":0.023641221,"about_ca_topic_score_gemma":0.012494482,"teacher_disagreement_score":0.023641221,"about_ca_system_score_codex":0.002155215,"about_ca_system_score_gemma":0.002327314,"threshold_uncertainty_score":0.07383847},"labels":[],"label_agreement":null},{"id":"W1971149402","doi":"10.1002/sim.3104","title":"Using the bootstrap to improve estimation and confidence intervals for regression coefficients selected using backwards variable elimination","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":76,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Institute of Health Services and Policy Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Statistics; Confidence interval; Regression analysis; Mathematics; Regression; Percentile; Linear regression; Segmented regression; Robust confidence intervals; Variable (mathematics); Econometrics; Polynomial regression","score_opus":0.1800998728779016,"score_gpt":0.5190475704760493,"score_spread":0.3389476975981477,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1971149402","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009727108,0.00025707242,0.98859733,0.00012075348,0.00003888927,0.00008011089,0.000082238264,0.000608442,0.00048805628],"genre_scores_gemma":[0.20600793,0.00035921938,0.79113114,0.00019351525,0.000109442626,0.0005733662,0.0006506614,0.00050835934,0.00046630081],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9471916,0.04218477,0.0020603032,0.002198984,0.0058075124,0.0005567999],"domain_scores_gemma":[0.62438834,0.32928553,0.010653304,0.021251472,0.013780787,0.00064059807],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06522285,0.001660944,0.0027604103,0.005726977,0.0009194262,0.0018736342,0.0029752464,0.0016335874,0.0031443122],"category_scores_gemma":[0.34084997,0.0009848399,0.0021478278,0.0039395364,0.0015027355,0.0024381587,0.0029254714,0.0042833607,0.000800754],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009625119,0.0003621619,0.030497761,0.0009003118,0.0016972868,0.0011455386,0.0014345573,0.28519934,0.0066957483,0.11930957,0.008033637,0.54376155],"study_design_scores_gemma":[0.00021771672,0.0002708384,0.0072940784,0.00033310452,0.00020581791,0.00038813127,0.00014513105,0.8991341,0.008048214,0.077745356,0.0061006877,0.00011685454],"about_ca_topic_score_codex":0.0037493068,"about_ca_topic_score_gemma":0.0024933135,"teacher_disagreement_score":0.06522285,"about_ca_system_score_codex":0.00079847616,"about_ca_system_score_gemma":0.0021040647,"threshold_uncertainty_score":0.34493554},"labels":[],"label_agreement":null},{"id":"W1971827711","doi":"10.1002/sim.789","title":"Incremental net benefit in randomized clinical trials","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":108,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Joseph’s Healthcare Hamilton; McMaster University; St. Joseph's Hospital","funders":"","keywords":"Censoring (clinical trials); Computer science; Clinical trial; Statistics; Econometrics; Randomized controlled trial; Medicine; Mathematics; Surgery","score_opus":0.5807754213420423,"score_gpt":0.5744626345624336,"score_spread":0.00631278677960867,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1971827711","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.022956677,0.24342342,0.62066513,0.02976835,0.012552212,0.01081763,0.004564356,0.002008443,0.05324378],"genre_scores_gemma":[0.58840024,0.06809539,0.3001886,0.010422389,0.006425378,0.017586065,0.0017907188,0.00051150523,0.0065796627],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7341204,0.2417572,0.006402635,0.003988716,0.012326505,0.001404534],"domain_scores_gemma":[0.58512026,0.3956445,0.0075772186,0.0052682343,0.0049352385,0.0014545977],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.14258108,0.0036352784,0.008009186,0.008147801,0.00069394003,0.0064018494,0.0047291415,0.006274253,0.013724992],"category_scores_gemma":[0.44847938,0.0013795412,0.004462205,0.0069955727,0.0036363944,0.00842523,0.0037795762,0.0062192287,0.0014293506],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.017678572,0.00043799295,0.005386447,0.028043313,0.011988776,0.000563479,0.0003959216,0.13490872,0.00025567206,0.37583145,0.024512073,0.39999756],"study_design_scores_gemma":[0.005644266,0.0022160017,0.0021900937,0.007554292,0.010417881,0.0007716636,0.00013656155,0.15320678,0.0008106215,0.791059,0.025719859,0.00027305444],"about_ca_topic_score_codex":0.0019461261,"about_ca_topic_score_gemma":0.0013321153,"teacher_disagreement_score":0.8574189,"about_ca_system_score_codex":0.006593041,"about_ca_system_score_gemma":0.006018459,"threshold_uncertainty_score":0.7540498},"labels":[],"label_agreement":null},{"id":"W1971972523","doi":"10.1002/sim.1463","title":"Hierarchical Bayesian spatial modelling of small‐area rates of non‐rare disease","year":2003,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic Associations and Epidemiology","field":"Biochemistry, Genetics and Molecular Biology","cited_by":42,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"British Columbia Centre of Excellence for Women's Health; University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Bayesian probability; Computer science; Bayesian hierarchical modeling; Statistics; Hierarchical database model; Bayesian inference; Econometrics; Artificial intelligence; Data mining; Mathematics","score_opus":0.02524430431112855,"score_gpt":0.30175547951473863,"score_spread":0.27651117520361007,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1971972523","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.027070867,0.0003450469,0.96922517,0.00071508886,0.00003456959,0.00009984782,0.00073676265,0.0003211719,0.0014514747],"genre_scores_gemma":[0.74249065,0.0013031693,0.24443452,0.00027375369,0.0001565817,0.00080474274,0.001869276,0.00018816952,0.008479205],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9944523,0.0034017933,0.00024400202,0.0009178979,0.00062693685,0.00035715906],"domain_scores_gemma":[0.9805761,0.0147431865,0.0022392645,0.0011675358,0.0009432525,0.00033077257],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011064024,0.00075904204,0.0016649021,0.0028851908,0.000839956,0.0023514077,0.004898393,0.0019365656,0.004908726],"category_scores_gemma":[0.036564704,0.0012431543,0.0024564702,0.0036707418,0.0024661853,0.002609332,0.0022735028,0.0018005506,0.0006871969],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006948358,0.000034909495,0.008744208,0.00008603741,0.00019167147,0.00020117017,0.00041404396,0.7567042,0.0003421379,0.21023679,0.0014081071,0.021567227],"study_design_scores_gemma":[0.000024288425,0.000019982162,0.0022792835,0.000025816063,0.00004029627,0.00007400416,0.00004524073,0.908308,0.00008017577,0.0878965,0.0011786363,0.000027751374],"about_ca_topic_score_codex":0.059135754,"about_ca_topic_score_gemma":0.050848283,"teacher_disagreement_score":0.059135754,"about_ca_system_score_codex":0.0027481003,"about_ca_system_score_gemma":0.0022153524,"threshold_uncertainty_score":0.117583096},"labels":[],"label_agreement":null},{"id":"W1972445384","doi":"10.1002/sim.5690","title":"On estimating average effects for multiple treatment groups","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital","funders":"","keywords":"Covariate; Statistics; Propensity score matching; Robustness (evolution); Population; Econometrics; Observational study; Variable (mathematics); Average treatment effect; Instrumental variable; Mathematics; Medicine; Biology","score_opus":0.10760751865554542,"score_gpt":0.4506370219169127,"score_spread":0.3430295032613673,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1972445384","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0064106006,0.0007319258,0.99126935,0.0005217336,0.00006828283,0.00016437614,0.0001492719,0.00012080997,0.00056377496],"genre_scores_gemma":[0.19233733,0.0019991093,0.7998372,0.0010961086,0.00038024643,0.0019552337,0.0005891422,0.000112451504,0.0016932526],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.93514043,0.0549181,0.0012493181,0.0049746414,0.0031081946,0.00060925214],"domain_scores_gemma":[0.79745966,0.18419266,0.0063718935,0.009533756,0.0019879893,0.0004540768],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10081509,0.0016421238,0.0038194223,0.0032058032,0.0010214896,0.0022467417,0.0038538366,0.0028723394,0.0043519214],"category_scores_gemma":[0.17153092,0.0011125803,0.004363994,0.0041326904,0.0034681023,0.0044399067,0.0032063338,0.0041392012,0.0005134536],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007435588,0.00040184142,0.023272805,0.0012980071,0.004604495,0.00033320973,0.001101859,0.17201622,0.0010391922,0.4670351,0.0024871707,0.3256665],"study_design_scores_gemma":[0.0003730832,0.00057295326,0.0060309204,0.00036489955,0.0010379574,0.00021392207,0.00015442049,0.28058007,0.0011397031,0.7023504,0.007088661,0.00009299421],"about_ca_topic_score_codex":0.004217518,"about_ca_topic_score_gemma":0.0032438706,"teacher_disagreement_score":0.89918494,"about_ca_system_score_codex":0.0017268066,"about_ca_system_score_gemma":0.00294185,"threshold_uncertainty_score":0.5331675},"labels":[],"label_agreement":null},{"id":"W1972795335","doi":"10.1002/sim.3989","title":"Estimation of relative risk and prevalence ratio","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Canadian Institutes of Health Research","keywords":"Estimator; Statistics; Mathematics; Binomial regression; Logistic regression; Binomial (polynomial); Negative binomial distribution; Confounding; Point estimation; Relative risk; Econometrics; Computer science; Confidence interval; Poisson distribution","score_opus":0.056982222668341694,"score_gpt":0.4327492605441972,"score_spread":0.3757670378758555,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1972795335","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.013561536,0.020472705,0.9508863,0.0023385864,0.00094668625,0.0018973836,0.0041865413,0.00095936446,0.0047509],"genre_scores_gemma":[0.22223395,0.01464046,0.7441075,0.0016687427,0.0013741584,0.009087064,0.00422549,0.00049585896,0.0021667061],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.85419184,0.107839376,0.011213841,0.012263523,0.013707838,0.0007835379],"domain_scores_gemma":[0.7094652,0.2482422,0.020143278,0.014150789,0.00742498,0.0005735632],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10925594,0.0022169002,0.006200499,0.009280175,0.00055044,0.0034560272,0.004774776,0.0030672844,0.0072560012],"category_scores_gemma":[0.39152455,0.0010279514,0.005029657,0.0073890295,0.0028641473,0.0041448814,0.0038251034,0.0052787266,0.0018239097],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013729916,0.00037114322,0.08224302,0.010933805,0.015160322,0.0008095126,0.0012676573,0.049175348,0.001361816,0.16082661,0.025149167,0.6513286],"study_design_scores_gemma":[0.00081133586,0.0016371071,0.050758407,0.006666841,0.007866381,0.0047443104,0.00142761,0.1797575,0.005721675,0.60621756,0.13370387,0.00068728655],"about_ca_topic_score_codex":0.0018203051,"about_ca_topic_score_gemma":0.00064281235,"teacher_disagreement_score":0.8907441,"about_ca_system_score_codex":0.0017520665,"about_ca_system_score_gemma":0.0030542572,"threshold_uncertainty_score":0.57780755},"labels":[],"label_agreement":null},{"id":"W1972837603","doi":"10.1002/sim.2367","title":"Comparison of variance estimation approaches in a two-state Markov model for longitudinal data with misclassification","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Natural Sciences and Engineering Research Council of Canada; Alberta Heritage Foundation for Medical Research","keywords":"Jackknife resampling; Unobservable; Statistics; Variance (accounting); Markov chain; Covariance; Resampling; Computer science; Mathematics; Econometrics; Observable; Estimator","score_opus":0.31867627678948385,"score_gpt":0.4844635956960856,"score_spread":0.16578731890660175,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1972837603","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.05929194,0.00095561513,0.93817604,0.00047297805,0.00004577094,0.00009179817,0.000045017176,0.00023813969,0.00068260095],"genre_scores_gemma":[0.5699182,0.0011403734,0.42645553,0.00020288331,0.000075786826,0.00047770838,0.00037998834,0.00023426091,0.0011151417],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9726415,0.023215272,0.0006478215,0.0011191719,0.001928312,0.0004479427],"domain_scores_gemma":[0.774984,0.21252945,0.0033143421,0.0049543087,0.0036691462,0.0005487325],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06467255,0.0010243694,0.0021058917,0.002646712,0.00065587007,0.0019748106,0.002437543,0.002582025,0.0009805878],"category_scores_gemma":[0.19063841,0.0009066836,0.0018334952,0.0018895359,0.0015498882,0.003890437,0.0024393396,0.0020765292,0.00024984826],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008979283,0.0001822482,0.015831957,0.00041574682,0.00095142255,0.00021607366,0.0014278213,0.70040303,0.0008753663,0.1390569,0.0010868587,0.13865466],"study_design_scores_gemma":[0.000058944777,0.00010135173,0.0025981402,0.00008017804,0.000087595756,0.000058140766,0.000093154704,0.946997,0.00038464414,0.04900829,0.00047720075,0.000055404867],"about_ca_topic_score_codex":0.006205496,"about_ca_topic_score_gemma":0.0046196496,"teacher_disagreement_score":0.06467255,"about_ca_system_score_codex":0.0019497455,"about_ca_system_score_gemma":0.002285758,"threshold_uncertainty_score":0.34202522},"labels":[],"label_agreement":null},{"id":"W1973819936","doi":"10.1002/sim.5504","title":"Skew‐elliptical spatial random effect modeling for areal data with application to mapping health utilization rates","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Spatial and Panel Data Analysis","field":"Economics, Econometrics and Finance","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Victoria","funders":"","keywords":"Skew; Computer science; Spatial analysis; Random effects model; Statistics; Econometrics; Mathematics; Medicine","score_opus":0.12102258270912948,"score_gpt":0.3523455106071009,"score_spread":0.23132292789797143,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1973819936","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017932935,0.00027750526,0.9802029,0.00033897473,0.000019562125,0.00006729463,0.00029771516,0.00029623843,0.00056685304],"genre_scores_gemma":[0.520349,0.0009328631,0.47280526,0.00026346382,0.00007832764,0.00043983082,0.0009048563,0.00019652657,0.0040297727],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99488455,0.003762515,0.00016790988,0.00066557125,0.0003406769,0.00017883733],"domain_scores_gemma":[0.9738132,0.020510197,0.0020793816,0.0020880993,0.0012270059,0.00028208297],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014983616,0.00074752694,0.0013953103,0.0017709731,0.00075630884,0.0014463409,0.0025098682,0.0015121807,0.0031939324],"category_scores_gemma":[0.036186337,0.0008300087,0.0019053075,0.0028861319,0.0017185189,0.0020141236,0.001987204,0.0018573774,0.00046505212],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008039398,0.000048615984,0.010071876,0.00009278837,0.00023176937,0.0002235201,0.00050835346,0.7202953,0.0004178538,0.21776168,0.0015885684,0.0486793],"study_design_scores_gemma":[0.000012559782,0.000014757856,0.0012218253,0.000017309092,0.000023160577,0.000037181933,0.000042883003,0.94601613,0.0000978788,0.05123134,0.0012663421,0.000018661978],"about_ca_topic_score_codex":0.06312165,"about_ca_topic_score_gemma":0.05758545,"teacher_disagreement_score":0.06312165,"about_ca_system_score_codex":0.0027906613,"about_ca_system_score_gemma":0.0027474405,"threshold_uncertainty_score":0.12550849},"labels":[],"label_agreement":null},{"id":"W1974321986","doi":"10.1002/sim.4099","title":"On the three‐arm non‐inferiority trial including a placebo with a prespecified margin","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":47,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"McMaster University","keywords":"Margin (machine learning); Assay sensitivity; Placebo; Medicine; Sample size determination; Randomized controlled trial; Statistics; Clinical trial; Randomization; Fraction (chemistry); Mathematics; Computer science; Surgery; Internal medicine; Machine learning","score_opus":0.4639363371681384,"score_gpt":0.5489701146468974,"score_spread":0.08503377747875895,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1974321986","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.018080693,0.0022054866,0.9553463,0.00347277,0.000819641,0.007961954,0.00042660514,0.00035724032,0.011329431],"genre_scores_gemma":[0.26186877,0.0025091723,0.6979103,0.009560125,0.0011339783,0.020746222,0.00037331565,0.00020702546,0.0056912145],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9302354,0.0572018,0.003106759,0.0031256003,0.0057867854,0.0005436404],"domain_scores_gemma":[0.90751326,0.078049965,0.004479741,0.007533254,0.0018414641,0.0005822033],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08947597,0.0012919466,0.0037334473,0.0016586337,0.00065043144,0.0017395548,0.0017772035,0.004041212,0.013540655],"category_scores_gemma":[0.11566368,0.00045370936,0.0043943655,0.0015143752,0.0039450675,0.0025836185,0.001750326,0.0053860922,0.0016826702],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.026802605,0.0006898108,0.002484264,0.0067454698,0.0017014913,0.0010256056,0.00082132686,0.021822322,0.00949389,0.40350688,0.013341168,0.51156515],"study_design_scores_gemma":[0.021686714,0.0121029485,0.007679237,0.0037379721,0.0029403109,0.0013315154,0.00021163406,0.15029332,0.019064587,0.6980596,0.082469516,0.00042262013],"about_ca_topic_score_codex":0.00026838994,"about_ca_topic_score_gemma":0.0003170617,"teacher_disagreement_score":0.08947597,"about_ca_system_score_codex":0.0010656223,"about_ca_system_score_gemma":0.0022668045,"threshold_uncertainty_score":0.47319978},"labels":[],"label_agreement":null},{"id":"W1974723978","doi":"10.1002/sim.6378","title":"A cautionary note concerning the use of stabilized weights in marginal structural models","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Concordia University; Hôpital du Sacré-Cœur de Montréal; Université Laval; Université du Québec à Montréal","funders":"Fonds de Recherche du Québec - Santé; Canadian Institutes of Health Research; Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada","keywords":"Marginal structural model; Inverse probability; Econometrics; Confounding; Inverse; Statistics; Computer science; Mathematics; Posterior probability; Bayesian probability","score_opus":0.20423091439008736,"score_gpt":0.4260492700041732,"score_spread":0.22181835561408583,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1974723978","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014610201,0.009863703,0.6458306,0.2916277,0.014632057,0.0004947173,0.0016118623,0.0015335073,0.01979571],"genre_scores_gemma":[0.25910616,0.0050739697,0.5837298,0.12216453,0.011269415,0.0017864419,0.00032287938,0.0010331883,0.01551353],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.87160546,0.0990234,0.008361461,0.007398121,0.012495981,0.0011156456],"domain_scores_gemma":[0.53780884,0.4004866,0.008746064,0.03246846,0.019418208,0.0010717822],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.15095167,0.0017531894,0.00293237,0.0024141003,0.0035790761,0.0054068286,0.010171702,0.006464516,0.005285103],"category_scores_gemma":[0.4746552,0.0012663202,0.0032579778,0.004971943,0.012598175,0.008105897,0.0038210573,0.027294362,0.002415156],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003670965,0.00008625444,0.007415635,0.0011471948,0.000580426,0.0014394508,0.0058751744,0.0060068197,0.0010358064,0.6676476,0.21992359,0.08847499],"study_design_scores_gemma":[0.00020481633,0.00019942525,0.0050165127,0.0016203297,0.0002513642,0.0013401776,0.0016615664,0.028037712,0.0032305042,0.8314678,0.12664215,0.00032761972],"about_ca_topic_score_codex":0.01084188,"about_ca_topic_score_gemma":0.01906889,"teacher_disagreement_score":0.8490483,"about_ca_system_score_codex":0.00218829,"about_ca_system_score_gemma":0.0035465069,"threshold_uncertainty_score":0.79831827},"labels":[],"label_agreement":null},{"id":"W1974830949","doi":"10.1002/sim.4435","title":"Global hypothesis testing for high‐dimensional repeated measures outcomes","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Center for Advancing Translational Sciences; National Institute of Dental and Craniofacial Research; National Institute on Drug Abuse; National Institute of Diabetes and Digestive and Kidney Diseases; National Heart, Lung, and Blood Institute; National Center for Research Resources; National Institute of Neurological Disorders and Stroke; National Institute on Alcohol Abuse and Alcoholism","keywords":"Sample size determination; Type I and type II errors; Covariate; False discovery rate; Multiple comparisons problem; Computer science; Inference; Univariate; Statistical power; Statistical hypothesis testing; Statistics; Statistical inference; Nominal level; Multivariate statistics; Mathematics; Machine learning; Artificial intelligence; Confidence interval","score_opus":0.7088825096546587,"score_gpt":0.5518508889249891,"score_spread":0.15703162072966959,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1974830949","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.003995535,0.00009845976,0.9948624,0.00012643386,0.000074197604,0.00013564892,0.00007442197,0.00017581129,0.00045715895],"genre_scores_gemma":[0.22726923,0.0004424295,0.7652937,0.0006068419,0.0003055102,0.0036154652,0.00047556672,0.00025533637,0.0017359214],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9623338,0.029044734,0.0012591827,0.0032185237,0.0035973482,0.0005463629],"domain_scores_gemma":[0.8538108,0.12721758,0.0046492037,0.01085925,0.0028609359,0.0006023514],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.050746445,0.0017301373,0.0030207643,0.00171134,0.0006045869,0.0012693312,0.002810621,0.001819432,0.010389785],"category_scores_gemma":[0.14582393,0.0005757635,0.0031435383,0.0014941286,0.0044855326,0.002601606,0.0023555972,0.0034917432,0.0011472158],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012814482,0.00042466298,0.014577528,0.0017601699,0.0023639065,0.0012458599,0.00090722204,0.07436339,0.008363163,0.39493883,0.009735829,0.49003807],"study_design_scores_gemma":[0.0005348839,0.002632993,0.010593197,0.00025085083,0.000516265,0.0007614854,0.000198428,0.38396236,0.0127868,0.5803896,0.007237914,0.00013514976],"about_ca_topic_score_codex":0.00032535486,"about_ca_topic_score_gemma":0.00037378923,"teacher_disagreement_score":0.050746445,"about_ca_system_score_codex":0.0008750307,"about_ca_system_score_gemma":0.0020068418,"threshold_uncertainty_score":0.26837605},"labels":[],"label_agreement":null},{"id":"W1975313883","doi":"10.1002/sim.4329","title":"Testing the evolution process of prostate‐specific antigen in early stage prostate cancer: what is the proper underlying model?","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Prostate Cancer Diagnosis and Treatment","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Princess Margaret Cancer Centre; University of Toronto; McMaster University; York University","funders":"","keywords":"Prostate-specific antigen; Stage (stratigraphy); Prostate; Prostate cancer; Process (computing); Cancer; Medicine; Oncology; Computer science; Internal medicine; Biology","score_opus":0.10325058714932615,"score_gpt":0.34510520111267246,"score_spread":0.2418546139633463,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1975313883","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92473984,0.00033990847,0.07005906,0.002657965,0.00003151902,0.00012904053,0.00024702869,0.00007550399,0.0017201394],"genre_scores_gemma":[0.9903137,0.00020075614,0.008492326,0.00015502585,0.000024049223,0.000075652846,0.00017894011,0.000010538551,0.0005489444],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9937862,0.0043546516,0.00015443226,0.0008290452,0.00043214334,0.00044343673],"domain_scores_gemma":[0.8876206,0.10252416,0.006363345,0.0016438749,0.0011798505,0.0006681463],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017033089,0.0008479099,0.0017951431,0.0008964899,0.00058907236,0.0017618538,0.0021155041,0.0018486998,0.0017549176],"category_scores_gemma":[0.07677726,0.00045250778,0.0014273184,0.00082284724,0.0022330754,0.0022887972,0.001089918,0.0017740272,0.00025015575],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011469724,0.00067788805,0.22936894,0.00039731862,0.00076634815,0.0006503137,0.00087381306,0.61793625,0.0019411616,0.109374024,0.0011207432,0.03574623],"study_design_scores_gemma":[0.00020881704,0.00083854015,0.025171619,0.000049573286,0.00017606275,0.00016637682,0.00044184233,0.91204035,0.0008196628,0.059430096,0.0006034064,0.000053552198],"about_ca_topic_score_codex":0.007300037,"about_ca_topic_score_gemma":0.0044199014,"teacher_disagreement_score":0.017033089,"about_ca_system_score_codex":0.0021551843,"about_ca_system_score_gemma":0.002122141,"threshold_uncertainty_score":0.09008062},"labels":[],"label_agreement":null},{"id":"W1975851557","doi":"10.1002/sim.3470","title":"Modeling conditional dependence between diagnostic tests: A multiple latent variable model","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":94,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; McGill University; McGill University Health Centre","funders":"Natural Sciences and Engineering Research Council of Canada; U.S. Public Health Service; Centers for Disease Control and Prevention","keywords":"Latent class model; Latent variable; Latent variable model; Covariate; Conditional dependence; Statistics; Bayesian probability; Computer science; Econometrics; Measure (data warehouse); Mathematics; Data mining","score_opus":0.054047281841197406,"score_gpt":0.31789441846119393,"score_spread":0.26384713661999654,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1975851557","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029885063,0.00088437396,0.96531606,0.0019107686,0.00007564216,0.00010586401,0.0003270053,0.00022749278,0.0012676942],"genre_scores_gemma":[0.7380809,0.0017642927,0.2526996,0.0005439981,0.00043236284,0.00084267085,0.001100535,0.00016443031,0.0043711783],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9814311,0.012304479,0.00060117064,0.002646526,0.0020214205,0.0009953041],"domain_scores_gemma":[0.90110844,0.08702731,0.0051522907,0.0035967466,0.002191607,0.000923513],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.025493113,0.0015906873,0.003148434,0.0036315392,0.0011846421,0.004238855,0.0058149206,0.0038863535,0.003704561],"category_scores_gemma":[0.088106796,0.0017510243,0.0024861642,0.003591597,0.004643741,0.0062581664,0.003949357,0.0058002053,0.00075735716],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00053217885,0.00035172576,0.024366962,0.0002970376,0.0009031148,0.00060884724,0.0014052844,0.35889065,0.0007791925,0.535556,0.0027479732,0.07356099],"study_design_scores_gemma":[0.00009728277,0.00008663308,0.0027655738,0.00009059806,0.00019508493,0.00018572096,0.000074738025,0.7150043,0.00023218177,0.27978468,0.0014083243,0.00007489029],"about_ca_topic_score_codex":0.00950305,"about_ca_topic_score_gemma":0.0071427855,"teacher_disagreement_score":0.025493113,"about_ca_system_score_codex":0.0030350233,"about_ca_system_score_gemma":0.0027958958,"threshold_uncertainty_score":0.13482207},"labels":[],"label_agreement":null},{"id":"W1975994236","doi":"10.1002/sim.3854","title":"The performance of different propensity-score methods for estimating differences in proportions (risk differences or absolute risk reductions) in observational studies","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":359,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Covariate; Statistics; Confidence interval; Observational study; Estimator; Mean squared error; Relative risk; Matching (statistics); Inverse probability weighting; Mathematics; Econometrics; Medicine","score_opus":0.4011751616883879,"score_gpt":0.5200551633034011,"score_spread":0.11888000161501316,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1975994236","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.034756232,0.013078188,0.94167495,0.0017615819,0.0005445868,0.004067523,0.0015214102,0.00046991062,0.0021255938],"genre_scores_gemma":[0.32929915,0.0063689854,0.65108514,0.0012509812,0.00041274802,0.008857602,0.0017233668,0.00043490963,0.000567083],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.508193,0.413272,0.036204834,0.016259324,0.024826331,0.001244452],"domain_scores_gemma":[0.27542666,0.62126666,0.04339208,0.041158132,0.017916646,0.0008398208],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4396123,0.002058965,0.0039445283,0.0076464596,0.00156653,0.005256041,0.0043506776,0.0032430969,0.0028180587],"category_scores_gemma":[0.70922977,0.0017790443,0.0104674315,0.012402589,0.003932175,0.0054547046,0.004308516,0.004812075,0.00058150606],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0042364323,0.00041647328,0.1859894,0.00841173,0.034373183,0.00047081528,0.0034810076,0.15720506,0.0012784553,0.091546625,0.0094453115,0.5031455],"study_design_scores_gemma":[0.003740691,0.0034452432,0.14565928,0.007521451,0.014589313,0.0017052256,0.0009385151,0.50861615,0.007220493,0.2627421,0.04248298,0.001338544],"about_ca_topic_score_codex":0.005815591,"about_ca_topic_score_gemma":0.003693299,"teacher_disagreement_score":0.56038773,"about_ca_system_score_codex":0.003350895,"about_ca_system_score_gemma":0.004824952,"threshold_uncertainty_score":0.6910578},"labels":[],"label_agreement":null},{"id":"W1976786126","doi":"10.1002/sim.911","title":"A general goodness‐of‐fit approach for inference procedures concerning the kappa statistic","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Goodness of fit; Statistics; Polytomous Rasch model; Statistic; Cohen's kappa; Inference; Mathematics; Statistical inference; Test statistic; Multinomial distribution; Dirichlet distribution; Outcome (game theory); Kappa; Range (aeronautics); Confidence interval; Econometrics; Statistical hypothesis testing; Computer science; Item response theory; Psychometrics; Artificial intelligence","score_opus":0.2882924109126018,"score_gpt":0.4485972092231941,"score_spread":0.1603047983105923,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1976786126","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0014451456,0.00019572937,0.9966989,0.00020420761,0.00005303442,0.00030994517,0.00006888304,0.00022276618,0.00080128055],"genre_scores_gemma":[0.0755984,0.00041244773,0.9193069,0.00044848424,0.00019587556,0.0029837321,0.00026359197,0.00024315766,0.0005474819],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.72980374,0.21978481,0.009939073,0.014206622,0.024517473,0.001748266],"domain_scores_gemma":[0.53927386,0.39773837,0.01441923,0.028407883,0.019188154,0.00097248895],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1990702,0.003046685,0.0048069824,0.01446625,0.0030317123,0.004993923,0.007185344,0.0065245754,0.0072053317],"category_scores_gemma":[0.5594519,0.0021054598,0.006545209,0.00837079,0.009444383,0.007726405,0.008252952,0.008232064,0.0026768008],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00051988865,0.00025334742,0.013343384,0.0027974849,0.0029693514,0.0007033141,0.004285059,0.07976567,0.0028313883,0.46637496,0.012542287,0.41361395],"study_design_scores_gemma":[0.00022058973,0.00070437137,0.0060994816,0.0009915559,0.00046812507,0.0011656906,0.000807247,0.19807033,0.0022495731,0.7758346,0.013044262,0.00034414956],"about_ca_topic_score_codex":0.0028964474,"about_ca_topic_score_gemma":0.0022632354,"teacher_disagreement_score":0.1990702,"about_ca_system_score_codex":0.0027174188,"about_ca_system_score_gemma":0.004752644,"threshold_uncertainty_score":0.987689},"labels":[],"label_agreement":null},{"id":"W1979134293","doi":"10.1002/sim.3310","title":"Estimating error rates in the classification of paired organs","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Retinal Imaging and Analysis","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Estimator; Resampling; Statistics; Cross-validation; Word error rate; Computer science; Sample size determination; Contrast (vision); Autocorrelation; Artificial intelligence; Variance (accounting); Pattern recognition (psychology); Mathematics","score_opus":0.06777729135711709,"score_gpt":0.380192824480855,"score_spread":0.31241553312373793,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1979134293","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.17569232,0.0007283382,0.8219744,0.00020029071,0.000088987705,0.00015571577,0.00019514748,0.00026415088,0.00070059067],"genre_scores_gemma":[0.7826317,0.00029295732,0.21455516,0.00018904902,0.00009383539,0.00038361494,0.000913942,0.00016756947,0.00077218283],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.96443254,0.021267584,0.0032668312,0.0046560355,0.005639364,0.0007376911],"domain_scores_gemma":[0.7394215,0.22008042,0.011070202,0.021676848,0.007138359,0.0006126678],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.058508962,0.0012152398,0.002028634,0.0026849296,0.00059933274,0.0019034568,0.0018085712,0.002385579,0.00064920087],"category_scores_gemma":[0.17185086,0.0006046708,0.0014897926,0.001488084,0.0027367142,0.0020638637,0.0025619832,0.0019355393,0.00043434286],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0021106636,0.0004647673,0.274696,0.0008266818,0.0017658884,0.00093065447,0.0017761759,0.2768288,0.017470678,0.021001685,0.0015406157,0.4005875],"study_design_scores_gemma":[0.00009612087,0.0007906796,0.1332368,0.00033339034,0.00046320073,0.0019981887,0.0004500008,0.7408658,0.045509107,0.07400269,0.0019623032,0.00029171957],"about_ca_topic_score_codex":0.0006851896,"about_ca_topic_score_gemma":0.00046368563,"teacher_disagreement_score":0.058508962,"about_ca_system_score_codex":0.0005781681,"about_ca_system_score_gemma":0.00056809187,"threshold_uncertainty_score":0.3094287},"labels":[],"label_agreement":null},{"id":"W1979394579","doi":"10.1002/sim.1974","title":"Influence analysis for linear mixed‐effects models","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":51,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences","funders":"National Cancer Institute","keywords":"Linear regression; Leverage (statistics); Mathematics; Measure (data warehouse); Linear model; Regression analysis; Statistics; Generalization; Regression; Applied mathematics; Simple linear regression; Generalized linear model; Infinitesimal; Simple (philosophy); Econometrics; Computer science; Data mining","score_opus":0.06358645407001984,"score_gpt":0.4203984667152716,"score_spread":0.35681201264525175,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1979394579","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0021169442,0.00030821547,0.99657995,0.000116806135,0.000023527035,0.000048385507,0.00004264658,0.000111784095,0.0006517418],"genre_scores_gemma":[0.2386879,0.0018820256,0.75319886,0.00047092434,0.00044106654,0.0014657272,0.0005995459,0.000303342,0.0029505745],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9736076,0.020147476,0.00083697174,0.0020428337,0.0030416152,0.00032346515],"domain_scores_gemma":[0.8681608,0.12091748,0.0036173444,0.0036373704,0.0031179744,0.0005490935],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026044691,0.0017035285,0.00226264,0.0042369496,0.0010043201,0.0018800757,0.0026832062,0.0015438999,0.0035825777],"category_scores_gemma":[0.116464615,0.0010005049,0.0033625867,0.0027896955,0.0023770502,0.002614364,0.0030956827,0.0027481748,0.0006117136],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001516616,0.00008239756,0.0055517456,0.00054532796,0.0010527471,0.00048864767,0.0006021416,0.12444476,0.00093583186,0.7319121,0.0020926488,0.13214009],"study_design_scores_gemma":[0.00003485787,0.00010106323,0.001154741,0.000086743814,0.00022740639,0.00017441073,0.00005437123,0.59796906,0.00074848015,0.3933662,0.0060407035,0.00004209862],"about_ca_topic_score_codex":0.0038449778,"about_ca_topic_score_gemma":0.003013823,"teacher_disagreement_score":0.026044691,"about_ca_system_score_codex":0.0017095389,"about_ca_system_score_gemma":0.0016049668,"threshold_uncertainty_score":0.13773918},"labels":[],"label_agreement":null},{"id":"W1979949183","doi":"10.1002/(sici)1097-0258(20000215)19:3<323::aid-sim372>3.0.co;2-d","title":"Saddlepoint approximations for small sample logistic regression problems","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"National Cancer Institute","keywords":"Inference; Mathematics; Logistic regression; Sequence (biology); Applied mathematics; Binomial distribution; Statistics; Binomial regression; Econometrics; Computer science; Artificial intelligence","score_opus":0.33454370973748493,"score_gpt":0.5060935000628242,"score_spread":0.17154979032533924,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1979949183","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0026576843,0.00039220587,0.9956927,0.00022552277,0.000029934967,0.00002691113,0.00003515595,0.00007705262,0.00086298375],"genre_scores_gemma":[0.39467004,0.003706057,0.58699924,0.0006972333,0.00031116817,0.0014001667,0.00071360846,0.00047763556,0.011024881],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99625254,0.0026101961,0.00017627202,0.00030557506,0.00052769016,0.00012769765],"domain_scores_gemma":[0.968071,0.028746977,0.000980233,0.0007678488,0.0011637702,0.00027024542],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012613322,0.0014137742,0.0021271585,0.0018174378,0.0005765997,0.0019220245,0.0019679,0.0017836603,0.0038934797],"category_scores_gemma":[0.056738444,0.0012093253,0.0014985051,0.0013937863,0.002194904,0.0026867113,0.002103265,0.0035855153,0.00088261225],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009712183,0.000034054658,0.0012172548,0.0002597621,0.00010500023,0.00024012911,0.00020110894,0.60806584,0.0006291842,0.3654297,0.0023459003,0.021374857],"study_design_scores_gemma":[0.000020499901,0.000022719383,0.000100395795,0.000031613814,0.000009102902,0.00003605927,0.000017083545,0.8867956,0.00014634036,0.1117927,0.001017585,0.000010339346],"about_ca_topic_score_codex":0.002634206,"about_ca_topic_score_gemma":0.001753019,"teacher_disagreement_score":0.012613322,"about_ca_system_score_codex":0.0013062155,"about_ca_system_score_gemma":0.001210578,"threshold_uncertainty_score":0.06670642},"labels":[],"label_agreement":null},{"id":"W1980747896","doi":"10.1002/sim.2326","title":"Checking stationarity of the incidence rate using prevalent cohort survival data","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":66,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University Health Centre; Montreal Children's Hospital; McGill University","funders":"","keywords":"Incidence (geometry); Cohort; Kaplan–Meier estimator; Estimator; Statistics; Survival analysis; Medicine; Econometrics; Demography; Mathematics","score_opus":0.18573745560773047,"score_gpt":0.466115744944905,"score_spread":0.2803782893371745,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1980747896","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.18322578,0.00022881286,0.8128597,0.00051882915,0.000040962976,0.00014640712,0.0011552843,0.00039917772,0.0014251145],"genre_scores_gemma":[0.8398974,0.00032442913,0.15671383,0.00015714618,0.00008368653,0.00023190529,0.0019486309,0.000093718576,0.0005493232],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9853233,0.008024211,0.0015596687,0.0022435659,0.002204993,0.00064437004],"domain_scores_gemma":[0.73935455,0.21373105,0.019176042,0.020578992,0.0062688743,0.0008904358],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03801329,0.00043544025,0.0011402598,0.004190271,0.0010239327,0.0015923853,0.0016531392,0.0015699913,0.0022927292],"category_scores_gemma":[0.26166257,0.0005692971,0.0016671591,0.0034843828,0.001659684,0.0025745342,0.0017019105,0.0016149761,0.00041233108],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012309892,0.0002781589,0.5771235,0.0005374554,0.0010118014,0.0021196892,0.0041751517,0.07126428,0.006164528,0.18370718,0.0030945446,0.14929274],"study_design_scores_gemma":[0.00030898422,0.00076395605,0.23080435,0.0002664398,0.00053173315,0.002283364,0.0009092739,0.47678328,0.012169787,0.26641572,0.008563894,0.0001991863],"about_ca_topic_score_codex":0.009277405,"about_ca_topic_score_gemma":0.0055443617,"teacher_disagreement_score":0.03801329,"about_ca_system_score_codex":0.0009006706,"about_ca_system_score_gemma":0.002058968,"threshold_uncertainty_score":0.20103592},"labels":[],"label_agreement":null},{"id":"W1982959065","doi":"10.1002/sim.5795","title":"Propensity scores used for analysis of cluster randomized trials with selection bias: a simulation study","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":40,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Propensity score matching; CRTS; Inverse probability weighting; Statistics; Selection bias; Weighting; Regression; Causal inference; Selection (genetic algorithm); Type I and type II errors; Matching (statistics); Randomization; Econometrics; Randomized controlled trial; Computer science; Mathematics; Medicine; Artificial intelligence; Internal medicine","score_opus":0.32206622457838024,"score_gpt":0.502459386483331,"score_spread":0.18039316190495075,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1982959065","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1906442,0.0047871005,0.7917142,0.0020565945,0.0002765046,0.0056628594,0.00057129096,0.0005064257,0.003780835],"genre_scores_gemma":[0.7488573,0.0016171574,0.24340427,0.0004703908,0.00009648567,0.004777638,0.0002823983,0.00007412163,0.0004202974],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.88969827,0.104735784,0.0020576478,0.001217954,0.0018255534,0.0004648118],"domain_scores_gemma":[0.5069065,0.46086642,0.010461975,0.014422093,0.0061518955,0.0011911911],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09905737,0.0011306588,0.0022897697,0.0017336599,0.0005726208,0.0015131811,0.0016638244,0.0020460286,0.0031614555],"category_scores_gemma":[0.32553363,0.00066955783,0.003061156,0.0028103387,0.0012062255,0.0015781703,0.001682845,0.00236792,0.00022759063],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.016097132,0.0018823321,0.03497265,0.0037835422,0.008827877,0.0009075506,0.00097291433,0.6722412,0.001078544,0.12947413,0.0056342524,0.12412783],"study_design_scores_gemma":[0.005829375,0.0030244056,0.002957641,0.0005985437,0.0019237369,0.0003120412,0.00014148491,0.9227362,0.00080105074,0.058282726,0.0032828,0.00010992961],"about_ca_topic_score_codex":0.0032079113,"about_ca_topic_score_gemma":0.0018531396,"teacher_disagreement_score":0.9009426,"about_ca_system_score_codex":0.0016278387,"about_ca_system_score_gemma":0.003600037,"threshold_uncertainty_score":0.52387166},"labels":[],"label_agreement":null},{"id":"W1983192352","doi":"10.1002/sim.3191","title":"A handbook of statistical analyses using Stata (2nd edn). Sophia Rabe‐Hesketh and Brian S. Everitt, Chapman &amp; Hall/CRC, Boca Raton, FL, 2007. No. of pages: ix + 342. Price: $49.46. ISBN 1‐58488‐756‐7","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Probability and Statistical Research","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Library science; Computer science","score_opus":0.23171456911003172,"score_gpt":0.4607068314387526,"score_spread":0.22899226232872086,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1983192352","genre_codex":"methods","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0027408903,0.05375082,0.60589576,0.015428757,0.008432756,0.0044751423,0.15529653,0.08634874,0.06763059],"genre_scores_gemma":[0.011773487,0.05213059,0.737377,0.006268232,0.0038971184,0.014270733,0.0735392,0.029972773,0.0707708],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98234737,0.008246385,0.0032220061,0.0013274177,0.0044876826,0.00036915333],"domain_scores_gemma":[0.8997897,0.072612725,0.006588305,0.0099088205,0.009786563,0.0013138595],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017717576,0.0021961222,0.0038427575,0.008547321,0.00095181714,0.0035091566,0.0040253256,0.0017430577,0.12676717],"category_scores_gemma":[0.06424962,0.0033356876,0.0017942856,0.01049518,0.0014613026,0.003873924,0.0026163992,0.006839549,0.11601168],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000076525015,0.00011322323,0.0006821771,0.0018904483,0.000105910614,0.00008503262,0.00013952523,0.0011286483,0.00046769812,0.002808182,0.7953566,0.19714598],"study_design_scores_gemma":[0.00020021875,0.00019236974,0.0057935026,0.0026797827,0.00014681532,0.000898368,0.0002499586,0.003421508,0.001138712,0.03506689,0.95001715,0.00019477567],"about_ca_topic_score_codex":0.0033011732,"about_ca_topic_score_gemma":0.0053557875,"teacher_disagreement_score":0.12676717,"about_ca_system_score_codex":0.0011827017,"about_ca_system_score_gemma":0.0069359145,"threshold_uncertainty_score":0.42407846},"labels":[],"label_agreement":null},{"id":"W1984151175","doi":"10.1002/sim.1271","title":"Dealing with competing risks: testing covariates and calculating sample size","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":63,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Princess Margaret Cancer Centre; Ontario Institute for Cancer Research","funders":"","keywords":"Covariate; Statistics; Econometrics; Proportional hazards model; Event (particle physics); Sample size determination; Hazard; Hazard ratio; Mathematics; Computer science; Confidence interval","score_opus":0.6075631836183745,"score_gpt":0.5529054365405092,"score_spread":0.054657747077865304,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1984151175","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008258816,0.0010601338,0.9839581,0.0018265295,0.00036741936,0.0028932975,0.0002646069,0.00037889168,0.0009922339],"genre_scores_gemma":[0.071387336,0.0007236675,0.91165185,0.00076682644,0.00030309288,0.014214792,0.0003455789,0.00015250906,0.00045439918],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.78195524,0.18642934,0.009158815,0.006085968,0.015730696,0.0006399462],"domain_scores_gemma":[0.43640688,0.5271728,0.0078924885,0.020700837,0.0068473113,0.0009796765],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.19555597,0.0013331877,0.0041164337,0.004352157,0.0012278529,0.0026215394,0.0044284263,0.0040424773,0.0032439095],"category_scores_gemma":[0.5930189,0.001097009,0.0020143336,0.003113124,0.0041332752,0.0043144994,0.00329939,0.003601785,0.0006929388],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020004215,0.0005225403,0.03462187,0.0026712697,0.0014968418,0.0009975259,0.0023509595,0.026097855,0.0034398888,0.12818918,0.012590369,0.78502136],"study_design_scores_gemma":[0.002052616,0.0051230127,0.018426914,0.001985064,0.0008646724,0.002447074,0.0007929351,0.18268216,0.010483948,0.7357814,0.039016183,0.00034410617],"about_ca_topic_score_codex":0.0011524403,"about_ca_topic_score_gemma":0.00079721597,"teacher_disagreement_score":0.19555597,"about_ca_system_score_codex":0.001093231,"about_ca_system_score_gemma":0.0031643303,"threshold_uncertainty_score":0.9920227},"labels":[],"label_agreement":null},{"id":"W1985369819","doi":"10.1002/sim.2403","title":"A workflow spatial scan statistic","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Data-Driven Disease Surveillance","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Scan statistic; Workflow; Statistic; Computer science; Completeness (order theory); Data mining; Statistics; Mathematics; Database","score_opus":0.010736343537186738,"score_gpt":0.3030262739491112,"score_spread":0.29228993041192447,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1985369819","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009808766,0.00012801168,0.9840577,0.0008779378,0.00018202183,0.0001631,0.0012425228,0.0016736358,0.0018663654],"genre_scores_gemma":[0.24345203,0.00026658564,0.74762136,0.0009723035,0.0003322981,0.0007570856,0.0035066162,0.00051164214,0.0025800793],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98841953,0.0062118033,0.00075556803,0.0015491057,0.0025724063,0.00049154286],"domain_scores_gemma":[0.95643103,0.025281563,0.0023360923,0.007054159,0.007720355,0.0011768969],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013918666,0.00073726627,0.001296187,0.0024562445,0.000590173,0.0022413395,0.0035935652,0.0016050922,0.0053279554],"category_scores_gemma":[0.052796517,0.00051133835,0.0014853708,0.0038475173,0.001537767,0.0027608108,0.0021198813,0.0023119552,0.0014345618],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015643386,0.00032203356,0.0499307,0.0004004403,0.00041356278,0.00070137077,0.0003343021,0.19968858,0.0052255755,0.22965279,0.036798414,0.47496793],"study_design_scores_gemma":[0.0003118714,0.00036549923,0.004001651,0.00007266689,0.00011014845,0.0008284526,0.00013318777,0.8040456,0.0042248424,0.1605981,0.025206612,0.00010140075],"about_ca_topic_score_codex":0.0059261736,"about_ca_topic_score_gemma":0.004795076,"teacher_disagreement_score":0.013918666,"about_ca_system_score_codex":0.0013916409,"about_ca_system_score_gemma":0.0064873532,"threshold_uncertainty_score":0.07360983},"labels":[],"label_agreement":null},{"id":"W1986440280","doi":"10.1002/sim.3276","title":"The performance of different propensity score methods for estimating marginal odds ratios, <i>Statistics in Medicine</i> 2007; <b>26</b>:3078–3094","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":34,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care","keywords":"Citation; Library science; Odds; Sociology; Medicine; Statistics; Mathematics; Computer science; Logistic regression","score_opus":0.19247240902296991,"score_gpt":0.4680823787825111,"score_spread":0.27560996975954116,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1986440280","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.13476305,0.0057316693,0.8531286,0.0020215537,0.00038810307,0.00026063118,0.00084461697,0.0012361186,0.0016255945],"genre_scores_gemma":[0.4551406,0.0023046695,0.5380627,0.0004574114,0.000205368,0.00042763073,0.0015041581,0.0005111231,0.001386359],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9675721,0.027199484,0.0014745404,0.0013817892,0.0020650935,0.0003070155],"domain_scores_gemma":[0.6931232,0.283589,0.0045205345,0.012471291,0.005324556,0.0009714675],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.076192655,0.0012332851,0.0018796991,0.0036760161,0.000887246,0.0028459744,0.0028499109,0.0026117656,0.0035424314],"category_scores_gemma":[0.29111487,0.0009145431,0.0033480087,0.0035622467,0.0017756426,0.0038870566,0.0025416985,0.0024441795,0.0007171134],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009570111,0.00058190926,0.0761856,0.0016056858,0.0069268816,0.00019913349,0.0013889786,0.13802549,0.0017064667,0.074481525,0.014473802,0.67485446],"study_design_scores_gemma":[0.0019664664,0.0008691674,0.03100358,0.00030126225,0.001829089,0.00057083264,0.00038403974,0.8370856,0.0043409243,0.11712655,0.004258585,0.0002639099],"about_ca_topic_score_codex":0.0047654766,"about_ca_topic_score_gemma":0.0050034453,"teacher_disagreement_score":0.9238073,"about_ca_system_score_codex":0.0011230733,"about_ca_system_score_gemma":0.002965254,"threshold_uncertainty_score":0.4029501},"labels":[],"label_agreement":null},{"id":"W1986551913","doi":"10.1002/sim.3922","title":"Optimizing the response to surveillance alerts in automated surveillance systems","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"COVID-19 epidemiological studies","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Robustness (evolution); Robust optimization; Operations research; Data mining; Realization (probability); Machine learning; Mathematical optimization; Statistics; Engineering","score_opus":0.21525991537945466,"score_gpt":0.4328536458959758,"score_spread":0.21759373051652114,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1986551913","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3258715,0.00040246753,0.67037576,0.0009241814,0.000037283637,0.00012761616,0.000088897956,0.00042701306,0.0017453289],"genre_scores_gemma":[0.9639884,0.000116390125,0.03508124,0.00007098784,0.000020887808,0.00007201838,0.0000469512,0.000024013603,0.0005790226],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9967801,0.0018861797,0.00012931599,0.00046541676,0.00042212493,0.00031698268],"domain_scores_gemma":[0.98903954,0.008883762,0.0011064385,0.000313059,0.00047622938,0.00018110068],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003881826,0.0009403591,0.0011166784,0.00061110465,0.0004228318,0.0010735891,0.00089889683,0.0012396962,0.00058147975],"category_scores_gemma":[0.01958372,0.0005731873,0.0003736743,0.0007039789,0.0010207585,0.0013246427,0.00093190593,0.0009356028,0.00012845315],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000078102705,0.000035038687,0.00082916097,0.000019967072,0.000016450549,0.000012175098,0.000028140554,0.98817855,0.00054515357,0.0015899529,0.00012078306,0.008546619],"study_design_scores_gemma":[0.000013179944,0.000055885503,0.00028940904,0.0000026786156,0.00000543024,0.000005181383,0.000014481302,0.9964495,0.00042220714,0.002664716,0.00007298137,0.000004375897],"about_ca_topic_score_codex":0.0065174648,"about_ca_topic_score_gemma":0.002489345,"teacher_disagreement_score":0.0065174648,"about_ca_system_score_codex":0.001489533,"about_ca_system_score_gemma":0.0020228135,"threshold_uncertainty_score":0.02052933},"labels":[],"label_agreement":null},{"id":"W1986927305","doi":"10.1002/sim.2782","title":"Methods for the statistical analysis of binary data in split‐mouth designs with baseline measurements","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Baseline (sea); Computer science; Context (archaeology); Binary number; Binary data; Outcome (game theory); Statistics; Statistical model; Statistical analysis; Artificial intelligence; Mathematics","score_opus":0.46426087040142916,"score_gpt":0.5858141738594048,"score_spread":0.12155330345797566,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1986927305","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00055562187,0.0002934586,0.9974758,0.000079022786,0.000086802625,0.0009797007,0.000110712564,0.00017743386,0.00024136298],"genre_scores_gemma":[0.011703495,0.00055539823,0.9715271,0.00015527473,0.00012363994,0.01519481,0.00021880903,0.00009849646,0.00042301172],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8187933,0.15624082,0.005437464,0.0062375506,0.012579979,0.00071093184],"domain_scores_gemma":[0.70261186,0.25261396,0.01433033,0.024958186,0.004871484,0.00061421463],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.14429681,0.0029753947,0.0041334694,0.0054664537,0.0010781813,0.0021054377,0.004203016,0.003695073,0.016655775],"category_scores_gemma":[0.2739505,0.0018585117,0.0043791067,0.004782813,0.004711526,0.002989196,0.0030791147,0.006900936,0.002845796],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00298055,0.0009172449,0.0026329793,0.005385634,0.0025125595,0.00029397596,0.0012805684,0.035438474,0.0040953695,0.38147342,0.01109588,0.5518933],"study_design_scores_gemma":[0.002991092,0.0044897087,0.005653761,0.0022167151,0.0011291917,0.0007000815,0.00023716914,0.22965212,0.007632072,0.7000626,0.044835217,0.00040033954],"about_ca_topic_score_codex":0.00061262696,"about_ca_topic_score_gemma":0.00067783054,"teacher_disagreement_score":0.14429681,"about_ca_system_score_codex":0.0015776099,"about_ca_system_score_gemma":0.0029020028,"threshold_uncertainty_score":0.7631236},"labels":[],"label_agreement":null},{"id":"W1987001026","doi":"10.1002/sim.3882","title":"The analysis of treatment effects for recurring episodic conditions","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; McMaster University","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Randomized controlled trial; Migraine; Disease; Neurology; Clinical trial; Asthma; Chronic Migraine; Intensive care medicine; Pediatrics; Psychiatry; Internal medicine","score_opus":0.04607669157443555,"score_gpt":0.44360969574707126,"score_spread":0.3975330041726357,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1987001026","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.027111497,0.041418597,0.91191494,0.00930185,0.0012388179,0.0015798522,0.0017429496,0.00037604955,0.0053154933],"genre_scores_gemma":[0.47799855,0.032110225,0.47064823,0.004509346,0.0022053209,0.005933591,0.0016151115,0.00021644044,0.004763198],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8973397,0.09249858,0.0024080523,0.0031471327,0.003998708,0.0006078227],"domain_scores_gemma":[0.6126915,0.37128028,0.008177493,0.005685147,0.001648465,0.0005171051],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11314958,0.0014821845,0.005741997,0.0026305465,0.0006781087,0.0025521964,0.0029658657,0.0037973209,0.007597959],"category_scores_gemma":[0.2572396,0.00090502645,0.005633494,0.0023758973,0.00353116,0.00399574,0.0021450245,0.0061022514,0.00051342667],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0034466882,0.00040819807,0.012154474,0.006961774,0.0104994625,0.00037233287,0.0006446549,0.10727567,0.00094983226,0.56431055,0.010209064,0.28276736],"study_design_scores_gemma":[0.0008819366,0.0013480837,0.0061136265,0.0010936637,0.0029828572,0.00017288941,0.00012883866,0.18776758,0.0009630801,0.78470963,0.013732251,0.00010557084],"about_ca_topic_score_codex":0.0029381278,"about_ca_topic_score_gemma":0.0018869891,"teacher_disagreement_score":0.11314958,"about_ca_system_score_codex":0.0023335721,"about_ca_system_score_gemma":0.0028314942,"threshold_uncertainty_score":0.59839934},"labels":[],"label_agreement":null},{"id":"W1987309974","doi":"10.1002/sim.1528","title":"Estimating the size of an illicit‐drug‐using population","year":2003,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Census and Population Estimation","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Population; Confidence interval; Statistics; Computer science; Population size; Sampling (signal processing); Sample size determination; Selection bias; Interval (graph theory); Econometrics; Medicine; Mathematics; Environmental health","score_opus":0.05992622483707203,"score_gpt":0.39626810304054444,"score_spread":0.3363418782034724,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1987309974","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09795423,0.00037217492,0.8945882,0.00063045544,0.00009265058,0.00021733504,0.00087349495,0.00024066526,0.0050308197],"genre_scores_gemma":[0.6450376,0.00079352356,0.3491749,0.00014060202,0.00016155916,0.00051528367,0.0015279471,0.000062279236,0.0025863103],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9981869,0.000908926,0.00007582211,0.0002849661,0.0004824198,0.00006099925],"domain_scores_gemma":[0.9873563,0.009009227,0.0011394541,0.0013949597,0.0009324865,0.00016763089],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0043667373,0.000361343,0.0006065178,0.0022846067,0.00037058012,0.00085216906,0.0011240302,0.0005457407,0.0017602829],"category_scores_gemma":[0.02574008,0.0003116487,0.00046769768,0.00086714484,0.0005469449,0.0018677954,0.0015423181,0.0008204017,0.00036167298],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00018874322,0.00018894888,0.25333735,0.00027277463,0.00031030274,0.00029263442,0.0014127151,0.15010378,0.0045915325,0.17692569,0.007708614,0.40466693],"study_design_scores_gemma":[0.000068916626,0.00021820473,0.06467176,0.00015687768,0.00011658504,0.00087311864,0.00069147075,0.8032774,0.006296018,0.09788795,0.02563264,0.00010912408],"about_ca_topic_score_codex":0.0041071787,"about_ca_topic_score_gemma":0.0030958124,"teacher_disagreement_score":0.0043667373,"about_ca_system_score_codex":0.0006006211,"about_ca_system_score_gemma":0.0010248218,"threshold_uncertainty_score":0.02309382},"labels":[],"label_agreement":null},{"id":"W1987856141","doi":"10.1002/sim.986","title":"Maximum likelihood estimation of a survival function with a change point for truncated and interval‐censored data","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Statistics; Interval (graph theory); Point estimation; Estimation; Maximum likelihood; Interval estimation; Likelihood function; Survival function; Survival analysis; Confidence interval; Econometrics; Function (biology); Mathematics; Computer science; Biology; Combinatorics; Economics","score_opus":0.19020068248610117,"score_gpt":0.3979920695445226,"score_spread":0.20779138705842143,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1987856141","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.012410451,0.00030070246,0.98672336,0.00014164478,0.000016158494,0.000027887652,0.00005882751,0.00014482204,0.00017618985],"genre_scores_gemma":[0.4142025,0.0013807862,0.57952434,0.00017777755,0.00014456741,0.00043282603,0.0011467459,0.00015029908,0.0028401674],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99634415,0.002337377,0.00017793216,0.0005827485,0.0004078894,0.00014994998],"domain_scores_gemma":[0.96148914,0.034431033,0.001735854,0.00123844,0.0008628703,0.00024265348],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014190772,0.0008797315,0.0019103714,0.0020718765,0.00050975085,0.0013309473,0.002105576,0.0023706858,0.0015599048],"category_scores_gemma":[0.062329847,0.0007160215,0.0015863971,0.0019534484,0.0020444053,0.0030289479,0.0016701039,0.0024032313,0.0006271059],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004983541,0.00013906909,0.00942946,0.0003508222,0.00029658098,0.00060705526,0.00046890005,0.74529254,0.0021943217,0.097776204,0.0015655376,0.14138116],"study_design_scores_gemma":[0.000045166184,0.00008697728,0.0016466985,0.000046038956,0.0000359199,0.0001679352,0.000037607555,0.9270003,0.00081976125,0.069184214,0.00088677823,0.000042681448],"about_ca_topic_score_codex":0.0021228266,"about_ca_topic_score_gemma":0.0012204563,"teacher_disagreement_score":0.014190772,"about_ca_system_score_codex":0.0010920474,"about_ca_system_score_gemma":0.0009911541,"threshold_uncertainty_score":0.07504892},"labels":[],"label_agreement":null},{"id":"W1987973071","doi":"10.1002/sim.5941","title":"Graphical assessment of internal and external calibration of logistic regression models by using loess smoothers","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":357,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; University of Toronto; Public Health Ontario; Sunnybrook Health Science Centre; Institute for Clinical Evaluative Sciences","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Overfitting; Calibration; Logistic regression; Statistics; Computer science; Covariate; Loess; Econometrics; Mathematics; Artificial intelligence","score_opus":0.4241021929435487,"score_gpt":0.5779164790022724,"score_spread":0.15381428605872371,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1987973071","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04190331,0.0005035912,0.9541317,0.00040940096,0.0000654274,0.0001816116,0.00020904542,0.0018830677,0.0007128027],"genre_scores_gemma":[0.6047517,0.00045172736,0.39135528,0.00034823394,0.00007691968,0.0006768335,0.0007867817,0.0007790966,0.00077340094],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.95629126,0.036306717,0.0019057344,0.0020019007,0.002995259,0.0004992372],"domain_scores_gemma":[0.6342739,0.31665128,0.017140592,0.020585854,0.01048043,0.0008679633],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07874872,0.0017154074,0.0016880423,0.004584369,0.00083550735,0.003093741,0.0021327517,0.002363902,0.0034459417],"category_scores_gemma":[0.34131733,0.00082585774,0.0020362185,0.0024918588,0.0022523606,0.0032700112,0.002967285,0.00340351,0.0007052567],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024755665,0.00033673013,0.0852329,0.0016805675,0.0021196222,0.0007577693,0.0030538614,0.501369,0.008543511,0.075257964,0.0068680304,0.31230456],"study_design_scores_gemma":[0.00017174808,0.0006038301,0.015881723,0.00028587642,0.00027850145,0.000429949,0.00025964886,0.91329455,0.004890879,0.060331583,0.0033427794,0.00022891653],"about_ca_topic_score_codex":0.0018430776,"about_ca_topic_score_gemma":0.0015452146,"teacher_disagreement_score":0.07874872,"about_ca_system_score_codex":0.0010845285,"about_ca_system_score_gemma":0.0012358262,"threshold_uncertainty_score":0.41646802},"labels":[],"label_agreement":null},{"id":"W1988018581","doi":"10.1002/sim.1748","title":"Simultaneous inference for longitudinal data with detection limits and covariates measured with errors, with application to AIDS studies","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Covariate; Censoring (clinical trials); Inference; Statistics; Computer science; Statistical inference; Gibbs sampling; Data set; Monte Carlo method; Observational error; Econometrics; Mathematics; Artificial intelligence; Bayesian probability","score_opus":0.054640807936469755,"score_gpt":0.3630267823652393,"score_spread":0.3083859744287695,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1988018581","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00571062,0.0007430741,0.9929739,0.0002211578,0.000026193815,0.000037382713,0.000045289245,0.00011217773,0.00013020865],"genre_scores_gemma":[0.205718,0.0020323836,0.7894478,0.00027723017,0.00019490998,0.0005720062,0.00039609493,0.00012321176,0.0012384296],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9800562,0.015304223,0.00070893334,0.001954995,0.0017512967,0.00022430113],"domain_scores_gemma":[0.83078164,0.15596,0.0059066783,0.0043570665,0.0023623016,0.0006323728],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.038250294,0.0016230769,0.003187225,0.0040365015,0.0012340924,0.0024602693,0.0030588591,0.0027789855,0.001550303],"category_scores_gemma":[0.19498855,0.001977551,0.002154195,0.0050820643,0.0031610003,0.003891851,0.0042349985,0.003821776,0.0003080554],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000619256,0.00020817683,0.019532952,0.0010472666,0.0019299528,0.0007049298,0.0015936779,0.36989832,0.002454536,0.28576544,0.0026190034,0.31362656],"study_design_scores_gemma":[0.00012687214,0.000120495766,0.0031424763,0.00013966387,0.0002505513,0.00024764266,0.000096181364,0.673583,0.0009441442,0.3187867,0.0024920988,0.00007018891],"about_ca_topic_score_codex":0.005918398,"about_ca_topic_score_gemma":0.0052942983,"teacher_disagreement_score":0.038250294,"about_ca_system_score_codex":0.0013163809,"about_ca_system_score_gemma":0.0021605429,"threshold_uncertainty_score":0.20228934},"labels":[],"label_agreement":null},{"id":"W1990732953","doi":"10.1002/sim.2435","title":"Pseudo-likelihood methods for longitudinal binary data with non-ignorable missing responses and covariates","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital","funders":"National Cancer Institute; National Heart, Lung, and Blood Institute; National Institute of General Medical Sciences; U.S. Public Health Service","keywords":"Covariate; Missing data; Categorical variable; Statistics; Parametric statistics; Estimating equations; Econometrics; Expectation–maximization algorithm; Outcome (game theory); Mathematics; Computer science; Maximum likelihood","score_opus":0.12276464088284904,"score_gpt":0.48109581167973214,"score_spread":0.3583311707968831,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1990732953","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0006501646,0.00033825496,0.9982374,0.00017388344,0.000040958428,0.000087513574,0.00010238353,0.00019302322,0.00017652637],"genre_scores_gemma":[0.042711336,0.0010966813,0.94989336,0.000473108,0.00029491735,0.0020263488,0.0011295342,0.00046037356,0.0019143516],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9615839,0.033136345,0.0010872608,0.0017103618,0.0021720561,0.00031016808],"domain_scores_gemma":[0.8279058,0.1525535,0.006452598,0.00820517,0.0041029747,0.00077997555],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.046261713,0.0021570884,0.0033999325,0.003692439,0.0011249739,0.0032719541,0.007829505,0.003449407,0.0069108903],"category_scores_gemma":[0.15441099,0.00249916,0.003433299,0.004806914,0.0038659226,0.0059055625,0.0054306122,0.006201401,0.002726726],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00069016346,0.00024927242,0.006473437,0.0022408452,0.0011123444,0.0009441327,0.0013813204,0.2539168,0.0011120671,0.4843242,0.0080442,0.23951131],"study_design_scores_gemma":[0.00014767365,0.000107648186,0.00070853956,0.00024353678,0.000070863236,0.00030075674,0.0001088219,0.58914447,0.0005021893,0.40015027,0.008423256,0.000091941154],"about_ca_topic_score_codex":0.0032051879,"about_ca_topic_score_gemma":0.0033566437,"teacher_disagreement_score":0.046261713,"about_ca_system_score_codex":0.0015772065,"about_ca_system_score_gemma":0.003973496,"threshold_uncertainty_score":0.24465823},"labels":[],"label_agreement":null},{"id":"W1990838844","doi":"10.1002/sim.714","title":"Testing for elevated disease rates using smoothed estimates","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Computer science; Statistics; Econometrics; Mathematics","score_opus":0.2191181211047542,"score_gpt":0.4805154219191844,"score_spread":0.2613973008144302,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1990838844","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.13677289,0.00046593763,0.8591506,0.00032587678,0.000092376315,0.00014813279,0.0008577697,0.00082690496,0.0013595615],"genre_scores_gemma":[0.8115231,0.0002403186,0.1855285,0.00013806182,0.000116389936,0.00027927698,0.0011510391,0.00009332023,0.0009299379],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98685455,0.0075865802,0.0006436486,0.0019295115,0.0025758492,0.0004098242],"domain_scores_gemma":[0.90582234,0.076438904,0.006395436,0.007610302,0.00308531,0.0006476327],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.018026141,0.00046426238,0.001344179,0.003287236,0.000511906,0.0015456739,0.001206075,0.00097356923,0.002312556],"category_scores_gemma":[0.1278616,0.00037548656,0.0013142482,0.0022617553,0.0011200107,0.0011214772,0.0013111725,0.0017627463,0.00033558538],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012001005,0.00026612877,0.42355508,0.00050669286,0.004456538,0.001527422,0.0020024157,0.08868022,0.0078013614,0.10533734,0.006980898,0.3576859],"study_design_scores_gemma":[0.00023315723,0.0013290354,0.28646436,0.00020996813,0.00096313714,0.0016311691,0.00083800516,0.5093048,0.0064842384,0.17932759,0.0128949275,0.00031955735],"about_ca_topic_score_codex":0.0045736036,"about_ca_topic_score_gemma":0.003139284,"teacher_disagreement_score":0.018026141,"about_ca_system_score_codex":0.0004666236,"about_ca_system_score_gemma":0.0011751715,"threshold_uncertainty_score":0.0953325},"labels":[],"label_agreement":null},{"id":"W1991135418","doi":"10.1002/1097-0258(20001115)19:21<3017::aid-sim785>3.0.co;2-p","title":"M. D. deB. Edwardes, ‘The generalization of the odds ratio, risk ratio and risk difference tor ×k tables’.Statistics in Medicine 2000;19(14): 1901-1914","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Royal Victoria Hospital","funders":"","keywords":"Statistics; Generalization; Medical statistics; Odds ratio; Mathematics; Odds; Medicine; Logistic regression","score_opus":0.15515203872518482,"score_gpt":0.46718523153196945,"score_spread":0.31203319280678465,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1991135418","genre_codex":"commentary","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00060816464,0.18687928,0.023507748,0.52447706,0.25882322,0.000068519534,0.0015130513,0.00022653527,0.0038964634],"genre_scores_gemma":[0.031095462,0.28856182,0.04204096,0.29447314,0.30760163,0.0005344677,0.0016855297,0.00078444835,0.03322262],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97280407,0.013773788,0.004680549,0.0014914327,0.006846737,0.00040344696],"domain_scores_gemma":[0.89353,0.084677294,0.004304504,0.0027458386,0.013830196,0.00091227517],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03140064,0.0018923033,0.0019255962,0.0044147237,0.0013814311,0.0035809577,0.003401056,0.006249613,0.010352653],"category_scores_gemma":[0.26007706,0.0017575761,0.0020553363,0.005273039,0.0037057754,0.005887911,0.0016459198,0.012042724,0.0060580904],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000753061,0.000009281223,0.0007526723,0.00049907673,0.000074395415,0.00015922292,0.0002051995,0.0001399766,0.000053468513,0.01040976,0.954028,0.033593763],"study_design_scores_gemma":[0.00013767237,0.00006990523,0.008523375,0.0020592643,0.00017120455,0.0018185878,0.00027246602,0.001152577,0.00063438667,0.0718794,0.9131111,0.00017003738],"about_ca_topic_score_codex":0.016190518,"about_ca_topic_score_gemma":0.023042385,"teacher_disagreement_score":0.03140064,"about_ca_system_score_codex":0.002701622,"about_ca_system_score_gemma":0.0032074158,"threshold_uncertainty_score":0.16606444},"labels":[],"label_agreement":null},{"id":"W1991362225","doi":"10.1002/sim.3383","title":"Sample size calculation for cluster randomized cross‐over trials","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":105,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"","keywords":"Intraclass correlation; Sample size determination; Cluster (spacecraft); Statistics; Sample (material); Cluster analysis; Cross over; Randomized controlled trial; Computer science; Correlation; External validity; Econometrics; Mathematics; Medicine; Physics; Psychometrics; Internal medicine; Thermodynamics","score_opus":0.5602429666646603,"score_gpt":0.6200809675466604,"score_spread":0.059838000882000086,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1991362225","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.004636487,0.001884911,0.98103493,0.00084376376,0.0009858061,0.008015568,0.0002988452,0.00044755748,0.0018521443],"genre_scores_gemma":[0.08289186,0.00087326433,0.8702727,0.0012963283,0.0004069471,0.042849824,0.00040234218,0.00022778264,0.00077903847],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8549059,0.1205636,0.008418242,0.004275285,0.0111601325,0.00067686324],"domain_scores_gemma":[0.6914732,0.27854303,0.007524927,0.010816568,0.01072417,0.00091819983],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.14836892,0.0013850809,0.0038718893,0.004125003,0.0007435881,0.0015691686,0.0028226746,0.003243516,0.0055873916],"category_scores_gemma":[0.44724697,0.00093871815,0.00251355,0.0029749942,0.0019004216,0.001969222,0.0020910148,0.0038841832,0.00086261274],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008621068,0.00039328355,0.0054224418,0.0068720262,0.0026897283,0.0003746344,0.00084083475,0.033619177,0.002460405,0.18512966,0.025659518,0.72791725],"study_design_scores_gemma":[0.014110786,0.0080608865,0.0087548075,0.004664871,0.003621106,0.0013454695,0.00020198862,0.423248,0.014023504,0.43295616,0.08868615,0.00032620522],"about_ca_topic_score_codex":0.00065835007,"about_ca_topic_score_gemma":0.0005029361,"teacher_disagreement_score":0.14836892,"about_ca_system_score_codex":0.0017888935,"about_ca_system_score_gemma":0.0030958948,"threshold_uncertainty_score":0.78465927},"labels":[],"label_agreement":null},{"id":"W1992221362","doi":"10.1002/sim.3316","title":"Archimedean copula model selection under dependent truncation","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"","keywords":"Copula (linguistics); Mathematics; Truncation (statistics); Multivariate statistics; Statistics; Econometrics; Applied mathematics","score_opus":0.15964420102335394,"score_gpt":0.431878468074725,"score_spread":0.2722342670513711,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1992221362","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.042203803,0.00021292563,0.95643824,0.0001789467,0.000025873955,0.000072443334,0.00014701264,0.00016012466,0.0005606923],"genre_scores_gemma":[0.6119524,0.00080135907,0.38039842,0.00021152692,0.00014550987,0.0007363354,0.0015682091,0.00023177933,0.0039544236],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9938943,0.0043220804,0.00019873343,0.00066574314,0.0006295296,0.00028972403],"domain_scores_gemma":[0.97313875,0.019676415,0.0016230333,0.0024900078,0.002529929,0.00054182776],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016304245,0.0013248022,0.0025037692,0.0014996199,0.0009009495,0.0015560485,0.0026941395,0.0010937914,0.0015085607],"category_scores_gemma":[0.03898967,0.00080778374,0.002215658,0.0017873778,0.0012685142,0.0018092204,0.0018206864,0.0026010294,0.0006215593],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002832516,0.00009436405,0.0056489613,0.00011415851,0.00030214843,0.0006126526,0.00029326216,0.85588086,0.0014014278,0.08453651,0.002292994,0.048539355],"study_design_scores_gemma":[0.000015841422,0.000028806226,0.0005493184,0.0000061144638,0.000014897941,0.0000392995,0.000013462711,0.98322403,0.00018382691,0.015705042,0.00020689533,0.00001255521],"about_ca_topic_score_codex":0.006021926,"about_ca_topic_score_gemma":0.003618801,"teacher_disagreement_score":0.016304245,"about_ca_system_score_codex":0.0011007823,"about_ca_system_score_gemma":0.0019878186,"threshold_uncertainty_score":0.086226106},"labels":[],"label_agreement":null},{"id":"W1992435907","doi":"10.1002/1097-0258(20010130)20:2<249::aid-sim641>3.0.co;2-l","title":"Statistical inferences for a twin correlation with multinomial outcomes","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Multinomial distribution; Sample size determination; Statistics; Point estimation; Goodness of fit; Statistical inference; Confidence interval; Interval estimation; Computer science; Econometrics; Inference; Sample (material); Monte Carlo method; Mathematics; Artificial intelligence","score_opus":0.4665853261819578,"score_gpt":0.5871577886326694,"score_spread":0.12057246245071163,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1992435907","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.023397963,0.002039397,0.9668797,0.0020446316,0.00067025254,0.00077495637,0.00027655225,0.00023324996,0.0036832897],"genre_scores_gemma":[0.34634015,0.0017059627,0.6434217,0.0017892043,0.00056434656,0.0040233163,0.0004828143,0.00019586745,0.0014765568],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.70420146,0.25515628,0.008516368,0.014079901,0.0168369,0.001209122],"domain_scores_gemma":[0.30286002,0.63187176,0.019392973,0.03700995,0.007778268,0.0010870362],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.20154758,0.0016785358,0.0035970823,0.0048439545,0.0019420044,0.004145295,0.0043213344,0.0031050486,0.00929131],"category_scores_gemma":[0.6963966,0.0011064432,0.0047719395,0.00516095,0.008924891,0.0064203986,0.0071787955,0.0068318127,0.0011256434],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016130932,0.00023773756,0.06396892,0.0026309253,0.0057084477,0.001982495,0.0041022776,0.026624743,0.00079911726,0.60355276,0.010556657,0.27822283],"study_design_scores_gemma":[0.0004114386,0.00049019634,0.013799231,0.0010908351,0.0012112891,0.0011334387,0.0005963041,0.088536836,0.0009610116,0.88440704,0.007252789,0.00010950398],"about_ca_topic_score_codex":0.0018789674,"about_ca_topic_score_gemma":0.0014111894,"teacher_disagreement_score":0.20154758,"about_ca_system_score_codex":0.0015920936,"about_ca_system_score_gemma":0.002694579,"threshold_uncertainty_score":0.984634},"labels":[],"label_agreement":null},{"id":"W1993310646","doi":"10.1002/sim.857","title":"Evaluation of an adjusted chi‐square statistic as applied to observational studies involving clustered binary data","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":31,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Statistics; Statistic; Chi-square test; Binary data; Test statistic; Observational study; Ancillary statistic; Mathematics; Binary number; Square (algebra); Statistical hypothesis testing; Pearson's chi-squared test; Econometrics; PRESS statistic; F-test; Arithmetic","score_opus":0.547003564024309,"score_gpt":0.5393908008446755,"score_spread":0.007612763179633553,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1993310646","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.048268437,0.0017578952,0.94505745,0.0008789556,0.00045266506,0.0007934494,0.00053992314,0.00069703034,0.0015541771],"genre_scores_gemma":[0.38550118,0.00070781814,0.6103069,0.00041314863,0.00016817045,0.0016398357,0.0005207958,0.00020776577,0.0005343704],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.90084475,0.07757143,0.0056799953,0.003975238,0.011291064,0.0006374103],"domain_scores_gemma":[0.5997497,0.35081944,0.017707828,0.014880326,0.015455319,0.0013874111],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0815065,0.0009860069,0.0024467164,0.0047630775,0.00075660617,0.0022756357,0.0027873416,0.0021493768,0.0028070936],"category_scores_gemma":[0.39923444,0.00055000605,0.0019850149,0.007082167,0.0023241893,0.002024284,0.0016572935,0.0020751196,0.00048304512],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0043289405,0.0003543837,0.1745183,0.0045687635,0.007917035,0.0012901681,0.0024790878,0.0649214,0.011121664,0.09230794,0.008939078,0.6272533],"study_design_scores_gemma":[0.0015643094,0.0124865,0.21173696,0.0014865161,0.0036404436,0.0047674924,0.0018175306,0.5277156,0.029019326,0.15259199,0.052066,0.0011073292],"about_ca_topic_score_codex":0.002160296,"about_ca_topic_score_gemma":0.0016724501,"teacher_disagreement_score":0.0815065,"about_ca_system_score_codex":0.0011221718,"about_ca_system_score_gemma":0.0026595653,"threshold_uncertainty_score":0.43105268},"labels":[],"label_agreement":null},{"id":"W1993607274","doi":"10.1002/sim.1297","title":"The European regulatory experience","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hatch (Canada)","funders":"","keywords":"Clinical trial; Regulatory science; Margin (machine learning); Risk analysis (engineering); Work (physics); Medicine; Computer science; Management science; Economics","score_opus":0.5295102756246849,"score_gpt":0.563117157145275,"score_spread":0.033606881520590104,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1993607274","genre_codex":"other","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01620522,0.16901389,0.017696686,0.21386544,0.011280846,0.00012551027,0.00087044743,0.00040658502,0.57053536],"genre_scores_gemma":[0.38115758,0.1233837,0.024318475,0.24261215,0.008710132,0.00046414763,0.0016424843,0.0007447317,0.21696666],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.97983366,0.007520142,0.001662542,0.003223103,0.0056554684,0.002105046],"domain_scores_gemma":[0.97981334,0.008429044,0.0021469248,0.0023322972,0.0049460316,0.0023324194],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.038670465,0.0007948493,0.00086449576,0.002321945,0.0025281613,0.011108067,0.002213737,0.009614641,0.021443244],"category_scores_gemma":[0.03434172,0.00044472766,0.0013715331,0.003977688,0.0043546227,0.006154678,0.0044176965,0.006976529,0.0051719835],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002982896,0.0002682737,0.0018818463,0.00070963294,0.00005911023,0.00115183,0.0011906652,0.0009427852,0.000812405,0.5868773,0.15292062,0.25288722],"study_design_scores_gemma":[0.000028569199,0.00006465623,0.0012457265,0.000558952,0.000016150927,0.00068602775,0.00030227812,0.00013076334,0.00045555984,0.016483234,0.9800096,0.00001853731],"about_ca_topic_score_codex":0.0038394602,"about_ca_topic_score_gemma":0.0025924179,"teacher_disagreement_score":0.038670465,"about_ca_system_score_codex":0.00497061,"about_ca_system_score_gemma":0.010529608,"threshold_uncertainty_score":0.2045114},"labels":[],"label_agreement":null},{"id":"W1993793629","doi":"10.1002/sim.3807","title":"Bootstrap‐based methods for estimating standard errors in Cox's regression analyses of clustered event times","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Canadian Institutes of Health Research; Natural Sciences and Engineering Research Council of Canada; McGill University","keywords":"Statistics; Proportional hazards model; Regression; Event (particle physics); Regression analysis; Computer science; Standard error; Econometrics; Mathematics","score_opus":0.21099525133564637,"score_gpt":0.5786252017706595,"score_spread":0.3676299504350131,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1993793629","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011740276,0.0001939988,0.99802446,0.000072109375,0.00006580609,0.0000717875,0.000056197197,0.00017512264,0.00016654475],"genre_scores_gemma":[0.06526439,0.0007096004,0.9301292,0.0002066565,0.0002815443,0.0019444721,0.00042647222,0.00040640237,0.0006313228],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9652413,0.02691278,0.0012665913,0.0018067636,0.0043393187,0.0004332165],"domain_scores_gemma":[0.86014926,0.10867636,0.0070560277,0.014959307,0.008615694,0.0005433122],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.049015325,0.0014655057,0.0021267019,0.0040605,0.00090412283,0.001535358,0.0052766916,0.002052788,0.0032830145],"category_scores_gemma":[0.21839787,0.000979069,0.0026594226,0.0047525354,0.0017571326,0.001979582,0.0028827433,0.004268901,0.0012512917],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004575961,0.00024617117,0.010929201,0.0010690042,0.0017426261,0.0005232125,0.0013837421,0.16624336,0.0025996198,0.31763446,0.01224479,0.48492622],"study_design_scores_gemma":[0.00036775798,0.0003544964,0.0071620625,0.00053569465,0.00042102986,0.00043249043,0.00023491247,0.64088345,0.004669089,0.3254118,0.019238446,0.00028878573],"about_ca_topic_score_codex":0.0024040511,"about_ca_topic_score_gemma":0.0019947535,"teacher_disagreement_score":0.049015325,"about_ca_system_score_codex":0.0010016722,"about_ca_system_score_gemma":0.0024873954,"threshold_uncertainty_score":0.25922096},"labels":[],"label_agreement":null},{"id":"W1994343730","doi":"10.1002/sim.1723","title":"Three validation metrics for automated probabilistic image segmentation of brain tumours","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Medical Image Segmentation Techniques","field":"Computer Science","cited_by":95,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"National Center for Research Resources; U.S. National Library of Medicine; National Cancer Institute; Agency for Healthcare Research and Quality; National Institutes of Health","keywords":"Segmentation; Computer science; Artificial intelligence; Markov random field; Metric (unit); Similarity (geometry); Pixel; Pattern recognition (psychology); Gold standard (test); Image segmentation; Image (mathematics); Mathematics; Statistics","score_opus":0.027769904315747322,"score_gpt":0.3596305153886217,"score_spread":0.33186061107287435,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1994343730","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.35959768,0.0015809562,0.6354725,0.00020673576,0.000044600423,0.00033741747,0.00029224646,0.0013494653,0.001118424],"genre_scores_gemma":[0.7264544,0.00015922639,0.27201292,0.000056521843,0.000020955868,0.00036110127,0.00058308063,0.00014281744,0.00020894594],"study_design_codex":"simulation_or_modeling","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9755824,0.013666201,0.0022959223,0.0017901434,0.0060541555,0.0006111562],"domain_scores_gemma":[0.8091411,0.14953092,0.015382337,0.008478515,0.016534109,0.00093299436],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.045761913,0.00138324,0.0015266867,0.00582204,0.0011596605,0.0019087229,0.0020485364,0.002850899,0.00038405386],"category_scores_gemma":[0.16912651,0.0006635835,0.0012471748,0.0021781267,0.0024656856,0.0018124852,0.0024534948,0.0011572328,0.00016287906],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0022716816,0.00031026424,0.07433685,0.0007108554,0.0008144088,0.00028423913,0.0010334761,0.61977684,0.024649847,0.013364675,0.0014726127,0.2609742],"study_design_scores_gemma":[0.000051386323,0.00038876964,0.019700546,0.00008075312,0.00005670908,0.0003283728,0.00007370515,0.95574975,0.018821558,0.004112745,0.0005164807,0.00011921932],"about_ca_topic_score_codex":0.002962226,"about_ca_topic_score_gemma":0.0029919127,"teacher_disagreement_score":0.045761913,"about_ca_system_score_codex":0.0024473788,"about_ca_system_score_gemma":0.0016574387,"threshold_uncertainty_score":0.242015},"labels":[],"label_agreement":null},{"id":"W1997239045","doi":"10.1002/sim.980","title":"A three‐stage clinical trial design for rare disorders","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sunnybrook Health Science Centre; Hospital for Sick Children; University of Toronto","funders":"","keywords":"Randomized controlled trial; Stage (stratigraphy); Clinical trial; Medicine; Sample size determination; Placebo; Research design; Clinical study design; N of 1 trial; Completely randomized design; Randomization; Physical therapy; Surgery; Internal medicine; Statistics; Alternative medicine; Mathematics; Pathology","score_opus":0.6420867572546721,"score_gpt":0.5536282516752941,"score_spread":0.08845850557937807,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1997239045","genre_codex":"protocol","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03556904,0.0017226419,0.34257874,0.002709357,0.0043015867,0.60644984,0.0008997134,0.0007841493,0.004985012],"genre_scores_gemma":[0.06661245,0.000482559,0.22794187,0.0019132013,0.00037063216,0.70067483,0.00026080213,0.000043426495,0.0017002511],"study_design_codex":"randomized_trial","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.82459986,0.15039408,0.009713602,0.0052230274,0.0080079585,0.002061454],"domain_scores_gemma":[0.9036601,0.059086602,0.009141569,0.015569645,0.009446966,0.0030951772],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.15341294,0.0024904758,0.0052041123,0.002187399,0.001647325,0.0017373635,0.0023723962,0.0055772406,0.01611356],"category_scores_gemma":[0.1687611,0.0014811906,0.004018688,0.0018834288,0.0032518825,0.0025950565,0.002387158,0.0051442166,0.0031161222],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.5071014,0.014755682,0.008767137,0.014035264,0.00532895,0.0007259594,0.002124863,0.026836038,0.014080209,0.10711822,0.020437997,0.27868822],"study_design_scores_gemma":[0.54052794,0.19150649,0.008631131,0.0017045308,0.0037858507,0.0007598615,0.00016497647,0.07497569,0.007228259,0.09024982,0.07972886,0.00073664315],"about_ca_topic_score_codex":0.00038633126,"about_ca_topic_score_gemma":0.0006018256,"teacher_disagreement_score":0.84658706,"about_ca_system_score_codex":0.0013760187,"about_ca_system_score_gemma":0.0057167467,"threshold_uncertainty_score":0.81133485},"labels":[],"label_agreement":null},{"id":"W1997256514","doi":"10.1002/sim.5572","title":"Comments on the FDA draft guidance on biosimilar products","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Biosimilars and Bioanalytical Methods","field":"Immunology and Microbiology","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Interchangeability; Biosimilar; Food and drug administration; Risk analysis (engineering); Computer science; Scientific evidence; Management science; Operations research; Business; Medicine; Engineering; Statistics","score_opus":0.051552394930830296,"score_gpt":0.3530185489860644,"score_spread":0.3014661540552341,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1997256514","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00026261457,0.0027031389,0.0008875338,0.9314507,0.04658368,0.00022513616,0.0011255441,0.00018946364,0.016572163],"genre_scores_gemma":[0.0023137613,0.0030227685,0.0028033156,0.9197748,0.027478449,0.00034793012,0.00074582535,0.0002239863,0.043289125],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.94780546,0.010250399,0.0062794327,0.0023453743,0.030333053,0.002986449],"domain_scores_gemma":[0.8271221,0.07221495,0.010015113,0.0034605842,0.08277821,0.004409046],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.038198553,0.0018073466,0.0021006353,0.0029444923,0.0045438376,0.006948795,0.0054605487,0.051132828,0.015201786],"category_scores_gemma":[0.175425,0.0014011913,0.0041411296,0.0037533178,0.0043092608,0.0051846122,0.0023881553,0.026871907,0.019433945],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000017425593,0.000008249403,0.00002741446,0.00004393378,0.0000023492355,0.00005026217,0.00005403468,0.000020357567,0.000049452403,0.00084670575,0.99653316,0.002346562],"study_design_scores_gemma":[0.000046539237,0.000039060717,0.00040740534,0.0003226304,0.000016983771,0.00010923917,0.0001720556,0.00009446535,0.00019372626,0.0015223153,0.99704355,0.000032009328],"about_ca_topic_score_codex":0.03948423,"about_ca_topic_score_gemma":0.035555914,"teacher_disagreement_score":0.96180147,"about_ca_system_score_codex":0.011141844,"about_ca_system_score_gemma":0.025933415,"threshold_uncertainty_score":0.20201564},"labels":[],"label_agreement":null},{"id":"W1998669587","doi":"10.1002/sim.1227","title":"Assessing the gain in diagnostic performance when combining two diagnostic tests","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":99,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McMaster University","funders":"","keywords":"Statistics; Likelihood-ratio test; Test (biology); Sensitivity (control systems); Component (thermodynamics); Score test; Diagnostic test; Computer science; Econometrics; Mathematics; Medicine","score_opus":0.44492644498609113,"score_gpt":0.5605836017274896,"score_spread":0.11565715674139843,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1998669587","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.43496835,0.020537708,0.5176272,0.005318824,0.0013537717,0.0020712835,0.0023212698,0.0014130478,0.014388606],"genre_scores_gemma":[0.74298805,0.002161409,0.25098315,0.0006930456,0.0007035287,0.0007104097,0.0008003726,0.0000974596,0.0008625216],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.86135304,0.09885579,0.010100437,0.0075518596,0.020950397,0.001188442],"domain_scores_gemma":[0.5749154,0.38731745,0.0155958785,0.012927231,0.008297751,0.0009463022],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.13818043,0.0020757616,0.0037011902,0.0086742425,0.00046277273,0.0040344107,0.0016236787,0.00309168,0.0027287328],"category_scores_gemma":[0.33851618,0.0009067523,0.0032886986,0.0050297333,0.002792901,0.003770712,0.0038915775,0.0025977336,0.0010202597],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.011517345,0.00073725235,0.2834662,0.003806985,0.015794612,0.0007101262,0.00093833747,0.02703327,0.013533837,0.007117192,0.0033162218,0.63202876],"study_design_scores_gemma":[0.0024548378,0.03192256,0.4271106,0.0021111832,0.03495716,0.01328933,0.0019935763,0.27477345,0.05862703,0.12694189,0.024498481,0.0013199834],"about_ca_topic_score_codex":0.00037726513,"about_ca_topic_score_gemma":0.0005288591,"teacher_disagreement_score":0.13818043,"about_ca_system_score_codex":0.0010203696,"about_ca_system_score_gemma":0.0012581012,"threshold_uncertainty_score":0.7307767},"labels":[],"label_agreement":null},{"id":"W1999968123","doi":"10.1002/sim.6043","title":"A marginal structural model for multiple-outcome survival data:assessing the impact of injection drug use on several causes of death in the Canadian Co-infection Cohort","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Liver Disease Diagnosis and Treatment","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University Health Centre; McGill University","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Marginal structural model; Inverse probability weighting; Confounding; Econometrics; Inverse probability; Estimator; Statistics; Weighting; Proportional hazards model; Outcome (game theory); Cohort; Medicine; Computer science; Mathematics; Bayesian probability","score_opus":0.09557594705627659,"score_gpt":0.41972797201469403,"score_spread":0.32415202495841744,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1999968123","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.26249146,0.000834556,0.72490937,0.003291978,0.00015244585,0.0008127759,0.0030116579,0.000542765,0.0039529228],"genre_scores_gemma":[0.90998536,0.00065941026,0.08112343,0.00027515623,0.000111976435,0.0007194677,0.0017256426,0.000050988278,0.005348563],"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","domain_scores_codex":[0.9928659,0.0044494434,0.0001944009,0.00088756817,0.0009169674,0.00068579573],"domain_scores_gemma":[0.97323453,0.020772325,0.0019955232,0.0016678891,0.001694665,0.00063506566],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.018355008,0.001137521,0.0017445575,0.001979047,0.00089778577,0.0016052283,0.004339035,0.0014814467,0.0046026767],"category_scores_gemma":[0.03901114,0.0006386699,0.0025791852,0.0019798637,0.0033054953,0.0012116133,0.0028179341,0.0021283086,0.00035675752],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00133656,0.00023243413,0.08102608,0.0003760344,0.0010712076,0.0008176058,0.0015861361,0.4846061,0.0013785057,0.34698826,0.004047595,0.07653346],"study_design_scores_gemma":[0.00020788207,0.00041463738,0.018454514,0.00009103529,0.0005269663,0.0002618197,0.00033444498,0.8950538,0.00043749265,0.08137071,0.0027370062,0.000109688],"about_ca_topic_score_codex":0.18670443,"about_ca_topic_score_gemma":0.13812982,"teacher_disagreement_score":0.8132956,"about_ca_system_score_codex":0.0049774544,"about_ca_system_score_gemma":0.007286587,"threshold_uncertainty_score":0.37123537},"labels":[],"label_agreement":null},{"id":"W2001885105","doi":"10.1002/sim.6235","title":"The estimation of calibration equations for variables with heteroscedastic measurement errors","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Canada","funders":"National Institutes of Health","keywords":"Heteroscedasticity; Calibration; Observational error; Transformation (genetics); Statistics; Computer science; Sample size determination; Set (abstract data type); Statistical hypothesis testing; Mathematics; Applied mathematics; Algorithm","score_opus":0.14812013680284633,"score_gpt":0.43135458447398406,"score_spread":0.2832344476711377,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2001885105","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0016101488,0.00024674812,0.9976841,0.00008801474,0.000023827532,0.000049814807,0.000031657608,0.00007080456,0.00019487948],"genre_scores_gemma":[0.14148884,0.0023900555,0.8508206,0.00032122093,0.00020743933,0.0013222649,0.0005305438,0.00018159543,0.0027373782],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98253644,0.011007002,0.0007028211,0.0024953699,0.0029296116,0.0003289056],"domain_scores_gemma":[0.93785286,0.05025831,0.0046987967,0.0027604417,0.004233297,0.00019626494],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.037003916,0.0019248073,0.0021443286,0.0025926854,0.0007394623,0.0017198913,0.003391654,0.0023917696,0.0027197849],"category_scores_gemma":[0.10891767,0.0011783269,0.002130429,0.0038579318,0.0027466593,0.0034855537,0.0032299166,0.0059387605,0.0009735572],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012108277,0.00011762601,0.008707524,0.00074538577,0.00045224853,0.00029098673,0.0007459881,0.3970089,0.0022185196,0.33151457,0.002961865,0.25511527],"study_design_scores_gemma":[0.000041974694,0.000082116414,0.0027025074,0.00015225768,0.000092443326,0.0001856171,0.000060025137,0.86390233,0.001774548,0.12618753,0.004737987,0.000080738224],"about_ca_topic_score_codex":0.003299611,"about_ca_topic_score_gemma":0.0024619424,"teacher_disagreement_score":0.037003916,"about_ca_system_score_codex":0.001662686,"about_ca_system_score_gemma":0.00271446,"threshold_uncertainty_score":0.19569778},"labels":[],"label_agreement":null},{"id":"W2002271480","doi":"10.1002/sim.1477","title":"Misclassification in a matched case‐control study with variable matching ratio: application to a study of c‐erbB‐2 overexpression and breast cancer","year":2003,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Mount Sinai Hospital","funders":"","keywords":"Odds ratio; Case-control study; Breast cancer; Risk factor; Statistics; Matching (statistics); Covariate; Medicine; Breast disease; Confidence interval; Oncology; Internal medicine; Mathematics; Computer science; Cancer","score_opus":0.15994251799571924,"score_gpt":0.5014059989025446,"score_spread":0.3414634809068253,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2002271480","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.057854056,0.0018203831,0.9378291,0.001110919,0.00031676443,0.00026029718,0.00008951728,0.0002491892,0.00046977607],"genre_scores_gemma":[0.50789475,0.0011452882,0.48774505,0.0008570735,0.0003622945,0.00074563565,0.00016741507,0.000072464216,0.001009952],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9219283,0.06677903,0.0026811042,0.0030266575,0.0052327337,0.00035214177],"domain_scores_gemma":[0.6866888,0.27806866,0.013784022,0.017802281,0.0030617565,0.000594538],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10191247,0.0008890413,0.0018716655,0.0022788122,0.0007865765,0.0017586071,0.003230474,0.0026107808,0.0015506351],"category_scores_gemma":[0.29470348,0.00067032716,0.0021930442,0.0029632754,0.0024699066,0.0015600206,0.00225693,0.0018162783,0.00021140507],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004328326,0.00043761428,0.16075496,0.001334729,0.0040969416,0.0032107648,0.0022720122,0.10684802,0.0025971802,0.17703766,0.004769271,0.5323126],"study_design_scores_gemma":[0.0016034426,0.0027836806,0.03778105,0.00042592705,0.0027398325,0.005707274,0.0003452716,0.66665965,0.0045483136,0.26857954,0.008538004,0.0002880682],"about_ca_topic_score_codex":0.0016176932,"about_ca_topic_score_gemma":0.0012343685,"teacher_disagreement_score":0.8980875,"about_ca_system_score_codex":0.00093139184,"about_ca_system_score_gemma":0.000936137,"threshold_uncertainty_score":0.53897107},"labels":[],"label_agreement":null},{"id":"W2004052735","doi":"10.1002/sim.1798","title":"Estimating episode lengths when some observations are probably censored","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Healthcare Policy and Management","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Memorial University of Newfoundland","funders":"National Institute on Drug Abuse","keywords":"Censoring (clinical trials); Estimator; Resampling; Statistics; Computer science; Econometrics; Monte Carlo method; Survival analysis; Mathematics","score_opus":0.08770789891825295,"score_gpt":0.3170221799295941,"score_spread":0.22931428101134113,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2004052735","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4998534,0.0018765945,0.49293736,0.001163453,0.0000941878,0.00018581741,0.0017396162,0.0003521067,0.0017974318],"genre_scores_gemma":[0.91939837,0.00068609376,0.07492628,0.00014396195,0.00020432862,0.00018745886,0.0029553222,0.000050409908,0.0014478005],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9964676,0.0015529282,0.00043045156,0.00079579715,0.00044771875,0.00030547785],"domain_scores_gemma":[0.9067554,0.07430852,0.012062337,0.0034752337,0.0021063357,0.0012921144],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010019835,0.00040517555,0.00097375014,0.0017962364,0.00057817524,0.0015682253,0.0013308922,0.0015606789,0.0019058953],"category_scores_gemma":[0.085038364,0.00068848336,0.00067933346,0.0017590767,0.000564882,0.0033186087,0.0012153307,0.0018736449,0.00044054407],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015541605,0.00033900028,0.50201994,0.00049275777,0.0008329745,0.00095176225,0.0013835123,0.22457874,0.0014983235,0.04460335,0.0044163717,0.21732911],"study_design_scores_gemma":[0.00013219244,0.00030266854,0.09233153,0.00017030044,0.00031308617,0.0006801196,0.00068188936,0.7874445,0.002347027,0.11038416,0.0050854688,0.00012704048],"about_ca_topic_score_codex":0.004763399,"about_ca_topic_score_gemma":0.0053710015,"teacher_disagreement_score":0.010019835,"about_ca_system_score_codex":0.001091622,"about_ca_system_score_gemma":0.0009375303,"threshold_uncertainty_score":0.052990615},"labels":[],"label_agreement":null},{"id":"W2004536301","doi":"10.1002/sim.3875","title":"On Bayesian shared component disease mapping and ecological regression with errors in covariates","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":36,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Child and Family Research Institute; University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Child and Family Research Institute","keywords":"Multivariate statistics; Covariate; Bayesian probability; Context (archaeology); Computer science; Markov chain Monte Carlo; Bayesian inference; Prior probability; Bayesian hierarchical modeling; Autoregressive model; Econometrics; Statistics; Machine learning; Artificial intelligence; Geography; Mathematics","score_opus":0.046227544003509174,"score_gpt":0.3725131808067419,"score_spread":0.32628563680323275,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2004536301","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0028034358,0.0016648264,0.9927492,0.0008553748,0.00004033609,0.000018664661,0.00009395197,0.000049152044,0.0017250363],"genre_scores_gemma":[0.2662813,0.015171066,0.7062295,0.0014062258,0.0009286743,0.00049917924,0.00081022014,0.00021832308,0.008455515],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98956573,0.007886868,0.0002669275,0.0010945571,0.0009577885,0.00022808908],"domain_scores_gemma":[0.967481,0.028424192,0.0013655714,0.0014541174,0.0010726367,0.00020249092],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017077686,0.0012342663,0.0017964015,0.0025380962,0.00070473365,0.0019475558,0.002453587,0.0019564126,0.0032888472],"category_scores_gemma":[0.049341995,0.0008721053,0.0015844912,0.0048529226,0.0043551787,0.0029131589,0.0031188563,0.0026757976,0.00058743265],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000019231742,0.000020059984,0.0020250373,0.00019592371,0.0001318058,0.00010523086,0.00026510918,0.15275155,0.00023066023,0.7831948,0.001528415,0.05953216],"study_design_scores_gemma":[0.00001082981,0.000026403357,0.0011578217,0.00011125455,0.000040982628,0.00009426521,0.000042029107,0.21859145,0.00013196643,0.7724273,0.0073274774,0.000038177866],"about_ca_topic_score_codex":0.01911893,"about_ca_topic_score_gemma":0.014224357,"teacher_disagreement_score":0.01911893,"about_ca_system_score_codex":0.0022710662,"about_ca_system_score_gemma":0.0028245396,"threshold_uncertainty_score":0.090316534},"labels":[],"label_agreement":null},{"id":"W2004618413","doi":"10.1002/sim.1030","title":"A measure of continuity of care based on the multiplicative intensity model","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Healthcare Policy and Management","field":"Economics, Econometrics and Finance","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Alberta","funders":"Natural Sciences and Engineering Research Council of Canada; Alberta Heritage Foundation for Medical Research","keywords":"Multiplicative function; Measure (data warehouse); Continuity of care; Covariate; Computer science; Health care; Intensity (physics); Econometrics; Mathematics; Data mining; Machine learning; Economics","score_opus":0.09829028965133782,"score_gpt":0.30572034029172646,"score_spread":0.20743005064038864,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2004618413","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1418724,0.0006804916,0.8480337,0.002162857,0.00010823615,0.0003396517,0.0007820211,0.00017646477,0.00584414],"genre_scores_gemma":[0.8662221,0.00050928653,0.1295313,0.00024152371,0.00025917566,0.00042928164,0.00068890175,0.000035573787,0.002082852],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99333405,0.0026105093,0.00044989816,0.001064076,0.0020835712,0.00045781507],"domain_scores_gemma":[0.97933584,0.011714672,0.0043132217,0.0022040172,0.0015493257,0.000882953],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0085401265,0.0004600497,0.0012594435,0.0028854734,0.0008416858,0.002228781,0.0033841275,0.0017596605,0.0027217735],"category_scores_gemma":[0.038501523,0.00039775344,0.0015398975,0.0035520447,0.0015455399,0.0038092032,0.0022413095,0.002403227,0.00026931067],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00032926555,0.00059330795,0.11347996,0.0003385132,0.00079374155,0.00034347072,0.0016340225,0.20400096,0.0029009285,0.5457344,0.0056476016,0.12420386],"study_design_scores_gemma":[0.00011445306,0.00058183237,0.054966312,0.000110344845,0.00037528362,0.0008774153,0.00056847266,0.71263915,0.0011994813,0.22127394,0.0071283826,0.00016489965],"about_ca_topic_score_codex":0.0048399633,"about_ca_topic_score_gemma":0.003121937,"teacher_disagreement_score":0.99145985,"about_ca_system_score_codex":0.0028997643,"about_ca_system_score_gemma":0.0021555729,"threshold_uncertainty_score":0.045165002},"labels":[],"label_agreement":null},{"id":"W2005293181","doi":"10.1002/sim.5606","title":"A generalized Jonckheere test against ordered alternatives for repeated measures in randomized blocks","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Acadia University","funders":"","keywords":"Null hypothesis; Null (SQL); Mathematics; Statistical hypothesis testing; Statistics; Randomized controlled trial; Applied mathematics; Null distribution; Block (permutation group theory); Computer science; Econometrics; Medicine; Test statistic; Data mining; Internal medicine; Combinatorics","score_opus":0.401823268684664,"score_gpt":0.5537626732144015,"score_spread":0.15193940452973753,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2005293181","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008085272,0.00043220966,0.9879816,0.00044417699,0.0002245379,0.0007825723,0.0002310715,0.00020369599,0.0016149009],"genre_scores_gemma":[0.12896505,0.00060470105,0.8608922,0.00085687794,0.00029232443,0.0063079516,0.0003759444,0.00011452127,0.0015905474],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8724991,0.10650672,0.0038295486,0.0049424493,0.010986818,0.0012353684],"domain_scores_gemma":[0.7728987,0.19685005,0.007989006,0.01315284,0.007677202,0.0014322242],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07566326,0.001442708,0.004462375,0.0033917977,0.0008145727,0.0020912427,0.0040180716,0.0032321387,0.010661138],"category_scores_gemma":[0.24566293,0.00058154407,0.0030192954,0.0036033392,0.004149393,0.0032929545,0.0027953854,0.0037622706,0.0015700863],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024543384,0.0005969807,0.007873569,0.0022381425,0.0016896209,0.0006953891,0.0010454858,0.043930065,0.003693656,0.5096281,0.006364879,0.41978973],"study_design_scores_gemma":[0.0021808,0.00714629,0.00844945,0.0008478359,0.0007457417,0.0012685114,0.00037736277,0.23880334,0.0047463793,0.7170502,0.017974636,0.0004094668],"about_ca_topic_score_codex":0.0009608078,"about_ca_topic_score_gemma":0.0010070393,"teacher_disagreement_score":0.07566326,"about_ca_system_score_codex":0.0012457127,"about_ca_system_score_gemma":0.004434904,"threshold_uncertainty_score":0.40015036},"labels":[],"label_agreement":null},{"id":"W2005628926","doi":"10.1002/sim.4182","title":"Estimation of reliability in a three‐factor model","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Northern British Columbia","funders":"","keywords":"Intraclass correlation; Reliability (semiconductor); Statistics; Interval estimation; Point estimation; Confidence interval; Markov chain Monte Carlo; Markov chain; Variance (accounting); Monte Carlo method; Random effects model; Estimation; Econometrics; Computer science; Mathematics; Medicine; Psychometrics; Meta-analysis","score_opus":0.14374485287821784,"score_gpt":0.4220546455536928,"score_spread":0.27830979267547495,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2005628926","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02369242,0.0004544972,0.9746552,0.00029162026,0.000030680698,0.00008373382,0.00013674967,0.00014686104,0.00050823786],"genre_scores_gemma":[0.5995353,0.0015072606,0.39496818,0.00018922087,0.00018478838,0.0008975262,0.0008351524,0.00008989332,0.0017925967],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96812606,0.0241127,0.0010538409,0.0032672381,0.0026919574,0.0007482999],"domain_scores_gemma":[0.9052067,0.08243018,0.0043640505,0.004989626,0.0025674498,0.0004419956],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.030978423,0.0015642957,0.0030357838,0.002615155,0.00060678576,0.0022133621,0.002942891,0.0024087767,0.0020350204],"category_scores_gemma":[0.0953797,0.0011903147,0.002662918,0.0033268488,0.0029637292,0.0024641964,0.002288818,0.0028241137,0.0009206224],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00055645616,0.00029374287,0.02188746,0.0007973904,0.0017466984,0.0005148759,0.0019740185,0.49302584,0.0016610883,0.31995058,0.0031532964,0.1544386],"study_design_scores_gemma":[0.00006223841,0.00018922643,0.005470328,0.00012888994,0.00014507219,0.00015495995,0.00011098489,0.7792011,0.000308274,0.21268988,0.0014576735,0.00008135826],"about_ca_topic_score_codex":0.0060707456,"about_ca_topic_score_gemma":0.002414749,"teacher_disagreement_score":0.030978423,"about_ca_system_score_codex":0.0013442463,"about_ca_system_score_gemma":0.0023543872,"threshold_uncertainty_score":0.16383153},"labels":[],"label_agreement":null},{"id":"W2006136142","doi":"10.1002/sim.5673","title":"A<i>K</i>‐nearest neighbors survival probability prediction method","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Institutes of Health","keywords":"Covariate; Proportional hazards model; Survival analysis; Statistics; Estimator; Nonparametric statistics; Metric (unit); Data set; Mathematics; Similarity (geometry); Computer science; Set (abstract data type); Artificial intelligence","score_opus":0.10052311971365267,"score_gpt":0.4154534579781773,"score_spread":0.3149303382645246,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2006136142","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0073776185,0.00019824898,0.9906709,0.000123654,0.00007038862,0.0000736605,0.00021509975,0.00032526304,0.00094516267],"genre_scores_gemma":[0.3186185,0.00031519184,0.67499954,0.00022591618,0.00020658941,0.00040181685,0.0010396476,0.000123862,0.0040689837],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9982356,0.0007219817,0.00010656839,0.00034496482,0.0004756069,0.00011518165],"domain_scores_gemma":[0.99697685,0.0017777141,0.00021361915,0.00043512275,0.0005045823,0.00009213442],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0034471636,0.0003644592,0.0008476889,0.0010168004,0.00059035193,0.00066143577,0.002082729,0.0008427897,0.0036300817],"category_scores_gemma":[0.010056616,0.0002682889,0.00081175764,0.00085030234,0.00051515055,0.0010056765,0.0007789788,0.0012323909,0.0010499601],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003228224,0.00027320284,0.014128617,0.00022174534,0.0001885941,0.0002589935,0.00015450548,0.2821543,0.00337585,0.050214447,0.014686412,0.63402045],"study_design_scores_gemma":[0.00003342516,0.000046886496,0.0014634451,0.000028465096,0.000021498696,0.00019349268,0.0000207371,0.96971434,0.001194715,0.023508547,0.0037421235,0.000032282784],"about_ca_topic_score_codex":0.003992426,"about_ca_topic_score_gemma":0.0036712843,"teacher_disagreement_score":0.003992426,"about_ca_system_score_codex":0.00039240072,"about_ca_system_score_gemma":0.0010359831,"threshold_uncertainty_score":0.018230557},"labels":[],"label_agreement":null},{"id":"W2007493872","doi":"10.1002/sim.2737","title":"Biased odds ratios from dichotomization of age","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":74,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Alberta Health Services; Columbia College","funders":"Eunice Kennedy Shriver National Institute of Child Health and Human Development; National Institute of Mental Health","keywords":"Odds; Odds ratio; Statistics; Demography; Medicine; Econometrics; Computer science; Logistic regression; Mathematics","score_opus":0.05253306517601675,"score_gpt":0.3872355764733088,"score_spread":0.33470251129729206,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2007493872","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.13495298,0.011277853,0.8284968,0.01178915,0.0007995826,0.00051406166,0.00092769664,0.0005678055,0.010674017],"genre_scores_gemma":[0.8321137,0.0026204772,0.15678549,0.0050027617,0.0006423759,0.00073515845,0.00054512144,0.00018675157,0.0013682053],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.903351,0.08016694,0.002715092,0.005729166,0.0069417977,0.0010959784],"domain_scores_gemma":[0.5620477,0.38887975,0.020514537,0.022330955,0.0053621246,0.00086490094],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08406337,0.001136472,0.0025190758,0.0027594063,0.00077208114,0.003318683,0.0025348992,0.0026012843,0.0034476519],"category_scores_gemma":[0.44553235,0.00074810244,0.0020679976,0.0026906736,0.0063459203,0.0058108675,0.0035584962,0.0044627544,0.00066708174],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0045506917,0.0001713406,0.1777812,0.002009297,0.0030407987,0.0020747136,0.004361209,0.037382055,0.0023373363,0.47771877,0.011993074,0.27657956],"study_design_scores_gemma":[0.00033857292,0.00030064603,0.031125657,0.0006886202,0.00077903667,0.0024162184,0.00070535485,0.07047093,0.0027607097,0.8825906,0.007678115,0.00014559229],"about_ca_topic_score_codex":0.001436657,"about_ca_topic_score_gemma":0.00092115137,"teacher_disagreement_score":0.91593665,"about_ca_system_score_codex":0.001681373,"about_ca_system_score_gemma":0.0012678572,"threshold_uncertainty_score":0.44457495},"labels":[],"label_agreement":null},{"id":"W2011410147","doi":"10.1002/sim.6356","title":"A model for time to fracture with a shock stream superimposed on progressive degradation: the Study of Osteoporotic Fractures","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bone health and osteoporosis research","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Ottawa Hospital","funders":"Eunice Kennedy Shriver National Institute of Child Health and Human Development; National Institute of Arthritis and Musculoskeletal and Skin Diseases; National Institute on Aging; University of Maryland","keywords":"Hip fracture; Fracture (geology); Event (particle physics); Shock (circulatory); Computer science; Econometrics; Perspective (graphical); Statistics; Medicine; Osteoporosis; Mathematics; Artificial intelligence; Engineering; Internal medicine","score_opus":0.027479904661868674,"score_gpt":0.3744236494380615,"score_spread":0.34694374477619283,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2011410147","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.37471214,0.0023513853,0.5905292,0.0133425165,0.00044150435,0.0003040156,0.0028081287,0.00048694358,0.015024187],"genre_scores_gemma":[0.9499605,0.0018572778,0.020425428,0.00059108454,0.00033568675,0.00046094402,0.001008492,0.00009372244,0.02526686],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9987349,0.0005549957,0.00006840527,0.0002600218,0.00014618628,0.00023543405],"domain_scores_gemma":[0.9922044,0.005210729,0.0011392881,0.00021854167,0.0005522595,0.0006748449],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0035945587,0.0010672803,0.0016514363,0.0014698949,0.0008118899,0.0022360394,0.0031911617,0.0038317882,0.006422994],"category_scores_gemma":[0.013811495,0.0008089276,0.0015796105,0.0013714078,0.0021380535,0.0021446755,0.0020615277,0.00282658,0.0008585197],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023681937,0.00017710803,0.011045985,0.00015060205,0.0001867997,0.0007280255,0.00046752483,0.73852515,0.0011843838,0.23943014,0.0027373119,0.0051301774],"study_design_scores_gemma":[0.00010146246,0.000115256575,0.0013995303,0.00002442751,0.00004998071,0.00010301967,0.00011596341,0.9417867,0.000075618766,0.05516604,0.0010180369,0.00004389884],"about_ca_topic_score_codex":0.023082688,"about_ca_topic_score_gemma":0.008462779,"teacher_disagreement_score":0.023082688,"about_ca_system_score_codex":0.0017335147,"about_ca_system_score_gemma":0.0020554585,"threshold_uncertainty_score":0.04589665},"labels":[],"label_agreement":null},{"id":"W2011513835","doi":"10.1002/sim.5436","title":"Misspecification of Cox regression models with composite endpoints","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Econometrics; Estimator; Copula (linguistics); Censoring (clinical trials); Proportional hazards model; Multivariate statistics; Statistics; Context (archaeology); Regression; Mathematics","score_opus":0.530858385765817,"score_gpt":0.5711734862838657,"score_spread":0.04031510051804865,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2011513835","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04243015,0.0013719792,0.9530389,0.0012568533,0.000093282964,0.000120537894,0.00020813578,0.0002439323,0.0012363107],"genre_scores_gemma":[0.84984106,0.0012630703,0.14313777,0.00081885874,0.00034979597,0.0007282997,0.00080330484,0.00016922233,0.0028885691],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.92414385,0.06304353,0.0025287124,0.0046076514,0.0044552754,0.0012209539],"domain_scores_gemma":[0.70853484,0.24273522,0.02365579,0.021650169,0.002739602,0.0006844101],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10141329,0.0010589614,0.0022165955,0.0019045097,0.0005341702,0.002532095,0.0030070464,0.001845888,0.001659508],"category_scores_gemma":[0.25998822,0.0008303232,0.0025799556,0.0030031907,0.0023097082,0.0026083207,0.0027083368,0.0038196978,0.00031678434],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011682468,0.00015551093,0.117552206,0.001013484,0.0022177107,0.0018734593,0.0016567375,0.297554,0.0015230747,0.4495855,0.004468486,0.121231616],"study_design_scores_gemma":[0.00011195834,0.00032968924,0.020844156,0.0001509829,0.00041547292,0.0007855463,0.00013599344,0.58060414,0.0011418617,0.39018494,0.0052012284,0.00009402767],"about_ca_topic_score_codex":0.0026086075,"about_ca_topic_score_gemma":0.0029882707,"teacher_disagreement_score":0.8985867,"about_ca_system_score_codex":0.0014154043,"about_ca_system_score_gemma":0.0022570556,"threshold_uncertainty_score":0.5363312},"labels":[],"label_agreement":null},{"id":"W2012127814","doi":"10.1002/1097-0258(20001030)19:20<2775::aid-sim545>3.0.co;2-g","title":"Numerical methods for the evaluation of individual bioequivalence criteria","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SNC-Lavalin (Canada)","funders":"","keywords":"Restricted maximum likelihood; Resampling; Fortran; Variance (accounting); Statistics; Mathematics; Bioequivalence; Computer science; Covariate; Mixed model; Maximum likelihood; Algorithm; Applied mathematics","score_opus":0.7410054186035245,"score_gpt":0.7031661636645462,"score_spread":0.03783925493897833,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2012127814","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000271554,0.00020898391,0.997692,0.0000806762,0.00004256465,0.00010218502,0.000049144066,0.00022541944,0.0013275272],"genre_scores_gemma":[0.011983343,0.0004433645,0.9848445,0.00009400641,0.000070119415,0.0013307563,0.00011434989,0.00019725636,0.0009223385],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98309547,0.011493291,0.0010032698,0.00068387506,0.003526601,0.00019746099],"domain_scores_gemma":[0.94914615,0.041099157,0.0026516537,0.003088151,0.0037378266,0.00027707167],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020084389,0.0017761608,0.0021195626,0.0040651634,0.0008304897,0.0021994559,0.002262203,0.0018832575,0.015205135],"category_scores_gemma":[0.093622446,0.0008303484,0.0014395851,0.0032313494,0.0022910163,0.0020967158,0.002671861,0.0037703055,0.004524401],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019355434,0.00015478485,0.0013874575,0.0012084423,0.00021267701,0.00022350023,0.000306286,0.1739789,0.0030000135,0.54327404,0.012757972,0.2633024],"study_design_scores_gemma":[0.00014514057,0.00018038145,0.0005843517,0.00036935828,0.000059876544,0.00023328204,0.00008361838,0.6493182,0.0016707538,0.31320992,0.03408086,0.000064185246],"about_ca_topic_score_codex":0.0015922323,"about_ca_topic_score_gemma":0.0013311121,"teacher_disagreement_score":0.020084389,"about_ca_system_score_codex":0.0015938954,"about_ca_system_score_gemma":0.0030746988,"threshold_uncertainty_score":0.10621768},"labels":[],"label_agreement":null},{"id":"W2013186997","doi":"10.1002/sim.1777","title":"Non‐parametric estimation of gap time survival functions for ordered multivariate failure time data","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":36,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Heart, Lung, and Blood Institute; McMaster University","keywords":"Censoring (clinical trials); Estimator; Identifiability; Parametric statistics; Multivariate statistics; Statistics; Mathematics; Nonparametric statistics; Bivariate analysis; Applied mathematics","score_opus":0.13900848002663804,"score_gpt":0.4338596289471855,"score_spread":0.2948511489205474,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2013186997","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.045670036,0.0002250554,0.95346653,0.00012196648,0.000010073931,0.000036888705,0.000104645405,0.00016745972,0.00019727627],"genre_scores_gemma":[0.77917737,0.0005697378,0.21803486,0.000069203714,0.000037023357,0.00033910095,0.0008041219,0.000087525244,0.0008809734],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9955343,0.0032972137,0.00019737818,0.00033570244,0.00047596407,0.00015945591],"domain_scores_gemma":[0.91433257,0.07472107,0.004981499,0.0043409565,0.0012399624,0.00038401474],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020009907,0.000652628,0.0012072824,0.0019020151,0.00029140507,0.0010151598,0.0015228695,0.000978728,0.001088583],"category_scores_gemma":[0.08099607,0.00038331756,0.00097179366,0.0016050036,0.0016120984,0.0016701722,0.0019611854,0.0020816806,0.0002544561],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00041230192,0.00018852892,0.03139441,0.00051594886,0.0002751078,0.00039645567,0.0011895224,0.5641921,0.0027970702,0.17524154,0.0012214361,0.22217552],"study_design_scores_gemma":[0.000018287208,0.000104695864,0.0064281197,0.00004473917,0.000025370226,0.00015869917,0.00014165891,0.9122981,0.0011702463,0.07885117,0.0007272027,0.000031664884],"about_ca_topic_score_codex":0.0013331071,"about_ca_topic_score_gemma":0.0010045193,"teacher_disagreement_score":0.020009907,"about_ca_system_score_codex":0.00062876573,"about_ca_system_score_gemma":0.00107172,"threshold_uncertainty_score":0.105823755},"labels":[],"label_agreement":null},{"id":"W2013245256","doi":"10.1002/sim.3764","title":"A weighted Cox model for modelling time‐dependent exposures in the analysis of case–control studies","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Université de Montréal","funders":"Canadian Institutes of Health Research; Health Canada; Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs; McGill University","keywords":"Proportional hazards model; Computer science; Econometrics; Statistics; Mathematics","score_opus":0.14315108448926323,"score_gpt":0.4545637737110838,"score_spread":0.3114126892218206,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2013245256","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0031441825,0.001964703,0.9917053,0.0006525217,0.0002975965,0.0006009729,0.0007412516,0.00022806705,0.00066541054],"genre_scores_gemma":[0.1970546,0.0089030685,0.7670521,0.0014695522,0.0012587919,0.011299488,0.0039310534,0.0003313955,0.008699853],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96220756,0.02975758,0.001597011,0.0033416832,0.0024983347,0.0005978423],"domain_scores_gemma":[0.9284231,0.061503924,0.0040192106,0.0034957228,0.0020897815,0.00046828666],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.060938936,0.0020766438,0.0027475397,0.004145216,0.0008270658,0.0028092435,0.005984206,0.0041600307,0.0064977836],"category_scores_gemma":[0.09774366,0.0014704318,0.0045630196,0.005550675,0.0017554222,0.0029954424,0.0020908725,0.005227666,0.0013659364],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000898121,0.00017543316,0.01599218,0.0018814572,0.002961656,0.0013414699,0.001132633,0.423824,0.0013292102,0.42813385,0.012463332,0.10986662],"study_design_scores_gemma":[0.0003506804,0.00045541333,0.001998965,0.00036947613,0.0007225769,0.00049829733,0.00010119567,0.7265714,0.0004919491,0.2510856,0.01722385,0.00013060957],"about_ca_topic_score_codex":0.009112414,"about_ca_topic_score_gemma":0.0059592547,"teacher_disagreement_score":0.060938936,"about_ca_system_score_codex":0.002029479,"about_ca_system_score_gemma":0.003967805,"threshold_uncertainty_score":0.32227975},"labels":[],"label_agreement":null},{"id":"W2013920814","doi":"10.1002/sim.4200","title":"Comparing paired vs non‐paired statistical methods of analyses when making inferences about absolute risk reductions in propensity‐score matched samples","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":325,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Statistics; Confidence interval; Confounding; Sample size determination; Selection bias; Observational study; Statistical significance; Statistical inference; Matching (statistics); Causal inference; Medicine; Mathematics","score_opus":0.6053372113795614,"score_gpt":0.5275261212026646,"score_spread":0.0778110901768968,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2013920814","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.021434557,0.0020797662,0.9641409,0.0013563719,0.0014501314,0.004995408,0.0005793584,0.000549677,0.003413794],"genre_scores_gemma":[0.23323654,0.0009323264,0.7443196,0.0019420466,0.0005003312,0.016895536,0.0005524331,0.0005322728,0.0010888737],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.33207357,0.6052327,0.01939249,0.01550575,0.026587034,0.0012083781],"domain_scores_gemma":[0.14402986,0.75664485,0.0317837,0.05181341,0.014913506,0.0008146727],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3921677,0.0019584787,0.0038270568,0.003772027,0.0015521995,0.005874407,0.004651726,0.0037557436,0.008487406],"category_scores_gemma":[0.7571114,0.0017094209,0.007897763,0.0045128763,0.0068211155,0.0068406416,0.0049218354,0.008159158,0.0011846179],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.013014706,0.0021766522,0.06633946,0.013094984,0.034942903,0.0011937494,0.010030798,0.044497408,0.0038848305,0.21430022,0.023765368,0.57275885],"study_design_scores_gemma":[0.006752149,0.016204834,0.07463906,0.00922901,0.014768853,0.0021743355,0.003535578,0.26513398,0.029514616,0.50069165,0.07604178,0.0013142325],"about_ca_topic_score_codex":0.0010846179,"about_ca_topic_score_gemma":0.0010714934,"teacher_disagreement_score":0.6078323,"about_ca_system_score_codex":0.002002545,"about_ca_system_score_gemma":0.0032877624,"threshold_uncertainty_score":0.7495655},"labels":[],"label_agreement":null},{"id":"W2014084548","doi":"10.1002/sim.2523","title":"Predicting case numbers during infectious disease outbreaks when some cases are undiagnosed","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"COVID-19 epidemiological studies","field":"Mathematics","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Outbreak; Infectious disease (medical specialty); Markov chain Monte Carlo; Basic reproduction number; Bayesian probability; Markov chain; Statistics; Disease; Computer science; Medicine; Mathematics; Virology; Environmental health; Pathology","score_opus":0.09159471013788364,"score_gpt":0.39795735505440083,"score_spread":0.3063626449165172,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2014084548","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3244093,0.0014740926,0.66333646,0.00027636878,0.00010063474,0.00026128965,0.0029969166,0.0014574867,0.0056875115],"genre_scores_gemma":[0.8064378,0.0005207001,0.18680853,0.000109250315,0.000072708004,0.00045457066,0.0043522473,0.00012573182,0.0011185334],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9959698,0.0019191263,0.0004361848,0.00075000915,0.0007833463,0.00014158065],"domain_scores_gemma":[0.9170394,0.06753895,0.007702346,0.0041991724,0.002838886,0.0006811842],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012573985,0.00072236505,0.0008335571,0.0039014604,0.00042513476,0.0011811244,0.0012955484,0.0010451085,0.0011469661],"category_scores_gemma":[0.077392116,0.0005352448,0.00063140376,0.0010737696,0.0005567348,0.0013431273,0.0010115731,0.0012437293,0.0006510653],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015440424,0.00024514852,0.42230964,0.0003937272,0.0004967121,0.00038086198,0.0014789497,0.31521347,0.004713172,0.012481367,0.0044150776,0.2363279],"study_design_scores_gemma":[0.00008343948,0.000545144,0.18358709,0.00035783445,0.00017754764,0.0010686729,0.00043932276,0.7632183,0.01302485,0.028639788,0.008559886,0.0002981806],"about_ca_topic_score_codex":0.0032612465,"about_ca_topic_score_gemma":0.0030160332,"teacher_disagreement_score":0.012573985,"about_ca_system_score_codex":0.00053411623,"about_ca_system_score_gemma":0.0005473139,"threshold_uncertainty_score":0.0664984},"labels":[],"label_agreement":null},{"id":"W2014423310","doi":"10.1002/sim.5569","title":"On the interchangeability of biologic drug products","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Biosimilars and Bioanalytical Methods","field":"Immunology and Microbiology","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canada Research Chairs; University of Toronto","funders":"","keywords":"Interchangeability; Bioequivalence; Drug; Risk analysis (engineering); Biosimilar; Tilmicosin; Medicine; Equivalence (formal languages); Pharmacology; Computer science; Mathematics; Chemistry; Pharmacokinetics","score_opus":0.06153093761929549,"score_gpt":0.3525918014453326,"score_spread":0.2910608638260371,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2014423310","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.045640133,0.036120728,0.30884376,0.08312403,0.008783654,0.0006877263,0.0017273484,0.0007284472,0.5143443],"genre_scores_gemma":[0.7206138,0.0420449,0.13785782,0.024996975,0.018364279,0.0012516914,0.0025866535,0.00082202564,0.05146195],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.922545,0.026195707,0.00790949,0.0065317233,0.033755016,0.0030630333],"domain_scores_gemma":[0.8882329,0.06991057,0.013400306,0.012892448,0.013530931,0.0020328981],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.033652592,0.00079205027,0.0014006023,0.0060486887,0.004306853,0.011428797,0.0024565181,0.0045778737,0.0104094045],"category_scores_gemma":[0.11580845,0.0005719818,0.0020726752,0.00898463,0.016362337,0.0132799605,0.006160782,0.0068489164,0.004985812],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009224422,0.000037718713,0.0013156402,0.000103116094,0.000011879772,0.00022521595,0.00031992028,0.00049015187,0.0003575131,0.9390369,0.008830769,0.049178857],"study_design_scores_gemma":[0.000029967327,0.0002002649,0.0026547883,0.0005946274,0.000037298498,0.0014010908,0.00048390514,0.0033295634,0.0029335709,0.76583475,0.22241913,0.00008093609],"about_ca_topic_score_codex":0.002230112,"about_ca_topic_score_gemma":0.0009839967,"teacher_disagreement_score":0.033652592,"about_ca_system_score_codex":0.007908342,"about_ca_system_score_gemma":0.006990503,"threshold_uncertainty_score":0.17797405},"labels":[],"label_agreement":null},{"id":"W2014828742","doi":"10.1002/sim.1720","title":"Two macroscopic and microscopic brain imaging studies of human hippocampus in early Alzheimer's disease and schizophrenia research","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Functional Brain Connectivity Studies","field":"Neuroscience","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Smiths Detection (Canada)","funders":"National Institute of Neurological Disorders and Stroke; National Institute of Mental Health; National Institutes of Health","keywords":"Neuroscience; Schizophrenia (object-oriented programming); Hippocampal formation; Hippocampus; Human brain; Functional magnetic resonance imaging; Disease; Brain Structure and Function; Psychology; Alzheimer's disease; Neuroimaging; Medicine; Psychiatry; Pathology","score_opus":0.08440019450687297,"score_gpt":0.42488422864518777,"score_spread":0.3404840341383148,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2014828742","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94524246,0.0066627227,0.03728668,0.0015772536,0.00008270632,0.0002556266,0.00066710485,0.00012131387,0.008104176],"genre_scores_gemma":[0.95840913,0.002679918,0.0357865,0.0002769656,0.000074179276,0.000107107044,0.00017921187,0.000021662427,0.0024653268],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.9997788,0.00007560902,0.00001305223,0.00004573672,0.000060524548,0.000026224243],"domain_scores_gemma":[0.9995757,0.00013466083,0.00010847723,0.000057886085,0.000057986977,0.00006526095],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00065433857,0.00025101847,0.00016529462,0.0010181803,0.0003350903,0.00027392243,0.00024689978,0.0004168279,0.0007734137],"category_scores_gemma":[0.0011618742,0.00028876553,0.00017904963,0.00031265398,0.00084620644,0.0003450969,0.00053023145,0.00033810173,0.000108431144],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002020389,0.00048164898,0.07594098,0.0007412733,0.00022047292,0.0033384704,0.0013583779,0.005352421,0.7875155,0.013357968,0.0015470958,0.10812548],"study_design_scores_gemma":[0.00022369313,0.0020033766,0.8648437,0.000099792866,0.0002233231,0.013075179,0.00091602025,0.0063127265,0.09275005,0.0123339845,0.007078399,0.00013974389],"about_ca_topic_score_codex":0.0060917214,"about_ca_topic_score_gemma":0.015299501,"teacher_disagreement_score":0.0060917214,"about_ca_system_score_codex":0.00046714433,"about_ca_system_score_gemma":0.0005033397,"threshold_uncertainty_score":0.012112498},"labels":[],"label_agreement":null},{"id":"W2016649643","doi":"10.1002/sim.5581","title":"Regression with incomplete covariates and left‐truncated time‐to‐event data","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Canadian Institutes of Health Research","keywords":"Covariate; Statistics; Truncation (statistics); Proportional hazards model; Econometrics; Censoring (clinical trials); Medicine; Survival analysis; Event (particle physics); Clinical trial; Mathematics; Internal medicine","score_opus":0.12745610875890212,"score_gpt":0.4359655878327633,"score_spread":0.30850947907386117,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2016649643","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.006997878,0.0005487689,0.99028045,0.00062753074,0.00007256358,0.00011143633,0.0008444969,0.00025269852,0.00026415207],"genre_scores_gemma":[0.26279536,0.0025042314,0.7188668,0.0011477191,0.000778515,0.002184593,0.0063425577,0.0004440611,0.004936043],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9647491,0.026612783,0.002042091,0.0038066292,0.0019085358,0.00088087184],"domain_scores_gemma":[0.8203188,0.14969352,0.009844547,0.016831942,0.0025742706,0.00073688524],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.076610915,0.002040067,0.0047743996,0.0021509319,0.000856138,0.0020061505,0.00628114,0.003002046,0.004907899],"category_scores_gemma":[0.18388082,0.0019966708,0.0040633455,0.0038069799,0.0029257636,0.005513338,0.003241639,0.005118508,0.0013656209],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019436282,0.00034582315,0.032545224,0.0017625353,0.0023271437,0.0015273486,0.0007983856,0.37732598,0.0022867622,0.38586268,0.009576741,0.18369776],"study_design_scores_gemma":[0.00030184933,0.00023545654,0.0058621513,0.00017672442,0.0002545948,0.00023916078,0.000055371296,0.6254927,0.0011369133,0.36055803,0.0056056445,0.00008149111],"about_ca_topic_score_codex":0.006072426,"about_ca_topic_score_gemma":0.004502804,"teacher_disagreement_score":0.076610915,"about_ca_system_score_codex":0.0014579063,"about_ca_system_score_gemma":0.0028995257,"threshold_uncertainty_score":0.4051621},"labels":[],"label_agreement":null},{"id":"W2016919144","doi":"10.1002/sim.2711","title":"Bayesian sensitivity analysis for unmeasured confounding in observational studies","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":184,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; University of British Columbia Hospital","funders":"","keywords":"Confounding; Statistics; Markov chain Monte Carlo; Bayesian probability; Observational study; Econometrics; Logistic regression; Prior probability; Sample size determination; Sensitivity (control systems); Mathematics; Computer science","score_opus":0.2203225194206481,"score_gpt":0.47620137534415125,"score_spread":0.2558788559235031,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2016919144","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017517373,0.0023400553,0.97508085,0.002014596,0.0001175295,0.00042277988,0.00022816389,0.00010730737,0.0021713565],"genre_scores_gemma":[0.739243,0.0029514662,0.25300166,0.0013037664,0.00029625627,0.001983278,0.00026676024,0.00009065766,0.0008630661],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.83659905,0.15007588,0.0025557834,0.0034951828,0.006378083,0.00089601474],"domain_scores_gemma":[0.39808345,0.5747087,0.011629346,0.010942355,0.0040790658,0.0005570442],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.16287005,0.0013851692,0.0032669397,0.0040063187,0.0010066346,0.0031571668,0.0030281025,0.0035227756,0.0028960784],"category_scores_gemma":[0.48156366,0.00096366165,0.0036800893,0.0030213213,0.003157869,0.0033721623,0.0036115563,0.0036136105,0.00017849226],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00067086006,0.00009664701,0.010567918,0.0015351665,0.0034358467,0.00091787387,0.00076006877,0.5666932,0.0006471625,0.35674554,0.002211173,0.055718504],"study_design_scores_gemma":[0.00018851893,0.00017490724,0.002877858,0.00048777644,0.0008702548,0.00036965837,0.00011552694,0.5058632,0.0005926207,0.48534966,0.003016271,0.000093805655],"about_ca_topic_score_codex":0.0039459257,"about_ca_topic_score_gemma":0.001653432,"teacher_disagreement_score":0.16287005,"about_ca_system_score_codex":0.0030125924,"about_ca_system_score_gemma":0.002308714,"threshold_uncertainty_score":0.86134946},"labels":[],"label_agreement":null},{"id":"W2017135200","doi":"10.1002/sim.5637","title":"Empirical likelihood‐based confidence intervals for length‐biased data","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University","funders":"National Cancer Institute","keywords":"Statistics; Confidence interval; Event (particle physics); Poisson distribution; Econometrics; Population; Sampling bias; Sample size determination; Survival analysis; Sample (material); Margin (machine learning); Simple random sample; Mathematics; Computer science; Demography","score_opus":0.38984294806424596,"score_gpt":0.5328989883676382,"score_spread":0.1430560403033922,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2017135200","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.006638828,0.002444094,0.98832405,0.00045646034,0.00012475677,0.00014170917,0.00026690573,0.0003856033,0.0012175549],"genre_scores_gemma":[0.2906763,0.0030798116,0.70009315,0.00078788935,0.00070689427,0.0017058206,0.0015247017,0.00042182335,0.0010036401],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9005031,0.078473166,0.0042367275,0.005620492,0.010320084,0.0008464417],"domain_scores_gemma":[0.28037953,0.6582108,0.020441867,0.029718874,0.010197804,0.0010510993],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.13146734,0.0017432269,0.0035099268,0.007865818,0.0010608031,0.0042250846,0.006694408,0.0036951397,0.0042631123],"category_scores_gemma":[0.6344421,0.001167008,0.002546036,0.006975049,0.004792021,0.0065258527,0.0049247695,0.00732867,0.0009263434],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007430101,0.00018799938,0.024991425,0.0019556049,0.0017592917,0.0011341679,0.0017633989,0.16283748,0.0010840135,0.5447697,0.008863321,0.24991058],"study_design_scores_gemma":[0.00020149194,0.00025874862,0.0064072516,0.0011499715,0.00033831064,0.00087583216,0.00030680065,0.50238806,0.0016077878,0.47732994,0.00897161,0.0001641837],"about_ca_topic_score_codex":0.0021167647,"about_ca_topic_score_gemma":0.0011357171,"teacher_disagreement_score":0.13146734,"about_ca_system_score_codex":0.0022340287,"about_ca_system_score_gemma":0.0016585167,"threshold_uncertainty_score":0.6952741},"labels":[],"label_agreement":null},{"id":"W2017258397","doi":"10.1002/sim.2895","title":"Tests of association under misclassification: Application to histological sampling in oncology","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"AI in cancer detection","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"National Cancer Institute","keywords":"Feature (linguistics); Medicine; Oligodendroglioma; Outcome (game theory); Association (psychology); Sampling (signal processing); Radiology; Pathology; Computer science; Psychology; Glioma; Mathematics; Astrocytoma","score_opus":0.0552569513599816,"score_gpt":0.3907814980580758,"score_spread":0.33552454669809423,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2017258397","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2453721,0.008721507,0.7292154,0.006581575,0.001451573,0.0017588758,0.0011918243,0.0013778302,0.0043294244],"genre_scores_gemma":[0.81874496,0.00084909436,0.17494254,0.001372997,0.0009680234,0.0014665175,0.000413931,0.0001945731,0.0010473239],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.36349815,0.57638997,0.012937149,0.024697777,0.02081175,0.0016652615],"domain_scores_gemma":[0.056653827,0.8997435,0.018686103,0.020733146,0.0030617334,0.0011217169],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.46104756,0.003410905,0.00676486,0.009317813,0.0036117362,0.0055319634,0.009282649,0.0075045554,0.0037097547],"category_scores_gemma":[0.755706,0.0016243234,0.007331435,0.012198964,0.02503944,0.00878191,0.0075200377,0.007899807,0.00041035237],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0060323854,0.00036564234,0.66025454,0.0024316844,0.025364768,0.0034459492,0.005495156,0.044727635,0.00082099764,0.06728221,0.004500561,0.17927839],"study_design_scores_gemma":[0.0012770259,0.0058843945,0.21009056,0.0010071769,0.00657587,0.005342024,0.0028591596,0.48937866,0.0021726722,0.26532632,0.009555196,0.00053106016],"about_ca_topic_score_codex":0.0060124686,"about_ca_topic_score_gemma":0.002502343,"teacher_disagreement_score":0.46104756,"about_ca_system_score_codex":0.0032200958,"about_ca_system_score_gemma":0.0044028624,"threshold_uncertainty_score":0.66462433},"labels":[],"label_agreement":null},{"id":"W2017498225","doi":"10.1002/sim.5791","title":"A comparison of Bayesian hierarchical modeling with group‐based exposure assessment in occupational epidemiology","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Occupational and environmental lung diseases","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Institute of Environmental Health Sciences","keywords":"Bayesian probability; Logistic regression; Computer science; Epidemiology; Bayesian hierarchical modeling; Hierarchical database model; Cohort; Statistics; Bayesian inference; Data mining; Econometrics; Medicine; Machine learning; Artificial intelligence; Mathematics; Pathology","score_opus":0.052482689336623825,"score_gpt":0.40254146372720645,"score_spread":0.35005877439058264,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2017498225","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.045765445,0.0012491919,0.9481802,0.0012652718,0.00012324026,0.00038180838,0.0003026246,0.00029333876,0.0024388626],"genre_scores_gemma":[0.332188,0.0012963773,0.6627196,0.00066047366,0.00016105105,0.0009016187,0.00060108857,0.00020524855,0.0012665726],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.96055394,0.034984343,0.0005832777,0.0010827052,0.002517808,0.00027795922],"domain_scores_gemma":[0.8497391,0.13700481,0.0030943106,0.0058130226,0.0035248438,0.00082394836],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.051077608,0.0009421823,0.0015783281,0.0020761904,0.00077463436,0.0018942425,0.0028569063,0.0016385457,0.002552184],"category_scores_gemma":[0.120814435,0.0007453465,0.0020458058,0.0019183652,0.0012607691,0.002902483,0.0023998236,0.0016596603,0.00037149308],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017519187,0.00044848563,0.01868624,0.0007820076,0.00185359,0.00014687028,0.0014541392,0.55623966,0.00077911705,0.16611804,0.0040246835,0.24771522],"study_design_scores_gemma":[0.00023860697,0.0003655998,0.0042458177,0.00014803199,0.00021593133,0.0000725178,0.00013569767,0.8923545,0.00020672135,0.09992534,0.0019941484,0.00009702418],"about_ca_topic_score_codex":0.018790456,"about_ca_topic_score_gemma":0.016205745,"teacher_disagreement_score":0.9489224,"about_ca_system_score_codex":0.002012152,"about_ca_system_score_gemma":0.003719024,"threshold_uncertainty_score":0.27012742},"labels":[],"label_agreement":null},{"id":"W2020884741","doi":"10.1002/sim.4193","title":"Regression models, scan statistics and reappearance probabilities to detect regions of association between gene expression and copy number","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Gene expression and cancer classification","field":"Biochemistry, Genetics and Molecular Biology","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Lunenfeld-Tanenbaum Research Institute; University of Toronto; Mount Sinai Hospital","funders":"Canadian Institutes of Health Research; University Health Network","keywords":"Covariate; Statistic; Statistics; Breast cancer; Regression; Chromosome; Regression analysis; Biology; Mathematics; Cancer; Genetics; Gene","score_opus":0.030693225007564433,"score_gpt":0.3028105837693933,"score_spread":0.27211735876182885,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2020884741","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.026311826,0.00078217254,0.9705639,0.00037504378,0.000054858137,0.00012629031,0.00046473657,0.0007619266,0.00055919925],"genre_scores_gemma":[0.5509405,0.0019299855,0.43654954,0.0004222505,0.0002990643,0.0016806509,0.0018541754,0.0007768701,0.005546904],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9869818,0.009373504,0.00036011575,0.0019724579,0.00088384293,0.00042830885],"domain_scores_gemma":[0.8856323,0.10191483,0.005826878,0.003991065,0.0021023853,0.0005324963],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.027932432,0.0018300471,0.0023841655,0.0034296168,0.00061202585,0.0020276546,0.0044300496,0.0028423257,0.0034863453],"category_scores_gemma":[0.09791305,0.0012353907,0.0027991282,0.0037233888,0.003308461,0.002987258,0.0018719095,0.0043423222,0.00091752503],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006371894,0.0001424434,0.037365913,0.00034427366,0.0008432751,0.00047206078,0.00045943007,0.73925537,0.0025137458,0.13985734,0.0026526088,0.07545639],"study_design_scores_gemma":[0.00006176663,0.00015356019,0.003751563,0.00003954659,0.00009423785,0.00018756273,0.0000535607,0.92387646,0.00087498635,0.06893545,0.0019128617,0.000058473222],"about_ca_topic_score_codex":0.008410694,"about_ca_topic_score_gemma":0.004833665,"teacher_disagreement_score":0.027932432,"about_ca_system_score_codex":0.0015525263,"about_ca_system_score_gemma":0.0016035571,"threshold_uncertainty_score":0.1477226},"labels":[],"label_agreement":null},{"id":"W2021071948","doi":"10.1002/sim.1473","title":"Bias correction of two‐state latent Markov process parameter estimates under misclassification","year":2003,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; University of Alberta","funders":"","keywords":"Estimator; Unobservable; Statistics; Mathematics; Computer science; Maximum likelihood; Econometrics","score_opus":0.16662811501022856,"score_gpt":0.4631804032302823,"score_spread":0.29655228822005375,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2021071948","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.122863606,0.0004542795,0.8751464,0.0004647532,0.00007422808,0.000059304984,0.000060009093,0.0002869921,0.000590439],"genre_scores_gemma":[0.87697446,0.00030375665,0.12125848,0.00015106065,0.00005289565,0.00011152183,0.00015390644,0.00006949304,0.0009244238],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99213696,0.0052702366,0.0003974933,0.0008789824,0.0009731025,0.00034317866],"domain_scores_gemma":[0.8555357,0.12501714,0.0074786707,0.0077767577,0.0037220244,0.00046969848],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03009889,0.00057660596,0.0011122767,0.0011932061,0.00051124574,0.0014529147,0.0014650326,0.0018447768,0.00079290476],"category_scores_gemma":[0.19347861,0.0005134333,0.00088239025,0.0012272642,0.0015398844,0.0021188557,0.0016413437,0.0018599853,0.00017999126],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006942547,0.00013901285,0.06548295,0.00036588387,0.000627463,0.00054429367,0.0016449683,0.5626412,0.004683487,0.13220681,0.0012526904,0.22971706],"study_design_scores_gemma":[0.000064139334,0.00009959089,0.01323554,0.0000824245,0.00011537804,0.0002138302,0.00008043348,0.90884,0.0034503944,0.07299344,0.00075063464,0.000074076466],"about_ca_topic_score_codex":0.0036581967,"about_ca_topic_score_gemma":0.002148789,"teacher_disagreement_score":0.03009889,"about_ca_system_score_codex":0.0013089485,"about_ca_system_score_gemma":0.0014671646,"threshold_uncertainty_score":0.15918005},"labels":[],"label_agreement":null},{"id":"W2021281962","doi":"10.1002/sim.6455","title":"Estimating survival probabilities by exposure levels: utilizing vital statistics and complex survey data with mortality follow‐up","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health disparities and outcomes","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto","funders":"","keywords":"Statistics; Estimator; Jackknife resampling; Categorical variable; Survey sampling; Variance (accounting); Econometrics; Hazard ratio; Hazard; Survey data collection; Mathematics; Computer science; Medicine; Population; Confidence interval; Economics","score_opus":0.3310131340846792,"score_gpt":0.46000764865395993,"score_spread":0.12899451456928074,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2021281962","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.028092442,0.00013587155,0.9704361,0.00016515228,0.000022352213,0.00024384243,0.00043733593,0.00012130554,0.00034568264],"genre_scores_gemma":[0.27658704,0.00043102048,0.7187322,0.00019264006,0.000072624505,0.0012635482,0.0013377826,0.00006575535,0.0013173388],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9929603,0.0046606148,0.00046859295,0.0008556215,0.0008589654,0.00019583588],"domain_scores_gemma":[0.97102106,0.021463769,0.0031016727,0.0032540702,0.0009463781,0.00021312188],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.015080219,0.00079573784,0.0011775218,0.0033434725,0.00039754517,0.0012894548,0.0016173626,0.0010027575,0.0024599382],"category_scores_gemma":[0.06262475,0.0007242903,0.0020392209,0.002951653,0.00081664824,0.0019800465,0.002598889,0.001395947,0.00039770585],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003661665,0.0005589403,0.3013016,0.0008001473,0.0019440196,0.00046308068,0.0018539124,0.14248273,0.0032544003,0.10293409,0.0037822768,0.44025868],"study_design_scores_gemma":[0.00015323918,0.00059597153,0.11520527,0.00034166974,0.0005124537,0.00059534237,0.0005414418,0.67683756,0.003743679,0.1915489,0.009696308,0.000228194],"about_ca_topic_score_codex":0.005207858,"about_ca_topic_score_gemma":0.0064201527,"teacher_disagreement_score":0.9849198,"about_ca_system_score_codex":0.00073206756,"about_ca_system_score_gemma":0.0019531474,"threshold_uncertainty_score":0.07975274},"labels":[],"label_agreement":null},{"id":"W2023282380","doi":"10.1002/sim.2968","title":"A comparison of bootstrap methods and an adjusted bootstrap approach for estimating the prediction error in microarray classification","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Gene expression and cancer classification","field":"Biochemistry, Genetics and Molecular Biology","cited_by":88,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Concordia University","funders":"","keywords":"Bootstrap aggregating; Computer science; Statistics; Sample size determination; Set (abstract data type); Sample (material); Data mining; Mathematics","score_opus":0.12683376374931365,"score_gpt":0.46886889527667613,"score_spread":0.3420351315273625,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2023282380","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0064795236,0.0028590786,0.9889452,0.00025334014,0.00018170063,0.00008677314,0.0000912752,0.0004202842,0.0006828417],"genre_scores_gemma":[0.15185863,0.0041710325,0.8393706,0.00043143635,0.0006371806,0.00079646095,0.00079222216,0.0005285542,0.0014138344],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.97353417,0.019365776,0.00088704744,0.0012251146,0.004660806,0.00032705104],"domain_scores_gemma":[0.94619966,0.042295855,0.0016933986,0.0041542286,0.0052952305,0.0003616327],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026293902,0.0016272981,0.0019955053,0.003609393,0.0007128657,0.0013530956,0.0028616034,0.0022291988,0.0015282393],"category_scores_gemma":[0.08179779,0.00064957625,0.0018195325,0.004002129,0.0013413363,0.0025146264,0.002180712,0.0030894922,0.0012016859],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009993843,0.00020022847,0.009193297,0.0010175396,0.0011873496,0.00023930981,0.00049548753,0.10218645,0.0042718058,0.05155138,0.0075779883,0.82107973],"study_design_scores_gemma":[0.00015759114,0.00053773774,0.009491789,0.00033821218,0.00026900967,0.00048987265,0.00020010938,0.8893548,0.0051725423,0.07998742,0.013784477,0.0002164495],"about_ca_topic_score_codex":0.0016987969,"about_ca_topic_score_gemma":0.0014566574,"teacher_disagreement_score":0.026293902,"about_ca_system_score_codex":0.00084924,"about_ca_system_score_gemma":0.0012744009,"threshold_uncertainty_score":0.1390571},"labels":[],"label_agreement":null},{"id":"W2023939511","doi":"10.1002/sim.2521","title":"Some design issues of strata‐matched non‐randomized studies with survival outcomes","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Barrie Urology Group","funders":"National Cancer Institute; National Institutes of Natural Sciences; Natural Sciences and Engineering Research Council of Canada","keywords":"Randomized controlled trial; Sample size determination; Statistics; Covariate; Mathematics; Test (biology); Completely randomized design; Log-rank test; Research design; Intervention (counseling); Medicine; Survival analysis; Surgery","score_opus":0.13716863590134443,"score_gpt":0.45099642339759155,"score_spread":0.3138277874962471,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2023939511","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0027712488,0.0017508869,0.98460174,0.005854328,0.00056586764,0.002507244,0.00013409839,0.00010327799,0.001711411],"genre_scores_gemma":[0.05628722,0.001493054,0.91959536,0.0050096703,0.00084310496,0.015167737,0.00008487671,0.0000838546,0.0014351591],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.60124373,0.36375278,0.011773935,0.007032429,0.015025917,0.0011712334],"domain_scores_gemma":[0.45097238,0.49464753,0.017434146,0.027807383,0.008401924,0.0007366459],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3511985,0.0017654351,0.003460234,0.0027021489,0.0019455964,0.0034691226,0.0047268164,0.006035193,0.0054144803],"category_scores_gemma":[0.4928198,0.002124774,0.0059007416,0.0042664716,0.0072781993,0.004292726,0.0036045483,0.007058556,0.0007978479],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010332424,0.00017340614,0.003225911,0.0022610873,0.0012464862,0.00039811342,0.0011848566,0.010322107,0.000822741,0.8477231,0.0057912185,0.12581758],"study_design_scores_gemma":[0.0021093083,0.0015959034,0.0023132875,0.0007585838,0.00085264817,0.0006212469,0.00018667566,0.040154096,0.002745265,0.9104331,0.038091265,0.00013859506],"about_ca_topic_score_codex":0.0010522677,"about_ca_topic_score_gemma":0.0011346801,"teacher_disagreement_score":0.6488015,"about_ca_system_score_codex":0.0031676965,"about_ca_system_score_gemma":0.0043490417,"threshold_uncertainty_score":0.8000878},"labels":[],"label_agreement":null},{"id":"W2025797841","doi":"10.1002/sim.2178","title":"Estimation of age‐specific sensitivity and sojourn time in breast cancer screening studies","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Global Cancer Incidence and Screening","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"National Cancer Institute; National Institutes of Health","keywords":"Estimator; Breast cancer; Sensitivity (control systems); Statistics; Selection (genetic algorithm); Model selection; Computer science; Econometrics; Breast cancer screening; Cancer; Mathematics; Medicine; Machine learning; Mammography; Internal medicine","score_opus":0.07831978169133007,"score_gpt":0.40168126907008117,"score_spread":0.3233614873787511,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2025797841","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.64189315,0.00809444,0.34593484,0.0009729688,0.000092193484,0.00034235095,0.00034376624,0.00025350926,0.0020727685],"genre_scores_gemma":[0.97910744,0.0009935268,0.019166613,0.00015002994,0.000054854154,0.000106408566,0.00013641166,0.000014744491,0.00026982708],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9530862,0.041968487,0.0014096684,0.0015921469,0.0015575864,0.00038599802],"domain_scores_gemma":[0.54686624,0.41389528,0.02261679,0.012981006,0.002822075,0.0008185913],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08182726,0.00078244123,0.002307819,0.0047759977,0.00035592233,0.0015444071,0.0019351118,0.0021668463,0.00078511913],"category_scores_gemma":[0.2762587,0.0008842587,0.0028665385,0.0025448254,0.0014950986,0.00251895,0.0018216139,0.0013582316,0.00012872388],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017342786,0.00023938251,0.48032296,0.0014135304,0.009505619,0.00083799683,0.0018439587,0.3415772,0.0031695946,0.0482464,0.0007335244,0.11037557],"study_design_scores_gemma":[0.00020319126,0.0013505961,0.24498412,0.00046159932,0.0028792787,0.0015117797,0.00038692128,0.6408838,0.0050856005,0.09896499,0.0030110492,0.0002770893],"about_ca_topic_score_codex":0.0016520134,"about_ca_topic_score_gemma":0.0010194184,"teacher_disagreement_score":0.08182726,"about_ca_system_score_codex":0.0011975813,"about_ca_system_score_gemma":0.0009624587,"threshold_uncertainty_score":0.4327491},"labels":[],"label_agreement":null},{"id":"W2027702724","doi":"10.1002/sim.2392","title":"Comparison of Cox's and relative survival models when estimating the effects of prognostic factors on disease‐specific mortality: a simulation study under proportional excess hazards","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Colorectal Cancer Screening and Detection","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Montreal General Hospital","funders":"","keywords":"Proportional hazards model; Relative survival; Medicine; Population; Survival analysis; Disease; Demography; Statistics; Mortality rate; Regression analysis; Relative risk; Cause of death; Oncology; Cancer registry; Internal medicine; Mathematics; Confidence interval; Environmental health","score_opus":0.08855982869920223,"score_gpt":0.40453573870942483,"score_spread":0.3159759100102226,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2027702724","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6152849,0.0042849598,0.3725568,0.0018724636,0.0002838698,0.0007354313,0.0010091844,0.00031670847,0.0036557508],"genre_scores_gemma":[0.9267677,0.0014088572,0.06866773,0.00026236696,0.00007596232,0.000929387,0.0005807716,0.00007587594,0.0012312788],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9832184,0.01473716,0.0003336476,0.0006163884,0.00063180947,0.00046263906],"domain_scores_gemma":[0.780106,0.21056886,0.003415387,0.002701587,0.0023839537,0.00082423765],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04598545,0.0015722497,0.0023874852,0.0019035076,0.0005838082,0.0014842638,0.0030605218,0.0022208188,0.002199731],"category_scores_gemma":[0.07208928,0.00090142916,0.00339221,0.0015326192,0.0015175041,0.0022246353,0.0018015219,0.0030082888,0.00018352277],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011968652,0.00019818415,0.006951951,0.00019942826,0.00039424864,0.00010177749,0.00019585749,0.969492,0.00016469245,0.0139915515,0.00044148616,0.006672039],"study_design_scores_gemma":[0.00023670678,0.00038531347,0.00077528274,0.000048641643,0.00011166105,0.000048108937,0.000071921706,0.99217236,0.00018841484,0.0056545855,0.00027448047,0.000032503896],"about_ca_topic_score_codex":0.0149741685,"about_ca_topic_score_gemma":0.0059476094,"teacher_disagreement_score":0.04598545,"about_ca_system_score_codex":0.002006351,"about_ca_system_score_gemma":0.0022010868,"threshold_uncertainty_score":0.2431972},"labels":[],"label_agreement":null},{"id":"W2029359551","doi":"10.1002/sim.781","title":"Testing for the presence of cured patients: a simulation study","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta; Memorial University of Newfoundland","funders":"Natural Sciences and Engineering Research Council of Canada; National Research Council Canada; Alberta Heritage Foundation for Medical Research","keywords":"Censoring (clinical trials); Weibull distribution; Statistics; Likelihood-ratio test; Mathematics; Hazard ratio; Score test; Null distribution; Null hypothesis; Null (SQL); Maximum likelihood; Asymptotic distribution; Sample size determination; Applied mathematics; Statistical hypothesis testing; Econometrics; Test statistic; Confidence interval; Computer science","score_opus":0.058367907964955994,"score_gpt":0.38105411797635275,"score_spread":0.32268621001139675,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2029359551","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94800293,0.0013621264,0.041572887,0.0016648705,0.00011260302,0.0008357228,0.0018206097,0.0001901609,0.004438033],"genre_scores_gemma":[0.9717156,0.0005725871,0.024071932,0.00027504654,0.000048112117,0.00076866854,0.0013933671,0.00002646125,0.0011282115],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9921705,0.0065303296,0.00025437278,0.00036600084,0.00034042378,0.00033832973],"domain_scores_gemma":[0.74123955,0.2416334,0.005081133,0.005586414,0.004540259,0.0019192635],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0231657,0.001061666,0.0018826508,0.0016658558,0.0008813277,0.0011957946,0.0017570549,0.002431962,0.004273598],"category_scores_gemma":[0.053494338,0.00052647403,0.002112601,0.0023216743,0.0012924173,0.001150151,0.0016887175,0.0026675786,0.00035247454],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.007062256,0.0036390056,0.07057412,0.00060943764,0.001138257,0.0017129707,0.00071091385,0.86096144,0.001204138,0.024049804,0.006115817,0.022221914],"study_design_scores_gemma":[0.0022536416,0.0020262715,0.007616577,0.00010537357,0.00041162083,0.00043049545,0.00036730093,0.97655505,0.00077985966,0.008230243,0.0011453929,0.00007819721],"about_ca_topic_score_codex":0.010255845,"about_ca_topic_score_gemma":0.009324519,"teacher_disagreement_score":0.0231657,"about_ca_system_score_codex":0.0016200244,"about_ca_system_score_gemma":0.0016113344,"threshold_uncertainty_score":0.12251341},"labels":[],"label_agreement":null},{"id":"W2030157325","doi":"10.1002/sim.4018","title":"A Bayesian approach to simultaneously adjusting for verification and reference standard bias in diagnostic test studies","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":43,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"","keywords":"Covariate; Statistics; Bayesian probability; Test (biology); Independence (probability theory); Sensitivity (control systems); Conditional independence; Selection bias; Computer science; Econometrics; Mathematics","score_opus":0.5507579548728604,"score_gpt":0.5716683765821122,"score_spread":0.020910421709251836,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2030157325","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.003728922,0.0008943725,0.99271214,0.0015552443,0.00007937546,0.00020756714,0.00010008758,0.00012937136,0.0005929131],"genre_scores_gemma":[0.16427875,0.0016544852,0.8293424,0.0010623652,0.000449835,0.001726692,0.00021560217,0.00010003214,0.0011698361],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8474282,0.13418774,0.0039047024,0.0054643615,0.007873346,0.0011416661],"domain_scores_gemma":[0.7273012,0.24370003,0.011894028,0.00983527,0.006103291,0.0011662592],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1376136,0.002139909,0.004396789,0.00469079,0.0018967035,0.0041286144,0.0073574805,0.00607343,0.0025411353],"category_scores_gemma":[0.31561342,0.0026993065,0.004602264,0.0050692055,0.003940253,0.005210008,0.0038468374,0.0057285135,0.00047269827],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006190947,0.00024098987,0.012463697,0.0009580217,0.0039395597,0.000893921,0.0016024677,0.27356806,0.00086543156,0.49669087,0.004281575,0.20387639],"study_design_scores_gemma":[0.00042882928,0.0002683023,0.00293988,0.00034643995,0.001077997,0.00046501102,0.00008659852,0.41438416,0.00054334715,0.5732767,0.005998166,0.0001846577],"about_ca_topic_score_codex":0.013140976,"about_ca_topic_score_gemma":0.014579666,"teacher_disagreement_score":0.8623864,"about_ca_system_score_codex":0.004721855,"about_ca_system_score_gemma":0.007750618,"threshold_uncertainty_score":0.7277789},"labels":[],"label_agreement":null},{"id":"W2030381164","doi":"10.1002/sim.1106","title":"Prediction trees with soft nodes for binary outcomes","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical and Computational Modeling","field":"Computer Science","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Montreal Heart Institute; McGill University","funders":"","keywords":"Computer science; Event (particle physics); Novelty; Tree (set theory); Data mining; Data set; Set (abstract data type); Node (physics); Algorithm; Statistics; Artificial intelligence; Mathematics","score_opus":0.04925586001394029,"score_gpt":0.2959766245404736,"score_spread":0.2467207645265333,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2030381164","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.023714269,0.0005619621,0.9734986,0.0005538594,0.00008560105,0.00006979675,0.0003323394,0.00043017368,0.00075341034],"genre_scores_gemma":[0.38569126,0.0009429965,0.6078492,0.00032008602,0.0003166225,0.00046807472,0.0016269174,0.00012442694,0.0026604491],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9986916,0.0007135376,0.000075563104,0.00022933907,0.00020047653,0.000089466215],"domain_scores_gemma":[0.98675877,0.01142553,0.0005033148,0.0004518059,0.00068819977,0.00017226403],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0042867893,0.0008161709,0.0012668492,0.0013056078,0.0007652806,0.0010333759,0.0016887554,0.0016484705,0.0025197915],"category_scores_gemma":[0.020292614,0.00054427114,0.0007877305,0.001888902,0.0008364138,0.002624808,0.0012845705,0.002179124,0.0011467744],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002870279,0.00009726813,0.0042235036,0.00028248943,0.00010183498,0.00035367295,0.00049471017,0.69386387,0.00131815,0.047100402,0.0069996435,0.24487741],"study_design_scores_gemma":[0.000021135771,0.00003701069,0.00029482305,0.000021997435,0.000014495887,0.00004058058,0.000019073465,0.93321073,0.00031760114,0.065059505,0.0009515829,0.0000113519945],"about_ca_topic_score_codex":0.0015175836,"about_ca_topic_score_gemma":0.0022970233,"teacher_disagreement_score":0.0042867893,"about_ca_system_score_codex":0.0004818929,"about_ca_system_score_gemma":0.00056833477,"threshold_uncertainty_score":0.022671044},"labels":[],"label_agreement":null},{"id":"W2032526975","doi":"10.1002/sim.2002","title":"The utility of prior information and stratification for parameter estimation with two screening tests but no gold standard","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Gold standard (test); Prior information; Estimation; Statistics; Computer science; Stratification (seeds); Econometrics; Risk stratification; Mathematics; Medicine; Artificial intelligence; Internal medicine; Biology; Economics","score_opus":0.05101232799794388,"score_gpt":0.398043093680317,"score_spread":0.3470307656823731,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2032526975","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0075608105,0.00071917864,0.98929906,0.0011914056,0.000056064888,0.00006218952,0.000052465355,0.000085494125,0.0009732463],"genre_scores_gemma":[0.31666577,0.0026232211,0.67675936,0.00074675854,0.00049602205,0.00050819037,0.00035206607,0.0001196855,0.0017289704],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9688118,0.025325635,0.0008956886,0.002096916,0.002426711,0.00044334136],"domain_scores_gemma":[0.7422467,0.2391447,0.0053213504,0.010090842,0.002320399,0.000876056],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07040168,0.002175902,0.0028556662,0.004577096,0.0012728422,0.0034601828,0.0028092847,0.0036360763,0.0019243886],"category_scores_gemma":[0.30150953,0.0014588375,0.0025527258,0.0026516884,0.008423052,0.0073229107,0.004956421,0.00456831,0.0004660639],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003208589,0.00008880858,0.0072235344,0.0003373804,0.00034939856,0.00031079573,0.00066334713,0.17804489,0.00055012584,0.72256744,0.0012794773,0.08826391],"study_design_scores_gemma":[0.00004913578,0.000098202465,0.0014831362,0.00012867873,0.00006773098,0.0001038983,0.000040011808,0.40034813,0.00038452697,0.59613055,0.0011056938,0.000060253766],"about_ca_topic_score_codex":0.004822091,"about_ca_topic_score_gemma":0.003243171,"teacher_disagreement_score":0.07040168,"about_ca_system_score_codex":0.0024366502,"about_ca_system_score_gemma":0.0026865222,"threshold_uncertainty_score":0.3723241},"labels":[],"label_agreement":null},{"id":"W2033533453","doi":"10.1002/sim.2608","title":"‘A memorandum on the Present Position and Prospects of Medical Statistics and Epidemiology’ by Major Greenwood","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Hydrology and Drought Analysis","field":"Environmental Science","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"University of Cambridge; London School of Hygiene and Tropical Medicine","keywords":"Memorandum; Staffing; Medical statistics; Quarter (Canadian coin); Position (finance); Summary statistics; Mathematical statistics; History; Statistics; Political science; Law; Economics; Mathematics; Archaeology","score_opus":0.011391141148016391,"score_gpt":0.28886839334842546,"score_spread":0.2774772522004091,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2033533453","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00093071756,0.015530227,0.005175471,0.5846281,0.38344392,0.00006294025,0.0011601446,0.0002567716,0.008811702],"genre_scores_gemma":[0.02272467,0.024077319,0.013595048,0.32573193,0.5482733,0.00040862834,0.0020988367,0.00089671544,0.062193613],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9933989,0.0024060686,0.0008165808,0.00089243817,0.0018960956,0.0005900173],"domain_scores_gemma":[0.9566324,0.028587898,0.002674219,0.0020405573,0.0077344202,0.0023305274],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0194826,0.0011238327,0.0014879205,0.0028192864,0.0024077054,0.0054454156,0.0024070274,0.0036102985,0.010781434],"category_scores_gemma":[0.069349,0.00071107416,0.0013234352,0.0018092332,0.0057919817,0.0059074275,0.0035842834,0.014196224,0.008528813],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000014360332,0.000005797795,0.000062120846,0.000037187707,0.000004863922,0.000036849742,0.00018905495,0.000038916227,0.00006212399,0.0039938744,0.9858891,0.009665638],"study_design_scores_gemma":[0.000010546658,0.000024493667,0.0005601504,0.00024084246,0.0000070011893,0.00018398777,0.00013256996,0.00011430731,0.00012603097,0.005604,0.9929677,0.00002839048],"about_ca_topic_score_codex":0.002767759,"about_ca_topic_score_gemma":0.002327602,"teacher_disagreement_score":0.0194826,"about_ca_system_score_codex":0.0030359684,"about_ca_system_score_gemma":0.0046309466,"threshold_uncertainty_score":0.10303503},"labels":[],"label_agreement":null},{"id":"W2034925381","doi":"10.1002/sim.1670","title":"Estimating the incidence of subclinical infections with<i>Legionella Pneumonia</i>using data augmentation: analysis of an outbreak in The Netherlands","year":2003,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Legionella and Acanthamoeba research","field":"Biochemistry, Genetics and Molecular Biology","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"","keywords":"Subclinical infection; Outbreak; Legionella; Incidence (geometry); Epidemiology; Pneumonia; Cluster (spacecraft); Legionnaires' disease; Medicine; Population; Immunology; Biology; Virology; Environmental health; Internal medicine; Legionella pneumophila; Bacteria","score_opus":0.0533357147632414,"score_gpt":0.4054226833367912,"score_spread":0.3520869685735498,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2034925381","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99659663,0.000053551572,0.002781416,0.000037516562,0.0000033393371,0.000020914753,0.0003723979,0.000022664806,0.00011171873],"genre_scores_gemma":[0.99334913,0.000056410172,0.0044517666,0.000008922542,0.000004401706,0.00003158824,0.0019511593,0.000010875196,0.0001358734],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99807787,0.001211732,0.00015684604,0.00028740152,0.00013893245,0.00012725944],"domain_scores_gemma":[0.9738635,0.020877399,0.002142146,0.0017460891,0.0010456097,0.00032522253],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004889754,0.00048676206,0.0006407712,0.0008589133,0.00028923916,0.00097022514,0.000758003,0.0006381153,0.00032739821],"category_scores_gemma":[0.020607965,0.0004864021,0.0008979856,0.0010334374,0.0006197436,0.0007013258,0.00059410697,0.00049528875,0.000086803564],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012720021,0.00043355272,0.74300283,0.0001316094,0.00065141974,0.00075547554,0.0007278791,0.22382648,0.0027517332,0.0005116259,0.0008629764,0.025072457],"study_design_scores_gemma":[0.00006386219,0.00020473708,0.28989017,0.000018975948,0.00012110401,0.000301027,0.00043932575,0.7055742,0.0020242066,0.0005481526,0.000768382,0.000045750505],"about_ca_topic_score_codex":0.06852025,"about_ca_topic_score_gemma":0.023773817,"teacher_disagreement_score":0.06852025,"about_ca_system_score_codex":0.0011751234,"about_ca_system_score_gemma":0.00071833335,"threshold_uncertainty_score":0.1362428},"labels":[],"label_agreement":null},{"id":"W2035743530","doi":"10.1002/sim.2816","title":"HIV viral dynamic models with dropouts and missing covariates","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"HIV Research and Treatment","field":"Immunology and Microbiology","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Covariate; Missing data; Dropout (neural networks); Human immunodeficiency virus (HIV); Drop out; Viral load; Medicine; Set (abstract data type); Econometrics; Statistics; Computer science; Immunology; Mathematics; Machine learning","score_opus":0.012216009390193109,"score_gpt":0.3083770029980744,"score_spread":0.2961609936078813,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2035743530","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.34540623,0.0038069643,0.63413364,0.0056180474,0.0005436058,0.0009128514,0.0038604026,0.0007120968,0.005006163],"genre_scores_gemma":[0.9104448,0.0025887017,0.06275357,0.0009405684,0.00041961286,0.00128873,0.0027957284,0.000105095576,0.018663254],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99355173,0.0043878458,0.0002510498,0.0006190132,0.0004824127,0.0007078453],"domain_scores_gemma":[0.9441085,0.044235263,0.006176068,0.0022051332,0.002109968,0.0011649418],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02831081,0.0015750298,0.004073485,0.0014249608,0.0012032992,0.0022906885,0.0038011558,0.0030962154,0.006420369],"category_scores_gemma":[0.058726322,0.001011685,0.0023556198,0.0023174544,0.0024121718,0.0029642489,0.0026603711,0.0043720747,0.00083985715],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013843734,0.00036928803,0.021374008,0.00042324848,0.00044602476,0.001162727,0.00083084777,0.78323215,0.00034530638,0.14817394,0.004938379,0.03731979],"study_design_scores_gemma":[0.0003731645,0.0002552293,0.0022720608,0.000083442,0.00012504215,0.00014380574,0.00016285754,0.9093447,0.00020811212,0.08484282,0.0021318325,0.00005704551],"about_ca_topic_score_codex":0.014149187,"about_ca_topic_score_gemma":0.006257451,"teacher_disagreement_score":0.02831081,"about_ca_system_score_codex":0.0018387843,"about_ca_system_score_gemma":0.0017576372,"threshold_uncertainty_score":0.14972365},"labels":[],"label_agreement":null},{"id":"W2036036016","doi":"10.1002/sim.1014","title":"Selection effects in randomized trials with count data","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Count data; Selection (genetic algorithm); Selection bias; Sample size determination; Statistics; Clinical trial; Econometrics; Computer science; Population; Sample (material); Medicine; Mathematics; Internal medicine; Artificial intelligence","score_opus":0.6251665613778234,"score_gpt":0.5981809771304896,"score_spread":0.026985584247333883,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2036036016","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009387672,0.022346387,0.95019966,0.007271061,0.0012430412,0.0028742908,0.00048167226,0.0005332376,0.0056629754],"genre_scores_gemma":[0.40102604,0.017449906,0.5460851,0.010401765,0.0027917665,0.015086871,0.00062435,0.0005667984,0.005967361],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.36143568,0.58692354,0.016721742,0.011693352,0.021779375,0.0014464149],"domain_scores_gemma":[0.101028256,0.85647243,0.021452498,0.016909236,0.0037252686,0.00041237107],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.47450206,0.00323372,0.007451783,0.005280104,0.0016215862,0.005840774,0.0037619774,0.008653087,0.008071347],"category_scores_gemma":[0.7565759,0.002162109,0.008144815,0.0059972964,0.01213485,0.010795973,0.0060198572,0.008519999,0.0011260444],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004627709,0.0001777852,0.017550275,0.007974998,0.0076278085,0.0011810107,0.001544316,0.04263042,0.0007009153,0.70200944,0.0076040686,0.20637132],"study_design_scores_gemma":[0.0027105196,0.0022612882,0.005760917,0.0029858314,0.0035431948,0.0013632189,0.00024693852,0.12297223,0.002678876,0.8347903,0.020458672,0.00022800498],"about_ca_topic_score_codex":0.0011459122,"about_ca_topic_score_gemma":0.00077872496,"teacher_disagreement_score":0.5254979,"about_ca_system_score_codex":0.004077981,"about_ca_system_score_gemma":0.004095753,"threshold_uncertainty_score":0.64803255},"labels":[],"label_agreement":null},{"id":"W2036916777","doi":"10.1002/sim.3102","title":"Analysis of a nonsusceptible fraction with current status data","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Distribution Estimation and Applications","field":"Mathematics","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; University of Waterloo","funders":"","keywords":"Covariate; Statistics; Nonparametric statistics; Censoring (clinical trials); Estimator; Parametric statistics; Event (particle physics); Econometrics; Confidence interval; Population; Event data; Survival analysis; Mathematics; Medicine; Environmental health","score_opus":0.13284856464219846,"score_gpt":0.4801103101450883,"score_spread":0.34726174550288985,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2036916777","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.22037409,0.0006516768,0.77696216,0.0003236587,0.00002819829,0.00013914527,0.0004381575,0.00021543885,0.000867386],"genre_scores_gemma":[0.8637043,0.00054017664,0.130704,0.000110139124,0.00007966489,0.00027033946,0.0016849695,0.00008803999,0.00281834],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9966924,0.0017393457,0.00022428356,0.0005356257,0.00063022063,0.00017804667],"domain_scores_gemma":[0.94924146,0.04226944,0.0027521313,0.0039963787,0.0013871036,0.0003535519],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017703822,0.0006572918,0.0012367978,0.0019717491,0.00040446589,0.0011582854,0.0016909015,0.0012951079,0.0026211017],"category_scores_gemma":[0.076430395,0.00040604957,0.0010599504,0.001610092,0.0008199926,0.0022468602,0.001378417,0.0009604402,0.00050027174],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017210325,0.00046473395,0.3700089,0.0006054428,0.00084462,0.0017413262,0.001342871,0.15616995,0.009809993,0.05519054,0.0017297379,0.40037084],"study_design_scores_gemma":[0.00006256094,0.00044620514,0.099541545,0.0000955112,0.00022530563,0.0010111828,0.00031437958,0.8084308,0.006171476,0.07896566,0.0046651657,0.00007025442],"about_ca_topic_score_codex":0.002095125,"about_ca_topic_score_gemma":0.0015320225,"teacher_disagreement_score":0.017703822,"about_ca_system_score_codex":0.00041145607,"about_ca_system_score_gemma":0.0007448317,"threshold_uncertainty_score":0.09362787},"labels":[],"label_agreement":null},{"id":"W2037113505","doi":"10.1002/sim.1347","title":"Incremental net benefit in randomized clinical trials with quality‐adjusted survival","year":2003,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":51,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; Hospital for Sick Children; University of Waterloo; University of Toronto","funders":"","keywords":"Censoring (clinical trials); Statistics; Survival analysis; Analysis of covariance; Covariance; Quality-adjusted life year; Mathematics; Econometrics; Cost effectiveness","score_opus":0.6064299540665855,"score_gpt":0.5651836531183877,"score_spread":0.04124630094819781,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2037113505","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.15155363,0.06587718,0.7518612,0.01295488,0.00199372,0.004093794,0.0020141974,0.0012405578,0.0084107695],"genre_scores_gemma":[0.86273855,0.008590196,0.12055841,0.0021969585,0.00089254597,0.0030067347,0.00069057127,0.00013124826,0.0011948479],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.7882535,0.19660662,0.003866793,0.0028207381,0.007313364,0.0011389782],"domain_scores_gemma":[0.5961817,0.38482213,0.010035181,0.0055467836,0.0025999756,0.0008142369],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.15694168,0.0025910903,0.0057162046,0.006672489,0.00039651452,0.0036688102,0.0025248588,0.0037350596,0.0029804462],"category_scores_gemma":[0.36293495,0.0012351336,0.0048935656,0.0048997365,0.0030139436,0.0052227806,0.0025713241,0.003967535,0.00030394955],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.028950112,0.00047185077,0.020409621,0.010601565,0.022441454,0.0007163142,0.0006222535,0.47272173,0.0006492244,0.14934377,0.0039693275,0.28910282],"study_design_scores_gemma":[0.0054065026,0.0035518766,0.008815612,0.0018228147,0.011752142,0.0006482923,0.000101934354,0.53357226,0.0012915576,0.428503,0.00425486,0.00027917203],"about_ca_topic_score_codex":0.002290612,"about_ca_topic_score_gemma":0.0014742132,"teacher_disagreement_score":0.84305835,"about_ca_system_score_codex":0.0036794036,"about_ca_system_score_gemma":0.003711327,"threshold_uncertainty_score":0.8299969},"labels":[],"label_agreement":null},{"id":"W2037652844","doi":"10.1002/sim.2352","title":"Modelling the transmission dynamics of acute haemorrhagic conjunctivitis: application to the 2003 outbreak in Mexico","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"COVID-19 epidemiological studies","field":"Mathematics","cited_by":40,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Outbreak; Transmission (telecommunications); Coronavirus disease 2019 (COVID-19); Virology; Disease transmission; Medicine; Computer science; Disease; Internal medicine; Telecommunications; Infectious disease (medical specialty)","score_opus":0.08694857396430807,"score_gpt":0.3994271032886009,"score_spread":0.3124785293242929,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2037652844","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9831661,0.00018047697,0.013086084,0.0008923431,0.000020050185,0.000083001236,0.0009415836,0.000108046195,0.0015223506],"genre_scores_gemma":[0.99114746,0.00025199476,0.0068082334,0.000043753473,0.00001786705,0.000095287345,0.0004525566,0.000016572723,0.001166272],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99964225,0.00018362347,0.000018336945,0.000053259308,0.000021110005,0.00008146735],"domain_scores_gemma":[0.9975737,0.0017003055,0.0004050481,0.00007055247,0.00015382926,0.000096488344],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0012746267,0.000643712,0.00062359986,0.00065518596,0.0006498769,0.0012076176,0.0010302656,0.001557321,0.0016476529],"category_scores_gemma":[0.00516971,0.00043142957,0.0008580446,0.00079376274,0.000539099,0.00070701947,0.00076366955,0.0007110833,0.00011288442],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001416508,0.00013443088,0.027442232,0.000030392608,0.00008625355,0.00021480191,0.00021219515,0.96575004,0.00040101586,0.0024852678,0.0004887151,0.0026130711],"study_design_scores_gemma":[0.000050240375,0.00007398977,0.0053576627,0.000008452286,0.000029896935,0.000025144096,0.00015719392,0.993206,0.000089835696,0.00071474444,0.0002727954,0.0000140165685],"about_ca_topic_score_codex":0.1631677,"about_ca_topic_score_gemma":0.07358585,"teacher_disagreement_score":0.1631677,"about_ca_system_score_codex":0.0028586385,"about_ca_system_score_gemma":0.0016464745,"threshold_uncertainty_score":0.32443595},"labels":[],"label_agreement":null},{"id":"W2037889113","doi":"10.1002/sim.5530","title":"Efficient analysis of case‐control studies with sample weights","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital","funders":"","keywords":"Estimator; Covariate; Statistics; Sample (material); Population; Sample size determination; Mathematics; Bootstrapping (finance); Robustness (evolution); Econometrics; Computer science","score_opus":0.1146731685975503,"score_gpt":0.4553308516164282,"score_spread":0.3406576830188779,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2037889113","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0075977673,0.00079212163,0.9902367,0.0003538218,0.000057906364,0.0004378643,0.000075781696,0.000077273195,0.00037073734],"genre_scores_gemma":[0.23923804,0.0015793404,0.752868,0.00037080096,0.00023713829,0.0044761957,0.00044092388,0.00006856123,0.00072106987],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8450673,0.13609628,0.0053353505,0.0044383137,0.008407385,0.0006553762],"domain_scores_gemma":[0.52558386,0.42874008,0.01672916,0.023092894,0.005342094,0.0005118879],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.15428536,0.0012467116,0.0037344105,0.0047080414,0.0007083461,0.0027519201,0.003099924,0.0024566173,0.002423664],"category_scores_gemma":[0.43598393,0.0015767067,0.0020872103,0.0041157496,0.002608171,0.0034480647,0.0034212582,0.002560854,0.00027724533],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008235381,0.00035476286,0.02667009,0.0021775977,0.004093574,0.00080805237,0.0006571353,0.28059602,0.0020763124,0.355106,0.002648629,0.3239883],"study_design_scores_gemma":[0.00053529034,0.00038843034,0.006086852,0.0005025428,0.0008616698,0.00032773014,0.00013115439,0.56730026,0.0014675966,0.41692615,0.005414659,0.00005766115],"about_ca_topic_score_codex":0.0010401595,"about_ca_topic_score_gemma":0.00083986676,"teacher_disagreement_score":0.15428536,"about_ca_system_score_codex":0.0016820288,"about_ca_system_score_gemma":0.0027959223,"threshold_uncertainty_score":0.8159487},"labels":[],"label_agreement":null},{"id":"W2038189661","doi":"10.1002/sim.2800","title":"Sampling an unknown universe: problems of researching mass casualty incidents (a history of ECRU's field research)","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Disaster Response and Management","field":"Health Professions","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Carleton University","funders":"","keywords":"Sampling (signal processing); Operations research; Disaster research; Field (mathematics); Library science; History; Sociology; Computer science; Telecommunications; Meteorology; Engineering; Geography; Mathematics","score_opus":0.3175418562933978,"score_gpt":0.5628297905198613,"score_spread":0.24528793422646344,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2038189661","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01699171,0.19078322,0.6255199,0.12969081,0.0051645045,0.0017570436,0.00037482183,0.00018761182,0.029530477],"genre_scores_gemma":[0.22703587,0.12159152,0.58263636,0.04483978,0.009131394,0.0087162675,0.00018836519,0.00026403257,0.0055964347],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","domain_scores_codex":[0.70055914,0.27139434,0.006122856,0.008016254,0.012657202,0.0012502538],"domain_scores_gemma":[0.3662556,0.5910913,0.008228385,0.022161407,0.010590058,0.0016733077],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2426639,0.001131378,0.0029465675,0.01051381,0.00951465,0.012385382,0.0037044745,0.005479981,0.001977438],"category_scores_gemma":[0.32251942,0.0017075975,0.0013967428,0.014299445,0.07589938,0.015315053,0.00757614,0.009597062,0.00061882223],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009018874,0.0001143673,0.0074037956,0.001727987,0.00012124063,0.00021976107,0.036844686,0.002267085,0.00015618834,0.6597576,0.018960332,0.27233687],"study_design_scores_gemma":[0.000063396954,0.00020745705,0.0042712297,0.0044184653,0.00004816018,0.00062890304,0.018566225,0.0030892065,0.0005334907,0.8445382,0.123498194,0.00013703082],"about_ca_topic_score_codex":0.0191501,"about_ca_topic_score_gemma":0.011253635,"teacher_disagreement_score":0.7573361,"about_ca_system_score_codex":0.00962683,"about_ca_system_score_gemma":0.012518196,"threshold_uncertainty_score":0.9339302},"labels":[],"label_agreement":null},{"id":"W2040330324","doi":"10.1002/sim.4486","title":"Optimal allocation of participants for the estimation of selection, preference and treatment effects in the two‐stage randomised trial design","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Preference; Selection (genetic algorithm); Randomized controlled trial; Sample size determination; Treatment and control groups; Treatment effect; Clinical trial; Medicine; Statistics; Mathematics; Computer science; Internal medicine","score_opus":0.5865424133203638,"score_gpt":0.5720278292840314,"score_spread":0.014514584036332434,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2040330324","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00814229,0.0014625732,0.9747747,0.0015683962,0.00027024213,0.012290294,0.00016558623,0.00017474785,0.0011510932],"genre_scores_gemma":[0.0944422,0.00086792256,0.865628,0.0009109583,0.00013872643,0.037242442,0.00013628826,0.000060258895,0.0005731098],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.44818562,0.5138119,0.013703429,0.010573126,0.012149926,0.001576029],"domain_scores_gemma":[0.5855312,0.3701392,0.016079027,0.022276359,0.0047058053,0.0012684076],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.33866426,0.0031649256,0.010993354,0.003361147,0.001101484,0.0040788683,0.0034855236,0.008027065,0.0052690133],"category_scores_gemma":[0.5205918,0.0028514697,0.005943477,0.0030768365,0.007426541,0.00832038,0.004458591,0.0076058875,0.0013124289],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.02049376,0.0011856959,0.006118289,0.00955271,0.0037440606,0.00046949193,0.0029536874,0.0894357,0.0042583956,0.53877574,0.0050388807,0.31797355],"study_design_scores_gemma":[0.016796527,0.0076707182,0.0034100467,0.0028062533,0.0016778868,0.00044714066,0.0002590696,0.2624953,0.005845531,0.6872484,0.010909647,0.00043342263],"about_ca_topic_score_codex":0.00066530367,"about_ca_topic_score_gemma":0.0006004013,"teacher_disagreement_score":0.6613357,"about_ca_system_score_codex":0.0035229044,"about_ca_system_score_gemma":0.008668136,"threshold_uncertainty_score":0.8155447},"labels":[],"label_agreement":null},{"id":"W2040984821","doi":"10.1002/sim.2507","title":"An innovative application of Bayesian disease mapping methods to patient safety research: a Canadian adverse medical event study","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":28,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Victoria; Children’s Health Research Institute; University of British Columbia","funders":"Canadian Institutes of Health Research; Ministry of Health, British Columbia","keywords":"Adverse effect; Event (particle physics); Bayesian probability; Computer science; Adverse Event Reporting System; Patient safety; Medicine; Statistics; Artificial intelligence; Internal medicine; Mathematics; Health care","score_opus":0.3067687218428774,"score_gpt":0.5485006337421326,"score_spread":0.24173191189925525,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2040984821","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3431278,0.01405364,0.5050819,0.0540882,0.0006714592,0.002681849,0.0043482827,0.00027431452,0.0756726],"genre_scores_gemma":[0.7648913,0.0054273913,0.22013344,0.001966885,0.0001421769,0.00073828147,0.00067447266,0.00003778489,0.005988255],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9933662,0.0042872257,0.00017007791,0.00044608078,0.0015081345,0.00022229884],"domain_scores_gemma":[0.9887002,0.00639436,0.00045623252,0.0008728605,0.0029082533,0.0006681466],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0146543635,0.00044411002,0.0005213286,0.0025740943,0.002763786,0.0016924646,0.0017844705,0.0008950573,0.0018900081],"category_scores_gemma":[0.03443428,0.00026778108,0.0007111023,0.004536978,0.0016966348,0.0007180404,0.0020987762,0.001218094,0.00009731954],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00046158483,0.00030611455,0.13799395,0.000665349,0.0005489564,0.0010199221,0.006951452,0.038656395,0.0015436596,0.30047503,0.024641894,0.48673573],"study_design_scores_gemma":[0.0010615109,0.00056021754,0.38043243,0.0009218737,0.000984611,0.0010239853,0.007641582,0.2134746,0.001298406,0.21794944,0.17420521,0.00044628244],"about_ca_topic_score_codex":0.95179725,"about_ca_topic_score_gemma":0.9646925,"teacher_disagreement_score":0.048202753,"about_ca_system_score_codex":0.0138831865,"about_ca_system_score_gemma":0.043431364,"threshold_uncertainty_score":0.10073006},"labels":[],"label_agreement":null},{"id":"W2041326463","doi":"10.1002/sim.5640","title":"Doubly robust estimation, optimally truncated inverse‐intensity weighting and increment‐based methods for the analysis of irregularly observed longitudinal data","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; Hospital for Sick Children; University of Toronto; St. Joseph’s Healthcare Hamilton","funders":"","keywords":"Estimator; Weighting; Truncation (statistics); Applied mathematics; Inverse; Mathematics; Computer science; Mean squared error; Statistics; Algorithm; Medicine","score_opus":0.41399082421169303,"score_gpt":0.48337734009618,"score_spread":0.06938651588448697,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2041326463","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0024314818,0.00027979803,0.9968966,0.00009942606,0.000025491709,0.00002454583,0.000038231403,0.0000642066,0.0001401424],"genre_scores_gemma":[0.09652239,0.0012296233,0.8985773,0.0001851623,0.00020111026,0.0004476699,0.0005007496,0.0001758344,0.0021601738],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9927979,0.0052944864,0.00031067105,0.000625186,0.00083459,0.00013721046],"domain_scores_gemma":[0.972429,0.020024506,0.0023011856,0.003411122,0.0015360519,0.00029807983],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01610067,0.00084642053,0.0012941961,0.001822397,0.000444356,0.0010780381,0.0024304711,0.0013476709,0.0015844128],"category_scores_gemma":[0.061023567,0.00072312885,0.0015517628,0.0019129134,0.0014690204,0.0018171119,0.0020801658,0.0024569784,0.00052316196],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00029797954,0.00018211598,0.008088259,0.00047951896,0.0006194226,0.00024111404,0.0003669135,0.39246473,0.0037254337,0.3280677,0.00403309,0.2614337],"study_design_scores_gemma":[0.00003238493,0.00008215111,0.0011177763,0.000042191812,0.000058943686,0.00006869398,0.000020690608,0.91285866,0.0007165747,0.081744514,0.0032121604,0.000045229783],"about_ca_topic_score_codex":0.002804984,"about_ca_topic_score_gemma":0.0027809395,"teacher_disagreement_score":0.01610067,"about_ca_system_score_codex":0.0007101623,"about_ca_system_score_gemma":0.0017280221,"threshold_uncertainty_score":0.08514953},"labels":[],"label_agreement":null},{"id":"W2041490883","doi":"10.1002/sim.6484","title":"A flexible mixed‐effect negative binomial regression model for detecting unusual increases in MRI lesion counts in individual multiple sclerosis patients","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada; Multiple Sclerosis Society of Canada","keywords":"Negative binomial distribution; Random effects model; Dirichlet process; Dirichlet distribution; Statistics; Bayesian probability; Parametric statistics; Mixed model; Conditional probability distribution; Mathematics; Computer science; Medicine; Poisson distribution; Meta-analysis; Pathology","score_opus":0.11343398538156718,"score_gpt":0.35615538315047834,"score_spread":0.24272139776891116,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2041490883","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017869892,0.00091159507,0.97835207,0.0009327249,0.000118199525,0.00020288612,0.00047751438,0.0003731832,0.0007619692],"genre_scores_gemma":[0.51506865,0.0017400326,0.47109288,0.0011801596,0.0003744192,0.002189724,0.0015215513,0.00022649708,0.006606097],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9884426,0.008419648,0.0003281021,0.0015802466,0.00083952805,0.00038982838],"domain_scores_gemma":[0.97564924,0.019942958,0.0018861332,0.0011280654,0.0010305699,0.0003629556],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026532132,0.00169054,0.0034623588,0.0025227533,0.0005382592,0.002086412,0.0061143865,0.0026829,0.003374527],"category_scores_gemma":[0.039588135,0.0014705474,0.004067017,0.0018826813,0.0016483645,0.0020158861,0.0018308399,0.003276894,0.0007303871],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009863606,0.00021108524,0.0140092885,0.0005447146,0.0018011421,0.0010004456,0.0005001744,0.7938787,0.0021844655,0.10384,0.0030862715,0.07795738],"study_design_scores_gemma":[0.00015582502,0.00011795124,0.0011899011,0.00005327777,0.00030562814,0.00014700225,0.000018946548,0.97252786,0.00029339685,0.023387853,0.0017531436,0.000049185335],"about_ca_topic_score_codex":0.008092685,"about_ca_topic_score_gemma":0.0067854705,"teacher_disagreement_score":0.026532132,"about_ca_system_score_codex":0.0014086532,"about_ca_system_score_gemma":0.0018229655,"threshold_uncertainty_score":0.14031696},"labels":[],"label_agreement":null},{"id":"W2042643721","doi":"10.1002/sim.2254","title":"Application of reliability coefficients in cDNA microarray data analysis","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Gene expression and cancer classification","field":"Biochemistry, Genetics and Molecular Biology","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Lunenfeld-Tanenbaum Research Institute; Mount Sinai Hospital","funders":"Canadian Institutes of Health Research","keywords":"Normalization (sociology); Computer science; Data mining; Microarray analysis techniques; Gene chip analysis; Reliability (semiconductor); Microarray; Statistics; Computational biology; Bioinformatics; Mathematics; Biology; Gene expression; Gene; Genetics","score_opus":0.016859038281986833,"score_gpt":0.33940636258979473,"score_spread":0.3225473243078079,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2042643721","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007713193,0.00051968783,0.99072766,0.00011221714,0.000038895392,0.000080359125,0.00006746915,0.00031404823,0.00042645296],"genre_scores_gemma":[0.22940949,0.0009405381,0.7674382,0.000115068615,0.00020228361,0.00055503234,0.00042571008,0.0003524755,0.0005612874],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.97113365,0.018683841,0.0014377033,0.0021290747,0.0061545162,0.00046117444],"domain_scores_gemma":[0.85209,0.11854394,0.007699459,0.0072482466,0.013696258,0.0007221445],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.033799168,0.0014523213,0.0019833078,0.0058241105,0.00093436777,0.0017908349,0.0015911773,0.0014521913,0.0006541499],"category_scores_gemma":[0.14402004,0.00083062897,0.001767825,0.004446277,0.0018945084,0.0015834182,0.001563167,0.0034142598,0.0006335564],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002993625,0.00013697633,0.023253826,0.00062019884,0.0006360889,0.00046889464,0.00076460524,0.5431889,0.01571255,0.07642381,0.003958229,0.33453655],"study_design_scores_gemma":[0.000029078352,0.00019287638,0.0051800557,0.00009436277,0.00010439877,0.0004299857,0.00011864989,0.92728996,0.009220205,0.05183761,0.005371611,0.00013119577],"about_ca_topic_score_codex":0.0021620914,"about_ca_topic_score_gemma":0.0014795081,"teacher_disagreement_score":0.033799168,"about_ca_system_score_codex":0.0013400613,"about_ca_system_score_gemma":0.0022698757,"threshold_uncertainty_score":0.17874926},"labels":[],"label_agreement":null},{"id":"W2045286429","doi":"10.1002/sim.3751","title":"Assessment of risks by predicting counterfactuals","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Risk and Safety Analysis","field":"Decision Sciences","cited_by":41,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"","keywords":"Counterfactual conditional; Econometrics; Computer science; Statistics; Mathematics; Psychology; Counterfactual thinking","score_opus":0.11143002454222618,"score_gpt":0.49855336493011776,"score_spread":0.3871233403878916,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2045286429","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.035512656,0.0016525978,0.95569086,0.0014271106,0.00026509797,0.00063172297,0.00050445093,0.0003050372,0.0040103914],"genre_scores_gemma":[0.57719964,0.0018661615,0.4165442,0.00066368544,0.00022137527,0.0019032223,0.0006286242,0.00008630722,0.0008867822],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9043729,0.077250004,0.003977449,0.0047898814,0.009116684,0.00049310655],"domain_scores_gemma":[0.58892673,0.35387403,0.026155915,0.025048893,0.0054547125,0.0005396412],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.109729,0.0020547064,0.0018670718,0.0026959237,0.00064723444,0.005041418,0.0022936894,0.0026113347,0.0034455277],"category_scores_gemma":[0.32465628,0.0006488851,0.002836979,0.001842833,0.0033577892,0.0062792804,0.0037108033,0.0038502524,0.00060371857],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013997221,0.0004148735,0.048672557,0.0014426894,0.0018100815,0.0005146425,0.001312767,0.20339195,0.0016552792,0.5068813,0.0038246827,0.22867942],"study_design_scores_gemma":[0.00020087618,0.0014428382,0.009861045,0.000949086,0.00079069607,0.0010028734,0.00037538132,0.3481295,0.004296827,0.6226126,0.010009559,0.00032872308],"about_ca_topic_score_codex":0.0007554219,"about_ca_topic_score_gemma":0.0004909638,"teacher_disagreement_score":0.109729,"about_ca_system_score_codex":0.0013067643,"about_ca_system_score_gemma":0.0017547107,"threshold_uncertainty_score":0.5803094},"labels":[],"label_agreement":null},{"id":"W2045335212","doi":"10.1002/sim.776","title":"Power comparison of robust approximate and non‐parametric tests for the analysis of cross‐over trials","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"HEC Montréal","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Parametric statistics; Sample size determination; Mathematics; Ordinary least squares; Statistics; Robustness (evolution); Covariance; Applied mathematics; Statistical hypothesis testing","score_opus":0.3388194152568817,"score_gpt":0.5831102429260211,"score_spread":0.2442908276691394,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2045335212","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01831309,0.0024538864,0.9753671,0.00036093243,0.00021062,0.0010462005,0.00016975372,0.0002743327,0.0018040108],"genre_scores_gemma":[0.39197066,0.0013517579,0.5915368,0.00062952755,0.00027003168,0.012180153,0.00055554847,0.00038619858,0.0011192731],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.7767181,0.1937647,0.004942778,0.008409897,0.015299743,0.00086481514],"domain_scores_gemma":[0.361268,0.5911325,0.012660278,0.024935702,0.009252546,0.0007509287],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.15711863,0.0014599945,0.0036853256,0.002910733,0.0005837078,0.0021668999,0.0031940085,0.003455564,0.004573278],"category_scores_gemma":[0.5137746,0.0009728767,0.0030314948,0.00243654,0.0035173981,0.0037912955,0.002639754,0.0026125011,0.0006769087],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.02066952,0.00072501873,0.012889894,0.0043253354,0.007857386,0.0004540986,0.0013514404,0.09671805,0.0049774284,0.2106388,0.0053660395,0.63402694],"study_design_scores_gemma":[0.008055664,0.018397752,0.03418986,0.0015382724,0.002911027,0.0013819565,0.0005328316,0.55918354,0.014210734,0.33454847,0.024499742,0.0005501779],"about_ca_topic_score_codex":0.00036387425,"about_ca_topic_score_gemma":0.00024925082,"teacher_disagreement_score":0.8428814,"about_ca_system_score_codex":0.0013334475,"about_ca_system_score_gemma":0.0019078456,"threshold_uncertainty_score":0.8309327},"labels":[],"label_agreement":null},{"id":"W2045862988","doi":"10.1002/sim.1696","title":"A non‐parametric procedure for evaluating treatment effect in the meta‐analysis of survival data","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Winnipeg","funders":"","keywords":"Statistics; Hazard ratio; Censoring (clinical trials); Mathematics; Logarithm; Confidence interval; Multiplicative function; Contrast (vision); Econometrics; Computer science; Artificial intelligence","score_opus":0.769066250341494,"score_gpt":0.6656981586317418,"score_spread":0.10336809170975225,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2045862988","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0020069466,0.0047168485,0.985329,0.00056696753,0.00071924034,0.003355624,0.0011102915,0.0012730956,0.0009220796],"genre_scores_gemma":[0.05801951,0.002355125,0.90487105,0.00097259955,0.0005294235,0.03074644,0.00089001254,0.0006295096,0.0009863954],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5995156,0.36460564,0.01091739,0.009789926,0.014295976,0.0008753762],"domain_scores_gemma":[0.6337366,0.31364992,0.010587437,0.037208173,0.004289033,0.0005288584],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.22118072,0.003132436,0.008926284,0.015033694,0.0012423783,0.0043431385,0.0054663704,0.004527929,0.008142273],"category_scores_gemma":[0.40844193,0.0017818122,0.030390635,0.011951713,0.0031808487,0.0041974518,0.005974961,0.0101371175,0.0013582982],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008721946,0.0004848578,0.006857715,0.03372641,0.1637007,0.0012181893,0.001650962,0.04300147,0.006210484,0.11474009,0.020414308,0.59927285],"study_design_scores_gemma":[0.009191752,0.011941325,0.020941237,0.007843395,0.10386539,0.0031584487,0.0010333676,0.17027488,0.017686179,0.4987487,0.1537256,0.0015897453],"about_ca_topic_score_codex":0.0010325005,"about_ca_topic_score_gemma":0.0007979509,"teacher_disagreement_score":0.77881926,"about_ca_system_score_codex":0.0018409841,"about_ca_system_score_gemma":0.0041079978,"threshold_uncertainty_score":0.9604228},"labels":[],"label_agreement":null},{"id":"W2046459096","doi":"10.1002/sim.6167","title":"Bootstrap confidence intervals for loess-based calibration curves","year":2014,"lang":"en","type":"letter","venue":"Statistics in Medicine","topic":"Soil Geostatistics and Mapping","field":"Environmental Science","cited_by":21,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sunnybrook Hospital; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Heart and Stroke Foundation of Canada","keywords":"Erasmus+; Library science; Citation; Confidence interval; Sociology; Operations research; Psychology; History; Computer science; Statistics; Mathematics; Art history","score_opus":0.04343705579199119,"score_gpt":0.3279541332850265,"score_spread":0.2845170774930353,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2046459096","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029186444,0.021683743,0.7691175,0.14291786,0.00709838,0.00012619942,0.0014914329,0.002280899,0.026097594],"genre_scores_gemma":[0.6133739,0.008331921,0.2847966,0.060133226,0.021359464,0.0005580892,0.00197164,0.0012700742,0.00820514],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9779221,0.014889134,0.0011702764,0.0011668785,0.004445757,0.00040572835],"domain_scores_gemma":[0.6576554,0.29144835,0.0061341505,0.02282828,0.020527802,0.0014059279],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.038791135,0.00056616997,0.0013440447,0.0029188418,0.0006322321,0.0026190735,0.0034716027,0.00774907,0.0040019257],"category_scores_gemma":[0.37575704,0.0004816357,0.0011018425,0.0030404804,0.0025634763,0.003152275,0.00185741,0.009641954,0.0015944951],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002150843,0.00023175277,0.011059155,0.00054624275,0.0004860052,0.001588288,0.0012225861,0.025546653,0.0017900262,0.17999159,0.20739856,0.56798834],"study_design_scores_gemma":[0.00053776737,0.00040161578,0.017285751,0.0018797334,0.00028807408,0.005770439,0.00036942904,0.15881756,0.0041638194,0.65620977,0.15396944,0.00030668863],"about_ca_topic_score_codex":0.0015135594,"about_ca_topic_score_gemma":0.0015968926,"teacher_disagreement_score":0.038791135,"about_ca_system_score_codex":0.0014473667,"about_ca_system_score_gemma":0.0009325991,"threshold_uncertainty_score":0.20514959},"labels":[],"label_agreement":null},{"id":"W2047326606","doi":"10.1002/sim.2519","title":"Joint estimation of time‐dependent and non‐linear effects of continuous covariates on survival","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":121,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Covariate; Mathematics; Generalization; Statistics; Applied mathematics; Parametric statistics; Maximum likelihood; Contrast (vision); Proportional hazards model; Regression; Linear regression; Regression analysis; Econometrics; Computer science; Mathematical analysis","score_opus":0.02742268541216722,"score_gpt":0.35343593669833295,"score_spread":0.32601325128616576,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2047326606","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.020500759,0.00060756295,0.97805834,0.00025206213,0.000030598207,0.000032277738,0.00010533317,0.00011489017,0.0002982341],"genre_scores_gemma":[0.7352803,0.003975526,0.25206354,0.00023287954,0.00028201856,0.00063516514,0.0009778164,0.00021226401,0.0063406304],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9852562,0.011246285,0.00045568118,0.0016001377,0.0010669186,0.00037470926],"domain_scores_gemma":[0.91158277,0.0783255,0.003640217,0.004735881,0.0013146577,0.00040091813],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.028900722,0.0010935658,0.0018893309,0.001613714,0.00032152326,0.0016990369,0.002180855,0.0014634682,0.002024702],"category_scores_gemma":[0.075301185,0.0011504622,0.0035653354,0.0021229086,0.0024139727,0.0031369345,0.002959142,0.002784364,0.00033982054],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00049244665,0.00019863265,0.030791158,0.000659895,0.0014861338,0.0004384954,0.0006933265,0.59088933,0.0027568028,0.22292268,0.00085845124,0.14781271],"study_design_scores_gemma":[0.00006233235,0.00035006687,0.008801004,0.00008068221,0.0002722347,0.00015932872,0.00006682522,0.85084236,0.0018353469,0.13537818,0.002078742,0.00007283166],"about_ca_topic_score_codex":0.0029108382,"about_ca_topic_score_gemma":0.0023931502,"teacher_disagreement_score":0.028900722,"about_ca_system_score_codex":0.0009199546,"about_ca_system_score_gemma":0.0026999637,"threshold_uncertainty_score":0.15284348},"labels":[],"label_agreement":null},{"id":"W2047363455","doi":"10.1002/sim.2890","title":"Mapping disability‐adjusted life years: a Bayesian hierarchical model framework for burden of disease and injury assessment","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":28,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Child and Family Research Institute; University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Michael Smith Health Research BC; Child and Family Research Institute","keywords":"Bayesian probability; Burden of disease; Computer science; Hierarchical database model; Disease; Statistics; Econometrics; Medicine; Artificial intelligence; Data mining; Mathematics; Internal medicine","score_opus":0.2060665600718926,"score_gpt":0.47105916337036907,"score_spread":0.26499260329847646,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2047363455","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0015226698,0.0006565556,0.9946713,0.00071010937,0.000038897608,0.000073913165,0.00064465415,0.00014924335,0.001532591],"genre_scores_gemma":[0.18926063,0.0043099327,0.79430455,0.00058918615,0.0003736615,0.0017647269,0.0029092892,0.00028169237,0.0062063322],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99526453,0.0034165003,0.00015884846,0.00044903718,0.0005565841,0.00015455553],"domain_scores_gemma":[0.9950152,0.003677724,0.00043925198,0.00030498955,0.00041141934,0.00015138058],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008884617,0.0011141484,0.0018614066,0.0028524788,0.0007833791,0.0025648202,0.0033946806,0.0014433896,0.00391023],"category_scores_gemma":[0.01909281,0.00090270897,0.002106505,0.0038583516,0.0012709161,0.0022488944,0.002112788,0.0024098172,0.00091067323],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00004426783,0.0000657688,0.004141923,0.000199322,0.00035120224,0.0001284612,0.0004298953,0.43299726,0.0002230395,0.4670553,0.007752393,0.08661127],"study_design_scores_gemma":[0.000024242143,0.000040641182,0.001446413,0.000094211304,0.00007237746,0.00008493508,0.00008834335,0.5152257,0.000059898,0.46984679,0.01296455,0.00005186479],"about_ca_topic_score_codex":0.049638722,"about_ca_topic_score_gemma":0.04200092,"teacher_disagreement_score":0.049638722,"about_ca_system_score_codex":0.0033484655,"about_ca_system_score_gemma":0.004079835,"threshold_uncertainty_score":0.09869957},"labels":[],"label_agreement":null},{"id":"W2048605164","doi":"10.1002/1097-0258(20000915/30)19:17/18<2421::aid-sim579>3.0.co;2-c","title":"Parametric bootstrap and penalized quasi-likelihood inference in conditional autoregressive models","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":59,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"BC Research (Canada); Simon Fraser University; British Columbia Centre of Excellence for Women's Health","funders":"","keywords":"Autoregressive model; Inference; Econometrics; Parametric statistics; Computer science; Parametric model; Statistics; Maximum likelihood; Mathematics; Artificial intelligence","score_opus":0.11185539004052535,"score_gpt":0.43240643843448806,"score_spread":0.3205510483939627,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2048605164","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00214781,0.00016027367,0.9971762,0.00010898729,0.000011103416,0.000013526787,0.000032956836,0.00008905554,0.00026014043],"genre_scores_gemma":[0.24405187,0.0008852059,0.752032,0.00023019877,0.00016058407,0.0004688763,0.00044804555,0.00021540248,0.0015077409],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9865306,0.011701884,0.00022648375,0.00046446687,0.0008683095,0.00020821726],"domain_scores_gemma":[0.9407039,0.052572526,0.0017633815,0.003432533,0.0012512091,0.00027639567],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016512658,0.0006471247,0.0011930804,0.0017177698,0.00061473,0.0012955243,0.0027039356,0.0013944787,0.0023934755],"category_scores_gemma":[0.08749331,0.00071755605,0.0013968971,0.0019097388,0.002575674,0.00198717,0.0022201776,0.0023804419,0.0005142206],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00010253845,0.00008577259,0.003549641,0.00027246674,0.00022987298,0.00027477136,0.00029898257,0.34274775,0.00076012954,0.5662777,0.0025814488,0.08281896],"study_design_scores_gemma":[0.000018423434,0.000029307335,0.0006531121,0.000030002739,0.0000137494935,0.00006873658,0.000025810361,0.77167153,0.0002750692,0.22579359,0.0014013492,0.000019282586],"about_ca_topic_score_codex":0.003920499,"about_ca_topic_score_gemma":0.0035911582,"teacher_disagreement_score":0.016512658,"about_ca_system_score_codex":0.00076955435,"about_ca_system_score_gemma":0.0012863606,"threshold_uncertainty_score":0.087328374},"labels":[],"label_agreement":null},{"id":"W2049450020","doi":"10.1002/sim.3804","title":"Analysis of interval‐censored disease progression data via multi‐state models under a nonignorable inspection process","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Canadian Institutes of Health Research","keywords":"Computer science; Process (computing); Interval (graph theory); Counting process; Estimation; Statistics; Expectation–maximization algorithm; Econometrics; Data mining; Maximum likelihood; Mathematics; Engineering","score_opus":0.13466788280961697,"score_gpt":0.47681463888454273,"score_spread":0.34214675607492573,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2049450020","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2022059,0.0004002952,0.795028,0.00071445794,0.0000258529,0.000111502064,0.0004466872,0.00026268803,0.00080469204],"genre_scores_gemma":[0.94350994,0.0004691612,0.051822703,0.000081905244,0.000055091357,0.0002809363,0.0006450449,0.000034716537,0.0031005212],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9970252,0.0016514309,0.00012408341,0.0005502732,0.00038043683,0.00026855135],"domain_scores_gemma":[0.9582656,0.03501404,0.003739849,0.001619696,0.00089222426,0.0004687595],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014049017,0.00093099044,0.0019196658,0.0015028071,0.00063947326,0.0016788122,0.0020153162,0.0016647704,0.002033841],"category_scores_gemma":[0.033187587,0.0008198294,0.0018297109,0.0015850645,0.0019367058,0.0022403756,0.0016779245,0.002658857,0.0002473168],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00035340906,0.00016874401,0.013348031,0.00012339937,0.00019626331,0.00038084536,0.00044141858,0.8768346,0.0011259952,0.0853931,0.00047243823,0.02116183],"study_design_scores_gemma":[0.00001703504,0.000061527215,0.0022852856,0.000009409526,0.000025873545,0.000033398184,0.000024274002,0.97806156,0.00019154571,0.019116417,0.00015499645,0.000018688981],"about_ca_topic_score_codex":0.008445511,"about_ca_topic_score_gemma":0.005850521,"teacher_disagreement_score":0.014049017,"about_ca_system_score_codex":0.0014056761,"about_ca_system_score_gemma":0.0015558891,"threshold_uncertainty_score":0.074299216},"labels":[],"label_agreement":null},{"id":"W2050134213","doi":"10.1002/sim.4132","title":"Estimating and testing for center effects in competing risks","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Medical College of Wisconsin","keywords":"Covariate; Statistics; Proportional hazards model; Random effects model; Context (archaeology); Computer science; Monte Carlo method; Econometrics; Mathematics; Medicine; Meta-analysis","score_opus":0.5113882250710012,"score_gpt":0.4856096886464182,"score_spread":0.025778536424583054,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2050134213","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02877371,0.00040980804,0.9690801,0.00047320029,0.000056984645,0.00024490277,0.00012273378,0.0002720113,0.0005666229],"genre_scores_gemma":[0.39019188,0.00044216923,0.6060092,0.00042086636,0.0001294407,0.0013794017,0.0006017772,0.00014124427,0.00068406045],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8218643,0.15101303,0.005628672,0.009381313,0.010130518,0.0019821383],"domain_scores_gemma":[0.28697404,0.674822,0.01451477,0.017243104,0.0052910983,0.0011550187],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1808051,0.0018958884,0.0042571793,0.0049843383,0.0016690977,0.003943009,0.0062376456,0.0037656836,0.0033441195],"category_scores_gemma":[0.45882377,0.0013601065,0.0052951775,0.00466468,0.006394695,0.0044635762,0.0060773175,0.0040649204,0.00035935035],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014923142,0.00048298048,0.13719802,0.0008954547,0.0037352375,0.0021972281,0.001962204,0.21729457,0.0014238376,0.35803065,0.0032230972,0.2720644],"study_design_scores_gemma":[0.00030319285,0.00083135994,0.0150784105,0.00020990614,0.00050674216,0.0010340275,0.0004994293,0.59741926,0.0024290048,0.37926066,0.0021625455,0.0002655762],"about_ca_topic_score_codex":0.0037482113,"about_ca_topic_score_gemma":0.0023721568,"teacher_disagreement_score":0.1808051,"about_ca_system_score_codex":0.0018694086,"about_ca_system_score_gemma":0.0050576036,"threshold_uncertainty_score":0.9562002},"labels":[],"label_agreement":null},{"id":"W2050464345","doi":"10.1002/sim.4501","title":"Ratio of geometric means to analyze continuous outcomes in meta‐analysis: comparison to mean differences and ratio of arithmetic means using empiric data and simulation","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":46,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Institute for Clinical Evaluative Sciences; Population Health Research Institute; Public Health Ontario; Hospital for Sick Children; Health Sciences Centre; Sunnybrook Health Science Centre; SickKids Foundation; University of Toronto; St. Michael's Hospital","funders":"Canadian Institutes of Health Research","keywords":"Pooling; Statistics; Standard deviation; Post-hoc analysis; Meta-analysis; Post hoc; Computer science; Confidence interval; Geometric mean; Econometrics; Mathematics; Medicine; Artificial intelligence; Internal medicine","score_opus":0.7130185644199137,"score_gpt":0.5692878424835579,"score_spread":0.14373072193635583,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2050464345","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015826592,0.0418123,0.9201401,0.0020887046,0.0034347821,0.0096656615,0.0026144877,0.0018220101,0.002595406],"genre_scores_gemma":[0.14561686,0.011317645,0.8069878,0.0015587943,0.00061613595,0.03039699,0.0016351244,0.0007223533,0.0011482286],"study_design_codex":"meta_analysis","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.49146232,0.46199265,0.016474174,0.013661028,0.01567848,0.0007313752],"domain_scores_gemma":[0.50335836,0.458304,0.016815837,0.015099511,0.00587985,0.00054252637],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.33150634,0.0053688744,0.021759918,0.013740149,0.00068376947,0.0053107487,0.00540057,0.005249742,0.009171785],"category_scores_gemma":[0.51065964,0.002428762,0.03932569,0.010478001,0.0037341232,0.008047255,0.0039644144,0.008150665,0.0013121357],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.015845792,0.0007649537,0.015203987,0.08298235,0.3347311,0.0007580582,0.0018464777,0.170671,0.0016867643,0.07542916,0.023327405,0.27675292],"study_design_scores_gemma":[0.015608912,0.010151755,0.010344051,0.021855865,0.12348771,0.0012993637,0.0007218371,0.50232077,0.005733597,0.26434082,0.042600058,0.0015352458],"about_ca_topic_score_codex":0.0016525578,"about_ca_topic_score_gemma":0.0011604738,"teacher_disagreement_score":0.6684936,"about_ca_system_score_codex":0.0035661077,"about_ca_system_score_gemma":0.0035705802,"threshold_uncertainty_score":0.8243717},"labels":[],"label_agreement":null},{"id":"W2052429290","doi":"10.1002/sim.4319","title":"On the association between variables with lower detection limits","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Distribution Estimation and Applications","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"","keywords":"Copula (linguistics); Estimator; Nonparametric statistics; Statistics; Mathematics; Random variable; Censoring (clinical trials); Applied mathematics; Econometrics","score_opus":0.08501645329480319,"score_gpt":0.3461242279515183,"score_spread":0.26110777465671514,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2052429290","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.20383325,0.007752368,0.78230286,0.0015253203,0.00019694077,0.00009571317,0.00038170756,0.0002591724,0.0036526327],"genre_scores_gemma":[0.8898824,0.0023831283,0.10389971,0.0008971212,0.0006440328,0.0002902602,0.00092627265,0.00009192998,0.0009851038],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9724502,0.0148102995,0.0017728828,0.0058033015,0.0041712024,0.0009921449],"domain_scores_gemma":[0.49059376,0.45809016,0.030467816,0.013778526,0.0054606926,0.0016089737],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05498343,0.001036772,0.0030670303,0.0048271967,0.0011179241,0.0050632884,0.0033955069,0.0033560758,0.0021142734],"category_scores_gemma":[0.25113118,0.0007020326,0.0013773908,0.0044323574,0.0064953524,0.007746878,0.0036937285,0.0040377327,0.0004293022],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012490965,0.0003337261,0.28827316,0.0014890173,0.002058176,0.0027077312,0.0026125542,0.11605333,0.0040666065,0.2854241,0.0021175256,0.293615],"study_design_scores_gemma":[0.00014047806,0.00094304275,0.0940506,0.00082361815,0.00082838157,0.004299268,0.00091869856,0.37794137,0.007361419,0.5008485,0.011447322,0.00039736103],"about_ca_topic_score_codex":0.0012384693,"about_ca_topic_score_gemma":0.0005005796,"teacher_disagreement_score":0.05498343,"about_ca_system_score_codex":0.0012236532,"about_ca_system_score_gemma":0.0013612598,"threshold_uncertainty_score":0.29078358},"labels":[],"label_agreement":null},{"id":"W2054140640","doi":"10.1002/sim.2210","title":"An appraisal of methods for the analysis of longitudinal categorical data with MAR drop‐outs","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Guelph; University of Waterloo","funders":"","keywords":"Categorical variable; Statistics; Generalized estimating equation; Marginal model; Mathematics; Gee; Missing data; Econometrics; Contingency table; Applied mathematics; Regression analysis","score_opus":0.1568631137230554,"score_gpt":0.5382222258091557,"score_spread":0.3813591120861003,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2054140640","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0016247865,0.23912911,0.7383711,0.013313429,0.0022294384,0.0006227828,0.00027620196,0.0004834416,0.003949711],"genre_scores_gemma":[0.015954943,0.16755633,0.80530214,0.0031256187,0.0025343855,0.0025001133,0.00033299765,0.0004185886,0.0022747864],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8113067,0.13863188,0.009584349,0.0036147728,0.036281757,0.00058048434],"domain_scores_gemma":[0.478027,0.46318272,0.009003396,0.009696177,0.039073005,0.001017629],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.19024833,0.0018639925,0.004176542,0.01285615,0.0014652733,0.005067888,0.006215861,0.00357869,0.0032354235],"category_scores_gemma":[0.3559669,0.0020744295,0.0044334116,0.011297375,0.005302486,0.0047898884,0.0026121482,0.0069105695,0.0015990401],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00030873303,0.00014748753,0.005982672,0.011311498,0.00091404206,0.00027738832,0.0012641671,0.010864082,0.0006845898,0.1253437,0.027610682,0.8152909],"study_design_scores_gemma":[0.0005354531,0.0012793024,0.029284712,0.030170426,0.00090054295,0.0035505472,0.0018817399,0.11772027,0.0026133496,0.32857403,0.48247582,0.0010138272],"about_ca_topic_score_codex":0.006763146,"about_ca_topic_score_gemma":0.0077449647,"teacher_disagreement_score":0.19024833,"about_ca_system_score_codex":0.00500204,"about_ca_system_score_gemma":0.007173662,"threshold_uncertainty_score":0.998568},"labels":[],"label_agreement":null},{"id":"W2054344391","doi":"10.1002/sim.3504","title":"Age‐ and size‐related reference ranges: A case study of spirometry through childhood and adulthood","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Hydrology and Drought Analysis","field":"Environmental Science","cited_by":165,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hospital for Sick Children","funders":"Medical Research Council; National Institute for Health and Care Research","keywords":"Skewness; Kurtosis; Statistics; Spirometry; Mathematics; Generalized additive model; Linear regression; Medicine; Internal medicine","score_opus":0.017518818533129855,"score_gpt":0.28501926695840435,"score_spread":0.2675004484252745,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2054344391","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9877759,0.0027371936,0.005187242,0.00054766453,0.00002289825,0.00009779348,0.00078325794,0.000032862503,0.0028151968],"genre_scores_gemma":[0.99033725,0.0017719701,0.0062635057,0.00010995536,0.000026911,0.00008315886,0.00049311924,0.000026715737,0.00088740943],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9947208,0.0027457273,0.0003761871,0.0006585109,0.00091298413,0.00058585324],"domain_scores_gemma":[0.988552,0.006541682,0.00167753,0.0012009409,0.0015904012,0.00043744443],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008070083,0.00047077614,0.0009312149,0.0019887735,0.001100025,0.0012208498,0.0010930919,0.0017348643,0.0011669842],"category_scores_gemma":[0.017790876,0.00034041327,0.001288747,0.0038780002,0.0008225286,0.001326453,0.0019924727,0.0014103592,0.00027107578],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00029464584,0.00028452495,0.9408535,0.00018862513,0.00013975876,0.009860576,0.007228293,0.0017501172,0.00029834485,0.002924887,0.0012907534,0.0348861],"study_design_scores_gemma":[0.00002974939,0.00071495085,0.94263303,0.00038501323,0.00021538328,0.018978132,0.015097271,0.010598564,0.00092988845,0.0015472164,0.008782686,0.0000881626],"about_ca_topic_score_codex":0.051076483,"about_ca_topic_score_gemma":0.047475584,"teacher_disagreement_score":0.051076483,"about_ca_system_score_codex":0.0014468258,"about_ca_system_score_gemma":0.0011653551,"threshold_uncertainty_score":0.10155833},"labels":[],"label_agreement":null},{"id":"W2054584201","doi":"10.1002/sim.3547","title":"Bayesian multivariate disease mapping and ecological regression with errors in covariates: Bayesian estimation of DALYs and ‘preventable’ DALYs","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Child and Family Research Institute; University of British Columbia","funders":"Canadian Institutes of Health Research; University of British Columbia; Ministry of Health, British Columbia","keywords":"Multivariate statistics; Covariate; Bayesian probability; Estimation; Disease; Bayesian inference; Multivariate analysis; Regression analysis; Statistics; Geography; Computer science; Environmental health; Medicine; Machine learning; Engineering; Mathematics","score_opus":0.03902389800503935,"score_gpt":0.3799327395614607,"score_spread":0.3409088415564213,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2054584201","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.004925994,0.0006148239,0.9925861,0.00065389957,0.000030782146,0.000024409694,0.000115751805,0.000054993827,0.0009932905],"genre_scores_gemma":[0.38633758,0.0049283104,0.60050946,0.0006755209,0.00045466088,0.0006380841,0.00080849644,0.00015002422,0.0054978356],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99292433,0.0049750363,0.00020013595,0.0008022688,0.00085067877,0.00024753305],"domain_scores_gemma":[0.98236084,0.014207069,0.0015623783,0.0010389138,0.0006518399,0.00017890261],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014243052,0.0011320258,0.0019303464,0.0022489864,0.00058065815,0.0020276392,0.002406028,0.001515533,0.0025070785],"category_scores_gemma":[0.05365126,0.0008126542,0.0017453241,0.0031817516,0.002140636,0.0031516908,0.0026513797,0.0027139708,0.0003497594],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000056753124,0.000050040628,0.008193427,0.00015840275,0.00023431024,0.00012258872,0.00025814297,0.31950903,0.00022949708,0.58310705,0.0018694351,0.08621125],"study_design_scores_gemma":[0.000023924098,0.00002887928,0.002251374,0.00006863615,0.0000530918,0.00010709825,0.00004362076,0.4605391,0.00013691203,0.53213674,0.0045723845,0.000038276747],"about_ca_topic_score_codex":0.013851262,"about_ca_topic_score_gemma":0.009487349,"teacher_disagreement_score":0.014243052,"about_ca_system_score_codex":0.0017626036,"about_ca_system_score_gemma":0.0021580532,"threshold_uncertainty_score":0.07532537},"labels":[],"label_agreement":null},{"id":"W2054826081","doi":"10.1002/sim.3036","title":"Use of instrumental variables in the analysis of generalized linear models in the presence of unmeasured confounding with applications to epidemiological research","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":72,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; University of Toronto; St. Paul's Hospital; University of British Columbia","funders":"Canadian Institutes of Health Research; Michael Smith Health Research BC","keywords":"Confounding; Generalized linear model; Instrumental variable; Epidemiology; Poisson distribution; Econometrics; Poisson regression; Linear model; Statistics; Logistic regression; Medicine; Computer science; Mathematics; Population; Environmental health; Internal medicine","score_opus":0.4857623800456914,"score_gpt":0.5493356677123323,"score_spread":0.06357328766664094,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2054826081","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.001464764,0.0007901925,0.996307,0.0006847647,0.00007771339,0.00003446712,0.000050335657,0.00008533543,0.00050542643],"genre_scores_gemma":[0.13855723,0.004330619,0.8519346,0.00077839143,0.0006756864,0.0008804359,0.00021341903,0.00016454668,0.0024650376],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96728045,0.028230477,0.0006778575,0.0012428911,0.0021634852,0.0004047919],"domain_scores_gemma":[0.8836389,0.10477941,0.0045388825,0.005167772,0.0015495629,0.00032545766],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.032968163,0.0018509483,0.0027181827,0.0033160297,0.0009385675,0.0024055543,0.0033833107,0.0026236635,0.0029663383],"category_scores_gemma":[0.11961974,0.001051728,0.0029773316,0.005087154,0.0042048227,0.002869256,0.004063635,0.0053109056,0.0005080233],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000631717,0.000047238667,0.0030051963,0.00035888734,0.000667702,0.00047771513,0.00037955798,0.07703005,0.00037016932,0.8695071,0.0016334365,0.046459645],"study_design_scores_gemma":[0.00006735956,0.000068988316,0.0006670454,0.0001454793,0.00014729584,0.00013456611,0.00006245385,0.2129383,0.00038351983,0.7794422,0.0058880732,0.000054804208],"about_ca_topic_score_codex":0.0040628784,"about_ca_topic_score_gemma":0.0030921118,"teacher_disagreement_score":0.032968163,"about_ca_system_score_codex":0.001514958,"about_ca_system_score_gemma":0.0038841257,"threshold_uncertainty_score":0.17435437},"labels":[],"label_agreement":null},{"id":"W2054869597","doi":"10.1002/sim.3808","title":"Threshold regression for survival data with time‐varying covariates","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":39,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"National Eye Institute; National Heart, Lung, and Blood Institute; National Institute for Occupational Safety and Health; Natural Sciences and Engineering Research Council of Canada; National Institutes of Health","keywords":"Covariate; Proportional hazards model; Statistics; Econometrics; Markov chain; Regression analysis; Event (particle physics); Computer science; Inference; Accelerated failure time model; Regression; Markov model; Mathematics; Artificial intelligence","score_opus":0.13762832443698783,"score_gpt":0.4428136105549579,"score_spread":0.30518528611797,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2054869597","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0048453906,0.00078622537,0.9922713,0.0004543867,0.00007151929,0.000077041346,0.00036390786,0.0003499691,0.0007803641],"genre_scores_gemma":[0.4070135,0.0055627776,0.5685623,0.0010501158,0.0007002174,0.0015790496,0.003821149,0.00072492065,0.010985982],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9884333,0.0072297454,0.0006139027,0.0018921116,0.0013694032,0.0004614942],"domain_scores_gemma":[0.94710386,0.04394305,0.0029980347,0.0041045025,0.0014683901,0.00038206554],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.023785343,0.00092713546,0.0019382298,0.0024275456,0.00071500684,0.0019900363,0.003173111,0.0023223856,0.0055470592],"category_scores_gemma":[0.083547816,0.0007776538,0.0024025182,0.0049922955,0.002057445,0.0037355851,0.0023488654,0.005654978,0.0013783908],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000229595,0.00009956588,0.010886824,0.000714321,0.00037708413,0.0006869338,0.00057446945,0.18002836,0.0026547927,0.6803948,0.005724277,0.117629014],"study_design_scores_gemma":[0.000046559533,0.00011174644,0.0025417407,0.00011944604,0.00006130907,0.0002817427,0.00008165112,0.6639,0.00071849045,0.32476094,0.0073246364,0.000051779378],"about_ca_topic_score_codex":0.0064801783,"about_ca_topic_score_gemma":0.0033863236,"teacher_disagreement_score":0.023785343,"about_ca_system_score_codex":0.0017747426,"about_ca_system_score_gemma":0.0021857936,"threshold_uncertainty_score":0.12579042},"labels":[],"label_agreement":null},{"id":"W2055260064","doi":"10.1002/sim.1442","title":"A comparison of several regression models for analysing cost of CABG surgery","year":2003,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":143,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Health Sciences Centre; Institute for Clinical Evaluative Sciences; University of Calgary; Sunnybrook Health Science Centre; Women's College Hospital; University of Toronto","funders":"","keywords":"Linear regression; Statistics; Poisson regression; Generalized linear model; Regression analysis; Regression; Proportional hazards model; Medicine; Linear model; Negative binomial distribution; Mathematics; Poisson distribution; Econometrics; Population","score_opus":0.5450433319250734,"score_gpt":0.5321256663837163,"score_spread":0.012917665541357093,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2055260064","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.36564437,0.007835736,0.60969406,0.0041324166,0.0004938952,0.0014636911,0.0031263325,0.0018270683,0.005782469],"genre_scores_gemma":[0.72544086,0.005480292,0.25807378,0.00059357693,0.00028441547,0.0020513793,0.004177182,0.0005627392,0.0033357514],"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","domain_scores_codex":[0.9443202,0.04833979,0.0013567863,0.0021383418,0.002748752,0.0010960851],"domain_scores_gemma":[0.77472496,0.20682125,0.005804945,0.0054410333,0.006478145,0.0007297062],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07162116,0.003231179,0.0030320396,0.00669043,0.00084232964,0.0027655186,0.004741076,0.0020093226,0.0023639032],"category_scores_gemma":[0.14075454,0.00148015,0.00806969,0.006737764,0.0010030422,0.0040512662,0.002209914,0.003708849,0.00085318304],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004755767,0.00089665316,0.12356052,0.0011711929,0.0068387655,0.00029198037,0.0008241464,0.678712,0.00047432474,0.028296221,0.0064666006,0.1477119],"study_design_scores_gemma":[0.00033293082,0.0010971347,0.019743223,0.00023338836,0.00084044173,0.00014875187,0.00046145456,0.9630321,0.00023135603,0.01211113,0.00161431,0.00015377338],"about_ca_topic_score_codex":0.032799367,"about_ca_topic_score_gemma":0.021782417,"teacher_disagreement_score":0.07162116,"about_ca_system_score_codex":0.0038546016,"about_ca_system_score_gemma":0.0036567473,"threshold_uncertainty_score":0.3787734},"labels":[],"label_agreement":null},{"id":"W2055353237","doi":"10.1002/sim.3092","title":"Comparison of algorithms to generate event times conditional on time‐dependent covariates","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":69,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre; McGill University","funders":"","keywords":"Covariate; Computer science; Toolbox; Event (particle physics); Algorithm; Stability (learning theory); Calibration; Flexibility (engineering); Binomial (polynomial); Statistics; Data mining; Mathematics; Machine learning","score_opus":0.37609532417178737,"score_gpt":0.6061969173828708,"score_spread":0.23010159321108348,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2055353237","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.072025485,0.0008809632,0.92204857,0.0006434682,0.000113028946,0.0005557692,0.00033173314,0.001391412,0.0020095867],"genre_scores_gemma":[0.26020595,0.0009365401,0.73446363,0.00021278039,0.000075328244,0.0016819222,0.0011510558,0.00036352256,0.0009092168],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9910323,0.006278892,0.0007617164,0.00074512296,0.00096689805,0.00021506393],"domain_scores_gemma":[0.8173009,0.16861553,0.0030035435,0.005571477,0.0046226867,0.0008859254],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.025374064,0.0010222587,0.0010586522,0.0026513732,0.00044991073,0.0012112228,0.0021068046,0.0016053491,0.002860726],"category_scores_gemma":[0.09699916,0.000531462,0.0010515578,0.0017781893,0.0007586035,0.0024622385,0.001545868,0.0021005485,0.0008130082],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0037081018,0.00083463645,0.0131920995,0.00047302333,0.0005590545,0.00007602506,0.00042104255,0.36095947,0.0017159126,0.05290479,0.004162806,0.5609931],"study_design_scores_gemma":[0.00096586253,0.00060291274,0.0024878597,0.00006113018,0.00008513828,0.00017931996,0.00009036212,0.9579288,0.0024537428,0.033098612,0.0019945088,0.000051838862],"about_ca_topic_score_codex":0.0021141798,"about_ca_topic_score_gemma":0.0018022122,"teacher_disagreement_score":0.025374064,"about_ca_system_score_codex":0.0012150881,"about_ca_system_score_gemma":0.0032332775,"threshold_uncertainty_score":0.13419253},"labels":[],"label_agreement":null},{"id":"W2055628766","doi":"10.1002/sim.4208","title":"Flexible modeling of the effects of continuous prognostic factors in relative survival","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Canadian Institutes of Health Research","keywords":"Covariate; Proportional hazards model; Statistics; Hazard ratio; Survival analysis; Relative survival; Hazard; Mathematics; Linear regression; Econometrics; Relative risk; Parametric statistics; Regression analysis; Medicine; Confidence interval; Internal medicine; Cancer; Biology; Cancer registry","score_opus":0.12102041667364608,"score_gpt":0.3824141029771552,"score_spread":0.2613936863035091,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2055628766","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02248825,0.00034322822,0.97600704,0.00028379983,0.00003669735,0.000045220633,0.00019603742,0.00016234833,0.00043736878],"genre_scores_gemma":[0.7963279,0.001040539,0.19596899,0.00029094503,0.00016502387,0.000625274,0.000566734,0.0001601235,0.004854572],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98878086,0.008375418,0.00026509966,0.001075846,0.00092527864,0.0005775397],"domain_scores_gemma":[0.9734874,0.021255024,0.002049489,0.0023291335,0.0005752057,0.00030368505],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02371284,0.0013509323,0.001607016,0.0013874102,0.00051265646,0.0018988622,0.003715176,0.0015169666,0.0023182142],"category_scores_gemma":[0.044927694,0.00082048815,0.0030601812,0.0022796157,0.002766868,0.0022232377,0.0024467818,0.0033015076,0.0004969467],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00027288264,0.0000733259,0.0073360736,0.00012492095,0.0002233101,0.00024789985,0.00039253072,0.82665974,0.00096287864,0.12382736,0.0006201135,0.03925899],"study_design_scores_gemma":[0.00006692089,0.00023457872,0.0024401043,0.000041642616,0.00012604792,0.00011686648,0.000037526544,0.89317477,0.00045721268,0.10190902,0.0013392605,0.000056150046],"about_ca_topic_score_codex":0.0060555846,"about_ca_topic_score_gemma":0.004697302,"teacher_disagreement_score":0.02371284,"about_ca_system_score_codex":0.0014296174,"about_ca_system_score_gemma":0.0016290366,"threshold_uncertainty_score":0.12540698},"labels":[],"label_agreement":null},{"id":"W2055909840","doi":"10.1002/sim.4343","title":"Comparison of alternative models for linking drug exposure with adverse effects","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":80,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Université de Montréal; McGill University Health Centre","funders":"Canadian Institutes of Health Research; Innovative Research Group Project of the National Natural Science Foundation of China","keywords":"A priori and a posteriori; Model selection; Metric (unit); Computer science; Information Criteria; Statistics; Econometrics; Mathematics; Data mining; Machine learning","score_opus":0.5145667783424172,"score_gpt":0.5664642533495956,"score_spread":0.05189747500717845,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2055909840","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.42369527,0.0020322646,0.5663995,0.0019250584,0.00019911019,0.00048277536,0.001866231,0.00063187355,0.0027679089],"genre_scores_gemma":[0.90427095,0.0008445728,0.08973688,0.00028121262,0.00009694232,0.00070085656,0.0019601712,0.000119784294,0.0019886238],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9852325,0.011707212,0.00064782694,0.0012306275,0.0006702065,0.00051171926],"domain_scores_gemma":[0.88312274,0.10485306,0.0042977734,0.0041622394,0.0026604552,0.0009036615],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.034812618,0.0018249082,0.0020428493,0.003723574,0.0007530054,0.0023104795,0.0036592605,0.0026973728,0.0028328367],"category_scores_gemma":[0.08255429,0.0010025463,0.0059052715,0.002065252,0.0013362768,0.0028275524,0.0019081407,0.003414633,0.00043613062],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00095552305,0.00017786786,0.01545465,0.00020768182,0.00077955856,0.00019371352,0.00034992892,0.9272239,0.0005248218,0.033747207,0.0007314637,0.019653711],"study_design_scores_gemma":[0.00008088675,0.00023258454,0.0029502164,0.00004213914,0.00014824451,0.00005593486,0.00006213713,0.9791612,0.00015201062,0.016662605,0.00040303686,0.000049067636],"about_ca_topic_score_codex":0.014877138,"about_ca_topic_score_gemma":0.009012151,"teacher_disagreement_score":0.9651874,"about_ca_system_score_codex":0.002259514,"about_ca_system_score_gemma":0.0021570865,"threshold_uncertainty_score":0.18410891},"labels":[],"label_agreement":null},{"id":"W2056285559","doi":"10.1002/sim.5402","title":"Improving the reporting of randomised trials: the CONSORT Statement and beyond","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":116,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"Medical Research Council","keywords":"Consolidated Standards of Reporting Trials; Context (archaeology); Alternative medicine; Systematic review; Clinical trial; Medicine; Medical research; Reliability (semiconductor); MEDLINE; Family medicine; Medical education; Political science","score_opus":0.7374228889944674,"score_gpt":0.5840386606266567,"score_spread":0.15338422836781074,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2056285559","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0021680288,0.048793845,0.71421206,0.08712387,0.070754275,0.064437225,0.0030213199,0.0043198946,0.005169409],"genre_scores_gemma":[0.023682963,0.020958548,0.69525874,0.024276407,0.014362765,0.21685398,0.0015642424,0.0015554583,0.0014869226],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.06856226,0.77540153,0.12479463,0.00898558,0.020714225,0.0015417425],"domain_scores_gemma":[0.030486021,0.7823531,0.07286438,0.064605966,0.04651207,0.0031784927],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.81652665,0.0071777143,0.024787702,0.021633431,0.0048544705,0.02114514,0.0114353765,0.030074442,0.0056643435],"category_scores_gemma":[0.90280336,0.009006935,0.018641919,0.034667317,0.02866593,0.017651739,0.010845707,0.045803502,0.0062558684],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0072749476,0.00044075324,0.005226774,0.1462995,0.020770622,0.0013534168,0.016647734,0.008242872,0.0022444976,0.20184542,0.2438297,0.34582376],"study_design_scores_gemma":[0.008314549,0.0013267467,0.0067120097,0.15237848,0.009907036,0.0019157735,0.00201179,0.03212276,0.004338572,0.4275201,0.35156113,0.00189107],"about_ca_topic_score_codex":0.0025335755,"about_ca_topic_score_gemma":0.0021879226,"teacher_disagreement_score":0.18347335,"about_ca_system_score_codex":0.016524116,"about_ca_system_score_gemma":0.041137155,"threshold_uncertainty_score":0.22625536},"labels":[],"label_agreement":null},{"id":"W2056730537","doi":"10.1002/sim.4305","title":"Monitoring binary outcomes using risk‐adjusted charts: a comparative study","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; University of Windsor; University of Alberta","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"CUSUM; Computer science; Control chart; Statistics; Statistical process control; Context (archaeology); Type I and type II errors; Shewhart individuals control chart; Binary number; Chart; Mathematics; EWMA chart; Process (computing)","score_opus":0.4789125639840743,"score_gpt":0.5315657088707874,"score_spread":0.05265314488671308,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2056730537","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.85710794,0.0690369,0.059191108,0.0009615227,0.0007752438,0.0011072179,0.002307561,0.0004042469,0.00910826],"genre_scores_gemma":[0.9687762,0.008686149,0.019706968,0.0001816048,0.00037045684,0.0005014078,0.0012929129,0.00007086419,0.00041346272],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.96755713,0.02513263,0.0015668205,0.001062439,0.004322641,0.00035836943],"domain_scores_gemma":[0.84320956,0.1287306,0.010364198,0.007108469,0.009552755,0.001034397],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.038699377,0.0008138601,0.0012446768,0.0042446395,0.0002928899,0.0018891179,0.0014019796,0.0012446744,0.0024272292],"category_scores_gemma":[0.11878402,0.0002624234,0.002264778,0.0034493299,0.0006626605,0.0021588923,0.0012116117,0.0009589137,0.00028930756],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.04534975,0.0027049903,0.20933041,0.0047940467,0.009895505,0.00034624926,0.0016452065,0.026581384,0.001187221,0.015426475,0.006082387,0.67665637],"study_design_scores_gemma":[0.008069076,0.08097003,0.543354,0.004399165,0.01856848,0.002633511,0.0042958157,0.26833567,0.0069526015,0.020389047,0.04075061,0.0012821232],"about_ca_topic_score_codex":0.0020889193,"about_ca_topic_score_gemma":0.00091540575,"teacher_disagreement_score":0.038699377,"about_ca_system_score_codex":0.0011425149,"about_ca_system_score_gemma":0.00084523414,"threshold_uncertainty_score":0.20466429},"labels":[],"label_agreement":null},{"id":"W2057909673","doi":"10.1002/sim.3758","title":"Estimation of prediction error for survival models","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":43,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Alberta Health Services; University of Waterloo","funders":"","keywords":"Estimator; Statistics; Confidence interval; Point estimation; Regression analysis; Regression; Prediction interval; Computer science; Standard error; Econometrics; Estimation; Mean squared error; Model selection; Mathematics","score_opus":0.15834846963960886,"score_gpt":0.45548541308610574,"score_spread":0.2971369434464969,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2057909673","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005352147,0.0006768222,0.9930875,0.0002470247,0.0000375277,0.000030599265,0.0000843982,0.00014238416,0.0003415765],"genre_scores_gemma":[0.48864534,0.0028521977,0.5031914,0.0003893944,0.0002903043,0.0006887639,0.0013730115,0.0002767993,0.0022927446],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97682035,0.017454255,0.0009124323,0.0014500808,0.0030200453,0.00034281716],"domain_scores_gemma":[0.8239791,0.15862516,0.0058685266,0.0068712537,0.0042836694,0.00037234952],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.046529353,0.0013082961,0.001835949,0.0028493174,0.00046033572,0.0019322002,0.0022045486,0.0021138247,0.0014010039],"category_scores_gemma":[0.21237664,0.0006218991,0.0015193773,0.0020674379,0.0017942394,0.0027817925,0.0027531048,0.003778573,0.00046841288],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023429164,0.00008481056,0.014670683,0.00050320075,0.0007317315,0.00019290694,0.00039729275,0.67657334,0.00087787013,0.13687585,0.002818418,0.16603959],"study_design_scores_gemma":[0.00002430371,0.000094109266,0.0022665262,0.00017818176,0.00006885588,0.00011365588,0.000051888725,0.87762845,0.00088653376,0.117017016,0.0016238664,0.00004655185],"about_ca_topic_score_codex":0.0030046087,"about_ca_topic_score_gemma":0.0011572322,"teacher_disagreement_score":0.046529353,"about_ca_system_score_codex":0.0011440886,"about_ca_system_score_gemma":0.0015103107,"threshold_uncertainty_score":0.24607366},"labels":[],"label_agreement":null},{"id":"W2058899694","doi":"10.1002/sim.4011","title":"Nonparametric sequential monitoring of longitudinal trials","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"University of Alberta","keywords":"Nonparametric statistics; Computer science; Simple (philosophy); Applied mathematics; Monte Carlo method; Statistics; Longitudinal data; Mathematics; Algorithm; Data mining","score_opus":0.6279650322204489,"score_gpt":0.6259249707736588,"score_spread":0.0020400614467900313,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2058899694","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0147237005,0.0018693155,0.98033273,0.0005926472,0.00017215464,0.00030997742,0.00019073201,0.00029757945,0.0015111588],"genre_scores_gemma":[0.5746333,0.0023485473,0.41750386,0.00039543537,0.00064539607,0.002338026,0.00046426945,0.00009091493,0.001580194],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.922677,0.06754853,0.0016558389,0.003082418,0.0044716224,0.0005647148],"domain_scores_gemma":[0.7928067,0.17332783,0.015877191,0.01219194,0.0046705585,0.0011257821],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06531986,0.0009230615,0.0023401703,0.0016218647,0.0005021713,0.0015508101,0.0020684807,0.001211655,0.0021005634],"category_scores_gemma":[0.23959424,0.000525241,0.0009911129,0.0022120937,0.0021217917,0.002638776,0.0019288844,0.0020236685,0.0003251158],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020815954,0.00023438828,0.014424561,0.00130495,0.0009580098,0.00036571934,0.0005709729,0.13253878,0.0014622052,0.27854276,0.0034988092,0.5640173],"study_design_scores_gemma":[0.00072890526,0.0020455192,0.0072383527,0.00028899673,0.00034986215,0.00045761082,0.00009475616,0.56685305,0.0022396443,0.40850246,0.011094151,0.00010673064],"about_ca_topic_score_codex":0.00088298577,"about_ca_topic_score_gemma":0.0005709398,"teacher_disagreement_score":0.06531986,"about_ca_system_score_codex":0.0010488309,"about_ca_system_score_gemma":0.00249241,"threshold_uncertainty_score":0.34544855},"labels":[],"label_agreement":null},{"id":"W2059447605","doi":"10.1002/sim.1973","title":"Concomitant information in bioassay and semi-parametric estimation","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; St. Joseph's Hospital; St. Joseph’s Healthcare Hamilton; University of Guelph","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Estimator; Interval estimation; Parametric statistics; Point estimation; Bioassay; Smoothing; Semiparametric model; Statistics; Confidence interval; Computer science; Estimation; Mathematics; Concomitant; Econometrics; Applied mathematics; Engineering","score_opus":0.04963630052769486,"score_gpt":0.38964238202597695,"score_spread":0.3400060814982821,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2059447605","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0029396478,0.00033545616,0.9960329,0.000088317865,0.000021139238,0.000023043067,0.000060139333,0.00006611864,0.00043335144],"genre_scores_gemma":[0.33847603,0.0017141325,0.65566707,0.00035816806,0.00042100073,0.00054410665,0.00058518565,0.00012342625,0.0021108454],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9783478,0.012443046,0.0011132344,0.002701257,0.005032021,0.00036263684],"domain_scores_gemma":[0.8999962,0.07911615,0.00576899,0.012635295,0.0020802142,0.0004031591],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02322169,0.0015106343,0.002972441,0.002734663,0.00065377756,0.003312698,0.00363321,0.002679842,0.0020777134],"category_scores_gemma":[0.10530924,0.0011803056,0.0025246032,0.0026391088,0.0044006705,0.00515862,0.0043046395,0.0050137606,0.0007462448],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005002182,0.0001999346,0.0053807157,0.0008883113,0.00064948,0.0007132535,0.00050641975,0.15108427,0.013309646,0.6158505,0.0008751609,0.21004215],"study_design_scores_gemma":[0.000050565683,0.00052493083,0.0027454933,0.00014007885,0.00022534856,0.00047053018,0.000050391845,0.37974563,0.010762428,0.598639,0.0064665433,0.00017902933],"about_ca_topic_score_codex":0.00072110153,"about_ca_topic_score_gemma":0.0005457833,"teacher_disagreement_score":0.02322169,"about_ca_system_score_codex":0.0010861679,"about_ca_system_score_gemma":0.0015266946,"threshold_uncertainty_score":0.12280947},"labels":[],"label_agreement":null},{"id":"W2059629136","doi":"10.1002/sim.3766","title":"Optimal two‐stage reliability studies","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Hemodynamic Monitoring and Therapy","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Intraclass correlation; Reliability (semiconductor); Measure (data warehouse); Stage (stratigraphy); Mathematics; Statistics; Limit (mathematics); Correlation; Correlation coefficient; Computer science; Reproducibility; Data mining; Power (physics); Physics; Biology","score_opus":0.03542734917062795,"score_gpt":0.41329293431337893,"score_spread":0.377865585142751,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2059629136","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09724526,0.0044016778,0.8920546,0.0008591941,0.00016382938,0.0014066661,0.0006538761,0.00035362973,0.002861141],"genre_scores_gemma":[0.57334304,0.0013380309,0.41954184,0.00038531297,0.00028464626,0.002543757,0.001058021,0.00016199055,0.0013434535],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95285344,0.035371326,0.0022459466,0.0045894952,0.0036979674,0.0012417716],"domain_scores_gemma":[0.7948806,0.15998845,0.011958361,0.019248942,0.011876673,0.0020469546],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06101019,0.0021780033,0.0035306115,0.0018275218,0.0007970903,0.0020608911,0.0032833826,0.002388928,0.003342613],"category_scores_gemma":[0.17522275,0.0020652942,0.0031647226,0.0015993027,0.0021217018,0.003916114,0.004272259,0.0030796619,0.0004832978],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.016643094,0.0015681128,0.072978415,0.0025419118,0.0042645284,0.001036777,0.0018492573,0.32768005,0.011702099,0.10357976,0.009384258,0.44677177],"study_design_scores_gemma":[0.0017342847,0.011254408,0.038206626,0.0006368387,0.0016535155,0.0010927898,0.0005175764,0.78358626,0.009655083,0.14104703,0.010248986,0.00036670102],"about_ca_topic_score_codex":0.0013561222,"about_ca_topic_score_gemma":0.0010976321,"teacher_disagreement_score":0.9389898,"about_ca_system_score_codex":0.001514022,"about_ca_system_score_gemma":0.0034230913,"threshold_uncertainty_score":0.32265663},"labels":[],"label_agreement":null},{"id":"W2059794838","doi":"10.1002/(sici)1097-0258(20000315)19:5<743::aid-sim338>3.0.co;2-f","title":"Retrospective estimation of the birth prevalence for delayed onset disorders: application to cystic fibrosis in Nova Scotia","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Cystic Fibrosis Research Advances","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto; Dalhousie University; University of British Columbia","funders":"Cystic Fibrosis Foundation","keywords":"Nova scotia; Medicine; Cystic fibrosis; Pediatrics; Maximum likelihood; Estimation; Demography; Statistics; Internal medicine; Geography","score_opus":0.01023350576769561,"score_gpt":0.34114217346828263,"score_spread":0.330908667700587,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2059794838","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.96505916,0.0008572004,0.026301889,0.00031009378,0.000027292117,0.00014629979,0.006371291,0.00011461681,0.00081223535],"genre_scores_gemma":[0.98837775,0.00035015255,0.007733418,0.00003401672,0.000011716109,0.00006915706,0.0030090234,0.000012183871,0.00040259652],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9974867,0.0015084547,0.00024545292,0.00038545238,0.00025570355,0.00011826369],"domain_scores_gemma":[0.98085016,0.0116271945,0.0039056505,0.0018286117,0.001369456,0.00041889804],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0061969473,0.0003465064,0.0005725886,0.0020844114,0.00040083914,0.0006798238,0.0008825428,0.00045266526,0.0008427515],"category_scores_gemma":[0.034984462,0.00054130686,0.000490175,0.002042649,0.00059302524,0.0002425735,0.00084465975,0.00051725324,0.00020570279],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013949318,0.000009147378,0.98424107,0.00004256923,0.00014598278,0.00021745855,0.00017554862,0.007768986,0.00012184576,0.0003561816,0.00043051702,0.006351232],"study_design_scores_gemma":[0.000027779934,0.000091288646,0.9563395,0.00006248018,0.000085436695,0.00047858348,0.00030334544,0.040990513,0.00017388891,0.00049704756,0.00092935126,0.00002090247],"about_ca_topic_score_codex":0.35069478,"about_ca_topic_score_gemma":0.17709765,"teacher_disagreement_score":0.6493052,"about_ca_system_score_codex":0.0015459468,"about_ca_system_score_gemma":0.0015145837,"threshold_uncertainty_score":0.697307},"labels":[],"label_agreement":null},{"id":"W2060230329","doi":"10.1002/sim.2197","title":"Bias-corrected maximum likelihood estimator of the intraclass correlation parameter for binary data","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Distribution Estimation and Applications","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Windsor","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Statistics; Intraclass correlation; Mathematics; Estimator; Maximum likelihood; Binary data; Restricted maximum likelihood; Estimation theory; Score test; Binary number","score_opus":0.12125405863141317,"score_gpt":0.4067263800309504,"score_spread":0.28547232139953727,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2060230329","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007398211,0.00034192263,0.99165136,0.00011716452,0.000020207654,0.000036897596,0.00008020739,0.00012483963,0.00022924939],"genre_scores_gemma":[0.18255313,0.0005036627,0.8146275,0.00023522844,0.00006956351,0.0003304676,0.00060125,0.000108940396,0.0009702008],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99129057,0.0064237886,0.00037234576,0.00073445763,0.001012568,0.0001662228],"domain_scores_gemma":[0.9514697,0.03951152,0.0023876468,0.0041526565,0.0023018275,0.00017661462],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.019432096,0.00049720943,0.0015014232,0.0016596114,0.00039125362,0.0009928666,0.0014485747,0.00131843,0.0017180404],"category_scores_gemma":[0.09632544,0.00040572268,0.000783413,0.0017693797,0.0009960586,0.0017705668,0.0015161406,0.0013156865,0.00055974076],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00062902394,0.00018186461,0.04205882,0.0010109242,0.000988595,0.0003696451,0.00062201393,0.22335689,0.009298955,0.13853915,0.005146696,0.57779735],"study_design_scores_gemma":[0.00014236987,0.00022761493,0.014021613,0.00020344321,0.00015124195,0.0007161426,0.000097087264,0.8010737,0.0062911743,0.17113099,0.0058488357,0.00009586296],"about_ca_topic_score_codex":0.0011132365,"about_ca_topic_score_gemma":0.0009659579,"teacher_disagreement_score":0.019432096,"about_ca_system_score_codex":0.00059440854,"about_ca_system_score_gemma":0.0012243405,"threshold_uncertainty_score":0.102768004},"labels":[],"label_agreement":null},{"id":"W2061407842","doi":"10.1002/sim.2618","title":"Conditioning on the propensity score can result in biased estimation of common measures of treatment effect: a Monte Carlo study","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":227,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Propensity score matching; Statistics; Covariate; Hazard ratio; Observational study; Mathematics; Econometrics; Confidence interval","score_opus":0.20389167045157128,"score_gpt":0.42466937770647534,"score_spread":0.22077770725490406,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2061407842","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.13272342,0.0032676551,0.8559098,0.0030779766,0.00012438741,0.00071364007,0.00017974981,0.00018504335,0.0038183236],"genre_scores_gemma":[0.79145455,0.0022342294,0.20181584,0.0011636517,0.00022239605,0.001207999,0.0002587823,0.00013184424,0.0015107505],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9323142,0.060450494,0.0013897805,0.0025868067,0.0025566318,0.0007019682],"domain_scores_gemma":[0.38426343,0.57759416,0.011918635,0.020673998,0.0047915885,0.00075812824],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.108449146,0.0009352491,0.0021855845,0.001543988,0.0012448292,0.0023884326,0.0019820556,0.0023933414,0.0038905775],"category_scores_gemma":[0.35288936,0.0009293345,0.0025632775,0.0020616935,0.003917318,0.003463022,0.0022682585,0.003538779,0.00034756376],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016179839,0.0006440202,0.059713844,0.00070209196,0.0029265974,0.0008825893,0.0012868638,0.2971936,0.0011870158,0.5228561,0.0042911377,0.10669817],"study_design_scores_gemma":[0.00043698415,0.00041436183,0.0072761755,0.00038271348,0.000696883,0.0003848098,0.00013120931,0.72067034,0.0016239857,0.26372036,0.0041526444,0.00010952247],"about_ca_topic_score_codex":0.004296117,"about_ca_topic_score_gemma":0.0030238605,"teacher_disagreement_score":0.89155084,"about_ca_system_score_codex":0.00214579,"about_ca_system_score_gemma":0.0027180535,"threshold_uncertainty_score":0.57354075},"labels":[],"label_agreement":null},{"id":"W2061691003","doi":"10.1002/sim.3200","title":"Impact of mis‐specification of the treatment model on estimates from a marginal structural model","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":66,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Estimator; Marginal structural model; Confounding; Inverse probability; Context (archaeology); Computer science; Econometrics; Specification; Mean squared error; Average treatment effect; Statistics; Sample size determination; Mathematics; Bayesian probability","score_opus":0.1925234186916368,"score_gpt":0.4631170529289025,"score_spread":0.2705936342372657,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2061691003","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.11762835,0.0020413373,0.8726347,0.0031953866,0.00018292085,0.00050875644,0.0004943679,0.00051907235,0.0027950949],"genre_scores_gemma":[0.7427659,0.00069584174,0.2532707,0.0010068264,0.00008775815,0.00065707887,0.0006480678,0.00023831648,0.0006295453],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.82712185,0.15550418,0.0052026506,0.0045085126,0.006659472,0.0010032633],"domain_scores_gemma":[0.19498135,0.7673295,0.009984944,0.022648742,0.0045366175,0.0005188508],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1822475,0.0015455687,0.0030327775,0.0014421042,0.0012548789,0.003018476,0.0027956793,0.002747963,0.0033182339],"category_scores_gemma":[0.6028127,0.0015041226,0.002557628,0.0020957883,0.0032179644,0.0042158724,0.003305566,0.0051173284,0.0005364305],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0029681935,0.0002305159,0.07197668,0.0016186078,0.0046192324,0.00082983664,0.0018891912,0.61772215,0.0014646042,0.14891237,0.0036711327,0.14409746],"study_design_scores_gemma":[0.00048007406,0.00057687174,0.011969246,0.0007540555,0.0008926656,0.0006114596,0.00041804468,0.7979594,0.0037121754,0.17895941,0.0034972262,0.00016929762],"about_ca_topic_score_codex":0.010335222,"about_ca_topic_score_gemma":0.0074810805,"teacher_disagreement_score":0.1822475,"about_ca_system_score_codex":0.0029952785,"about_ca_system_score_gemma":0.0038033396,"threshold_uncertainty_score":0.96382844},"labels":[],"label_agreement":null},{"id":"W2063846874","doi":"10.1002/sim.3460","title":"Bayesian propensity score analysis for observational data","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":117,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto; University of British Columbia; Simon Fraser University","funders":"","keywords":"Propensity score matching; Observational study; Statistics; Confounding; Confidence interval; Bayesian probability; Markov chain Monte Carlo; Credible interval; Econometrics; Odds ratio; Outcome (game theory); Medicine; Mathematics","score_opus":0.5909869107336676,"score_gpt":0.490360892353943,"score_spread":0.10062601837972462,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2063846874","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0015200055,0.00043512374,0.9970471,0.0002928573,0.000027095684,0.00011844811,0.000118972945,0.00017011349,0.00027022458],"genre_scores_gemma":[0.1478337,0.002442408,0.84404916,0.00049942214,0.00036135863,0.0020355864,0.0012676907,0.00021610393,0.0012944874],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95127076,0.040541932,0.0015169596,0.0024832103,0.003717237,0.0004698537],"domain_scores_gemma":[0.851211,0.12892343,0.007827906,0.007819886,0.003556833,0.0006609784],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06296529,0.0014678178,0.0032457197,0.0062193205,0.001218935,0.0028857575,0.0032141493,0.002833264,0.004752407],"category_scores_gemma":[0.22713225,0.0014069834,0.0024209493,0.006348886,0.0032413246,0.004603465,0.0039506406,0.004547124,0.001159238],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00027162393,0.000119878736,0.011425284,0.0008230133,0.0011624537,0.0003174461,0.00049282593,0.1894581,0.0005250712,0.5808557,0.0059101786,0.20863853],"study_design_scores_gemma":[0.0001626168,0.0000731428,0.0020126218,0.00015633067,0.00012095112,0.00013319781,0.00004052274,0.42823339,0.00027642245,0.5637193,0.0050130263,0.00005853629],"about_ca_topic_score_codex":0.005453109,"about_ca_topic_score_gemma":0.0029534665,"teacher_disagreement_score":0.06296529,"about_ca_system_score_codex":0.0024328309,"about_ca_system_score_gemma":0.00484258,"threshold_uncertainty_score":0.33299625},"labels":[],"label_agreement":null},{"id":"W2064530264","doi":"10.1002/sim.1060","title":"Longitudinal profiles of health care providers","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Healthcare Policy and Management","field":"Economics, Econometrics and Finance","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto","funders":"Harvard University","keywords":"Mahalanobis distance; Univariate; Health care; Computer science; Profiling (computer programming); Baseline (sea); Longitudinal data; Statistics; Data mining; Multivariate statistics; Econometrics; Actuarial science; Machine learning; Artificial intelligence; Business; Mathematics","score_opus":0.09542272310886057,"score_gpt":0.3368681650707985,"score_spread":0.2414454419619379,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2064530264","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.84839463,0.00077299477,0.110289566,0.0027311225,0.00007707022,0.00061662385,0.024144504,0.00046935325,0.012504151],"genre_scores_gemma":[0.951756,0.0004733924,0.03276918,0.00019944627,0.000050477087,0.0003908219,0.010569717,0.000037197256,0.003753753],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99643385,0.0018086041,0.0003291436,0.0004523891,0.00067646365,0.00029947495],"domain_scores_gemma":[0.974098,0.008344274,0.007083314,0.0032029867,0.005994882,0.001276515],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010029868,0.00036011837,0.000322857,0.0022184795,0.0005734425,0.0012770285,0.0006504194,0.0006808891,0.004359233],"category_scores_gemma":[0.0473853,0.0003220559,0.0004597102,0.0022824893,0.0002531185,0.001975761,0.0013283702,0.0007666091,0.001426046],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00031503008,0.00016566063,0.8879819,0.00007198492,0.0001562364,0.000086219596,0.0012237238,0.019139485,0.0005076307,0.010278283,0.0070058084,0.073068015],"study_design_scores_gemma":[0.000078696845,0.0007558659,0.6696238,0.00017415323,0.00014584807,0.0006153007,0.0028253752,0.27437446,0.0017497222,0.033810694,0.015687393,0.00015884257],"about_ca_topic_score_codex":0.010302537,"about_ca_topic_score_gemma":0.008955834,"teacher_disagreement_score":0.010302537,"about_ca_system_score_codex":0.00097097043,"about_ca_system_score_gemma":0.0011650316,"threshold_uncertainty_score":0.053043664},"labels":[],"label_agreement":null},{"id":"W2064554092","doi":"10.1002/sim.2885","title":"Non‐inferiority trial design for recurrent events","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Sample size determination; Statistics; Poisson distribution; Robustness (evolution); Null hypothesis; Econometrics; Computer science; Event (particle physics); Marginal distribution; Random effects model; Mathematics; Medicine; Random variable; Meta-analysis; Internal medicine","score_opus":0.6391600063244661,"score_gpt":0.6329558923146955,"score_spread":0.0062041140097706116,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2064554092","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0016853483,0.002068099,0.9790247,0.0010609647,0.00052191364,0.014149136,0.00011784235,0.0002559479,0.0011160288],"genre_scores_gemma":[0.025361316,0.0015765433,0.8950529,0.0011610963,0.00048602253,0.07527229,0.00009482427,0.000082870785,0.0009120565],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8423026,0.13904275,0.006857203,0.002729807,0.008518664,0.0005488753],"domain_scores_gemma":[0.8208726,0.15004665,0.012439217,0.01196448,0.0035910236,0.0010860688],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10404967,0.0015671446,0.0031496899,0.0023185937,0.00054612494,0.002109224,0.0030096048,0.0039457013,0.008636054],"category_scores_gemma":[0.21103175,0.0009104043,0.0029446671,0.0015929563,0.0019216695,0.0018086169,0.0019325891,0.0054760245,0.0015083539],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00958786,0.0011922439,0.001714705,0.015271682,0.0023044937,0.00054279546,0.00079966336,0.033555105,0.0030745824,0.31487584,0.0134731475,0.6036079],"study_design_scores_gemma":[0.049633365,0.0195415,0.0019929786,0.0054050605,0.003401799,0.0021515538,0.00015041114,0.23749162,0.00826353,0.5649733,0.10652303,0.0004719155],"about_ca_topic_score_codex":0.000111104,"about_ca_topic_score_gemma":0.00023390555,"teacher_disagreement_score":0.8959503,"about_ca_system_score_codex":0.0010494926,"about_ca_system_score_gemma":0.0028352758,"threshold_uncertainty_score":0.5502738},"labels":[],"label_agreement":null},{"id":"W2064608069","doi":"10.1002/sim.4126","title":"Assessment of joint symmetry in arthritis","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Rheumatoid Arthritis Research and Therapies","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"Medical Research Council; University of Toronto","keywords":"Joint (building); Symmetry (geometry); Psoriatic arthritis; Arthritis; Consistency (knowledge bases); Joint disease; Contingency table; Disease; Medicine; Table (database); Goodness of fit; Econometrics; Mathematics; Computer science; Statistics; Artificial intelligence; Data mining; Pathology; Internal medicine; Alternative medicine; Structural engineering","score_opus":0.04119543443743018,"score_gpt":0.3491132775485651,"score_spread":0.3079178431111349,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2064608069","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1534023,0.0017865634,0.8308844,0.00039466305,0.00023395312,0.0017213898,0.0028424112,0.0011790689,0.0075552724],"genre_scores_gemma":[0.62340784,0.0007017365,0.36915034,0.000088384164,0.00016621382,0.0016441335,0.002350597,0.00025736977,0.0022333553],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9622616,0.021164937,0.003562365,0.0023265972,0.009979719,0.0007047814],"domain_scores_gemma":[0.8252577,0.11578351,0.024998631,0.018795623,0.013569059,0.0015954316],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.037820045,0.0010522212,0.0015741828,0.007994247,0.0006130738,0.0021029646,0.0018823094,0.0010521319,0.0053424602],"category_scores_gemma":[0.16036841,0.00048576135,0.0024521227,0.0044166087,0.0016716144,0.0018081479,0.0033044459,0.0012031295,0.00092595554],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016240546,0.00035642553,0.44731075,0.001004184,0.0021835086,0.00030925882,0.0019369541,0.022556638,0.0034069594,0.021082273,0.005419944,0.49280897],"study_design_scores_gemma":[0.0005475597,0.003577138,0.6854306,0.00057892106,0.0008683394,0.0030271814,0.0019777236,0.2125597,0.0099248355,0.06360942,0.017171381,0.0007272631],"about_ca_topic_score_codex":0.0017513069,"about_ca_topic_score_gemma":0.002418859,"teacher_disagreement_score":0.037820045,"about_ca_system_score_codex":0.0010506359,"about_ca_system_score_gemma":0.0014902921,"threshold_uncertainty_score":0.20001388},"labels":[],"label_agreement":null},{"id":"W2066581144","doi":"10.1002/sim.2546","title":"Bivariate models for co‐aggregation of dichotomous traits in twins","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Bivariate analysis; Twin study; Correlation; Dizygotic twin; Trait; Statistics; Similarity (geometry); Econometrics; Dizygotic twins; Population; Genetic correlation; Monozygotic twin; Psychology; Heritability; Mathematics; Demography; Biology; Computer science; Medicine; Genetics; Genetic variation; Artificial intelligence","score_opus":0.13561655662648225,"score_gpt":0.4075748772475037,"score_spread":0.27195832062102143,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2066581144","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.12290319,0.0017908494,0.86464787,0.0020375496,0.00039078947,0.000621673,0.0023455275,0.0004880384,0.0047745327],"genre_scores_gemma":[0.78744364,0.0031690889,0.17929514,0.00044675617,0.00056634273,0.0035041138,0.0035362644,0.0002093948,0.021829283],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.97543406,0.018199915,0.00096694764,0.0024245135,0.001609945,0.0013646679],"domain_scores_gemma":[0.91889894,0.06363723,0.006030258,0.0065647285,0.0034957682,0.0013730734],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04060368,0.0021568134,0.004255789,0.0038689903,0.0014687938,0.0036373413,0.0060120304,0.0033332186,0.009657776],"category_scores_gemma":[0.07732524,0.0016932946,0.0036856756,0.0046132538,0.0034864428,0.003299774,0.0049484274,0.0046066255,0.0020767644],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007396013,0.0003271084,0.0475716,0.00029931997,0.0016166078,0.0012621561,0.0023627193,0.27810195,0.00053020933,0.6239213,0.0048925406,0.038374793],"study_design_scores_gemma":[0.00010671445,0.00017484145,0.0056819767,0.000121968704,0.00028358572,0.00021805736,0.00022182311,0.8257619,0.00013159229,0.16417705,0.0030248913,0.00009554024],"about_ca_topic_score_codex":0.02010086,"about_ca_topic_score_gemma":0.009987945,"teacher_disagreement_score":0.04060368,"about_ca_system_score_codex":0.0020746961,"about_ca_system_score_gemma":0.0021329161,"threshold_uncertainty_score":0.21473539},"labels":[],"label_agreement":null},{"id":"W2066948187","doi":"10.1002/sim.3953","title":"A binning method for analyzing mixed longitudinal data measured at distinct time points","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Poisson distribution; Mixed model; Smoothing; Poisson regression; Generalized linear mixed model; Longitudinal data; Computer science; Statistics; Preprocessor; Event (particle physics); Generalized linear model; Econometrics; Mathematics; Data mining; Artificial intelligence; Medicine","score_opus":0.14269810999708532,"score_gpt":0.45700158563559795,"score_spread":0.31430347563851263,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2066948187","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0005776917,0.0001039801,0.9988519,0.000029377417,0.000032030493,0.00004475238,0.00007082669,0.00022070076,0.00006876966],"genre_scores_gemma":[0.013540616,0.00026288274,0.9842689,0.00009081355,0.00006634512,0.0006111705,0.00042571287,0.00016550753,0.00056807016],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9927499,0.0046262834,0.00047492335,0.0009187811,0.0010522677,0.00017768808],"domain_scores_gemma":[0.98299384,0.012516375,0.00089382875,0.0021144731,0.0012038819,0.00027746568],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.015099998,0.0010449776,0.0019363066,0.003840272,0.0009571006,0.0014258376,0.0025885974,0.0012974852,0.0050283493],"category_scores_gemma":[0.030739158,0.0010165549,0.0020462507,0.0049810195,0.0014067328,0.0021332013,0.0020883468,0.0031602357,0.0011664496],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006561188,0.0001868351,0.006226985,0.0008553412,0.0011104257,0.00040502666,0.0009096029,0.061748,0.008538937,0.3015878,0.007912522,0.6098624],"study_design_scores_gemma":[0.00016398192,0.00025037653,0.0035702446,0.00018348299,0.00029382162,0.00072034827,0.00014018615,0.6452652,0.004785388,0.3135103,0.030919006,0.00019763083],"about_ca_topic_score_codex":0.0034892522,"about_ca_topic_score_gemma":0.0032896742,"teacher_disagreement_score":0.9849,"about_ca_system_score_codex":0.0008763198,"about_ca_system_score_gemma":0.0022843208,"threshold_uncertainty_score":0.07985741},"labels":[],"label_agreement":null},{"id":"W2068528322","doi":"10.1002/sim.2093","title":"Extending logistic regression to model diffuse interactions","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; St. Paul's Hospital; University of British Columbia","funders":"","keywords":"Logistic regression; Pairwise comparison; Outcome (game theory); Econometrics; Observational study; Computer science; Regression analysis; Ordered logit; Statistics; Extension (predicate logic); Additive model; Mathematics; Mathematical economics","score_opus":0.22472706457083738,"score_gpt":0.5077382236377517,"score_spread":0.2830111590669143,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2068528322","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.021970123,0.00058187486,0.9734248,0.0009870641,0.00006810846,0.00010782887,0.0002428988,0.0003073209,0.0023099969],"genre_scores_gemma":[0.6003381,0.0022630142,0.3786051,0.0011514316,0.0006497788,0.0012597793,0.00094902155,0.0003913527,0.0143925045],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9811717,0.013015057,0.00059039076,0.0024885957,0.0017804308,0.0009537065],"domain_scores_gemma":[0.9333946,0.054023333,0.0052381908,0.00480561,0.0019332112,0.0006050617],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02883075,0.0021356062,0.003192735,0.0033272852,0.00087988825,0.0028148487,0.005125596,0.0036844045,0.004013281],"category_scores_gemma":[0.07724453,0.0014645158,0.003786855,0.0041078944,0.003604069,0.005635337,0.0060448255,0.00611078,0.0017461551],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003493831,0.00017840642,0.014276672,0.0003761264,0.000691804,0.0017878982,0.001147489,0.50928676,0.0014892915,0.41316685,0.0022140902,0.055035233],"study_design_scores_gemma":[0.0000628051,0.000103030965,0.0011608241,0.000037109898,0.000065451815,0.0002867744,0.000057372174,0.7654313,0.00016966097,0.23041603,0.0021587713,0.000050881838],"about_ca_topic_score_codex":0.008114164,"about_ca_topic_score_gemma":0.0046686134,"teacher_disagreement_score":0.02883075,"about_ca_system_score_codex":0.0015735286,"about_ca_system_score_gemma":0.0014275352,"threshold_uncertainty_score":0.15247339},"labels":[],"label_agreement":null},{"id":"W2068953434","doi":"10.1002/sim.2053","title":"The use of the propensity score for estimating treatment effects: administrative versus clinical data","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":225,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; Sunnybrook Health Science Centre; Institute for Clinical Evaluative Sciences; Women's College Hospital; University of Toronto","funders":"","keywords":"Propensity score matching; Statistics; Econometrics; Medicine; Mathematics","score_opus":0.7319400186447765,"score_gpt":0.5861486378777672,"score_spread":0.14579138076700937,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2068953434","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0364692,0.0041089007,0.9514685,0.003570345,0.00027585315,0.0005688406,0.0009930548,0.00013629052,0.0024090211],"genre_scores_gemma":[0.5472874,0.005537903,0.43988684,0.0015729341,0.0008182395,0.0020212666,0.001690907,0.000096226846,0.0010883813],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.85470194,0.13003151,0.0042553744,0.0033543592,0.0071845897,0.0004722185],"domain_scores_gemma":[0.72279996,0.22674589,0.023218872,0.022041984,0.004294283,0.0008990102],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.095158994,0.0008925841,0.0020318988,0.006047806,0.0007102516,0.0032147602,0.0016511565,0.002075332,0.0023562617],"category_scores_gemma":[0.30990645,0.00061488873,0.0018132059,0.009455675,0.0030704862,0.0038919342,0.0027785907,0.0027932182,0.00038834792],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001110565,0.00044092053,0.18150733,0.0017057343,0.0050639706,0.00031222534,0.0011785183,0.045426257,0.0010096373,0.36768228,0.0050201965,0.38954234],"study_design_scores_gemma":[0.001220227,0.002130459,0.11269043,0.0012174647,0.0016851872,0.0011136697,0.00074880134,0.19971742,0.0029143845,0.64831156,0.027934873,0.000315627],"about_ca_topic_score_codex":0.0020962171,"about_ca_topic_score_gemma":0.0014703604,"teacher_disagreement_score":0.904841,"about_ca_system_score_codex":0.0012378779,"about_ca_system_score_gemma":0.0023278897,"threshold_uncertainty_score":0.5032549},"labels":[],"label_agreement":null},{"id":"W2069688121","doi":"10.1002/1097-0258(20001030)19:20<2867::aid-sim551>3.0.co;2-j","title":"Properties of the estimated variance component for subject-by-formulation interaction in studies of individual bioequivalence","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":28,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Bioequivalence; Variance (accounting); Statistics; Restricted maximum likelihood; Econometrics; Variance components; Component (thermodynamics); Mathematics; Maximum likelihood; Medicine; Economics; Pharmacology","score_opus":0.6061784706403508,"score_gpt":0.5724391072053919,"score_spread":0.03373936343495887,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2069688121","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.20311746,0.005021232,0.7875547,0.0007088897,0.00011739823,0.00043637742,0.0005428177,0.00060722366,0.0018938768],"genre_scores_gemma":[0.85722375,0.0006031255,0.14003785,0.00031666938,0.00005513389,0.00054387905,0.00062453945,0.00022773193,0.00036729523],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9030275,0.085772276,0.0032182038,0.0035044637,0.0041293264,0.00034827905],"domain_scores_gemma":[0.354654,0.61718583,0.009562506,0.014874867,0.0034227492,0.00030012595],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.121513136,0.0008673143,0.0015564493,0.0015849144,0.0002862001,0.002061517,0.00096213835,0.0022396275,0.0010901693],"category_scores_gemma":[0.341341,0.00059855153,0.0027538785,0.0016487961,0.0015870036,0.0019356239,0.0008994893,0.0021082286,0.00025837985],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.013502629,0.0004266605,0.21755797,0.0037535292,0.016385075,0.0010531953,0.002026991,0.32656872,0.020982048,0.051344268,0.0028714966,0.34352735],"study_design_scores_gemma":[0.0007737236,0.006314373,0.1588297,0.00089508307,0.0052666864,0.002798006,0.00036756264,0.72326195,0.01890648,0.074456155,0.0075257625,0.0006044892],"about_ca_topic_score_codex":0.00062961155,"about_ca_topic_score_gemma":0.00052246853,"teacher_disagreement_score":0.8784869,"about_ca_system_score_codex":0.0010445521,"about_ca_system_score_gemma":0.0010576914,"threshold_uncertainty_score":0.6426306},"labels":[],"label_agreement":null},{"id":"W2071109714","doi":"10.1002/sim.2655","title":"Analysing and interpreting competing risk data","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":184,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ontario Institute for Cancer Research","funders":"","keywords":"Hazard; Computer science; Event (particle physics); Population; Interpretation (philosophy); Econometrics; Statistics; Risk analysis (engineering); Mathematics; Medicine; Environmental health","score_opus":0.11141470997664706,"score_gpt":0.46070952155919104,"score_spread":0.349294811582544,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2071109714","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.018013746,0.0045151818,0.96241254,0.0060393712,0.0006741503,0.0009587769,0.0028505358,0.00063578127,0.003899962],"genre_scores_gemma":[0.20004313,0.0035948763,0.7856989,0.0025936444,0.0014021136,0.0021682554,0.0031058681,0.0003927074,0.0010005557],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8442545,0.12791862,0.008531388,0.004872096,0.013522712,0.0009007505],"domain_scores_gemma":[0.35922506,0.596873,0.01804977,0.01848089,0.0063100387,0.0010612769],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.13751419,0.0023197818,0.0040415754,0.010610085,0.0011941895,0.0076174103,0.0048409966,0.00507532,0.0048368364],"category_scores_gemma":[0.39639276,0.0011364656,0.0038345777,0.0075119,0.004155857,0.005939,0.004699115,0.0052844263,0.00077139755],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015064839,0.00035175378,0.03594054,0.008025832,0.0031501027,0.0040959893,0.011514056,0.053201716,0.0032725565,0.4905457,0.022580722,0.3658146],"study_design_scores_gemma":[0.0002659672,0.00033938297,0.0075704553,0.0019399994,0.00061931467,0.0019387205,0.0024292073,0.062376745,0.0023657416,0.88041645,0.039483674,0.00025432868],"about_ca_topic_score_codex":0.0016834465,"about_ca_topic_score_gemma":0.0010780714,"teacher_disagreement_score":0.13751419,"about_ca_system_score_codex":0.0018657517,"about_ca_system_score_gemma":0.0033598486,"threshold_uncertainty_score":0.7272532},"labels":[],"label_agreement":null},{"id":"W2071379517","doi":"10.1002/sim.2770","title":"A comparison of regression trees, logistic regression, generalized additive models, and multivariate adaptive regression splines for predicting AMI mortality","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":155,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Logistic regression; Multivariate adaptive regression splines; Statistics; Receiver operating characteristic; Multivariate statistics; Regression; Regression analysis; Logistic model tree; Predictive modelling; Mathematics; Bayesian multivariate linear regression; Medicine","score_opus":0.11830730882407893,"score_gpt":0.4368427016114135,"score_spread":0.31853539278733456,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2071379517","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7057297,0.014308092,0.27147225,0.0026506865,0.00040222978,0.00043483276,0.0010932623,0.0009278963,0.0029810688],"genre_scores_gemma":[0.90093285,0.003975611,0.092698164,0.00022835752,0.00023330584,0.00027447523,0.00092689076,0.00014861308,0.0005817569],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.96772534,0.028271515,0.00075965654,0.0009085169,0.0019047181,0.00043019297],"domain_scores_gemma":[0.89666474,0.091066934,0.0037148972,0.0023757229,0.0050847903,0.0010929228],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.044299833,0.0019087229,0.0021765148,0.0068656923,0.00059643795,0.001533744,0.0015089486,0.0014366577,0.00057297375],"category_scores_gemma":[0.08113194,0.00067814585,0.002478045,0.0037679435,0.00069284736,0.002683205,0.0016020271,0.002051917,0.0003255075],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0061999513,0.00076369615,0.18485364,0.00086390396,0.0035130393,0.0002323017,0.0010240799,0.48803273,0.00064054364,0.0070830015,0.0038881777,0.3029049],"study_design_scores_gemma":[0.00013536119,0.0009911947,0.020062307,0.00012322837,0.00023538324,0.000096054864,0.00017401866,0.971915,0.00013248342,0.005515409,0.00054142997,0.00007817632],"about_ca_topic_score_codex":0.012184601,"about_ca_topic_score_gemma":0.010291224,"teacher_disagreement_score":0.044299833,"about_ca_system_score_codex":0.0011770799,"about_ca_system_score_gemma":0.0022830432,"threshold_uncertainty_score":0.23428267},"labels":[],"label_agreement":null},{"id":"W2071834496","doi":"10.1002/sim.5536","title":"Estimation methods for marginal and association parameters for longitudinal binary data with nonignorable missing observations","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Missing data; Covariate; Pairwise comparison; Computer science; Statistics; Robustness (evolution); Econometrics; Binary data; Data mining; Binary number; Mathematics; Machine learning","score_opus":0.26870786049482126,"score_gpt":0.5006180560725655,"score_spread":0.23191019557774423,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2071834496","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00090546446,0.0003357942,0.9982192,0.00019238258,0.00002707574,0.000042819924,0.00007575135,0.000067536435,0.00013394104],"genre_scores_gemma":[0.04107237,0.0017282204,0.95341116,0.00019824039,0.00026417145,0.0008579087,0.00072045793,0.00016860454,0.001578856],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9890814,0.008038207,0.00049464626,0.0012092944,0.00096861535,0.00020787733],"domain_scores_gemma":[0.93189776,0.05788461,0.003439171,0.003798699,0.0024472235,0.0005325369],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.040021483,0.0017158581,0.0022167969,0.004062342,0.0009786073,0.0018127951,0.00528402,0.0027188938,0.005955999],"category_scores_gemma":[0.13469627,0.0012374528,0.0029442066,0.004129124,0.002676555,0.0048839753,0.0033092129,0.0064450237,0.0017889871],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019233022,0.00016279155,0.009157925,0.00078895135,0.0006040957,0.00026302665,0.00086293806,0.14923897,0.0011911879,0.56251216,0.0060699414,0.26895577],"study_design_scores_gemma":[0.00006948256,0.00007990052,0.0020904094,0.00019647254,0.00013051982,0.00028338382,0.00014134217,0.3984043,0.0007453503,0.5901602,0.0076106754,0.000087909444],"about_ca_topic_score_codex":0.0026933623,"about_ca_topic_score_gemma":0.0035973573,"teacher_disagreement_score":0.040021483,"about_ca_system_score_codex":0.0013232013,"about_ca_system_score_gemma":0.0031792056,"threshold_uncertainty_score":0.21165633},"labels":[],"label_agreement":null},{"id":"W2072082323","doi":"10.1002/sim.4374","title":"A Bayesian approach to risk‐adjusted outcome monitoring in healthcare","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Pacific Northwest National Laboratory; University of Waterloo; Virginia Polytechnic Institute and State University","keywords":"Computer science; Bayes' theorem; Bayesian probability; Outcome (game theory); Prior probability; Health care; Bayes factor; Data mining; Set (abstract data type); Bayesian inference; Machine learning; Artificial intelligence; Mathematics","score_opus":0.2882576065015896,"score_gpt":0.4814126593832788,"score_spread":0.19315505288168922,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2072082323","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0031542825,0.00040609797,0.9946912,0.00056089077,0.000027880773,0.00006746713,0.00008532793,0.00010156618,0.0009051273],"genre_scores_gemma":[0.20971803,0.0018471311,0.78444296,0.00041878564,0.00040204648,0.00074439065,0.0004884883,0.0000881839,0.0018500029],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9861288,0.009640745,0.0005490457,0.0011902363,0.0022021357,0.0002889363],"domain_scores_gemma":[0.9676073,0.027295496,0.001963312,0.0010118077,0.0018241263,0.0002978678],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.021800194,0.0012407824,0.0022655053,0.002786014,0.0008917451,0.0025840336,0.0030282466,0.0022337746,0.0019494402],"category_scores_gemma":[0.06586932,0.0012894776,0.0014841916,0.0026881157,0.0019595341,0.0035899337,0.0022947688,0.003234681,0.0004757899],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017798807,0.00013502862,0.0040015043,0.00020384318,0.00031415222,0.00021379336,0.00038722128,0.5636986,0.00087610225,0.2835777,0.0032480722,0.14316602],"study_design_scores_gemma":[0.000041006242,0.000054257816,0.0007842482,0.000048338785,0.000043413493,0.000082431965,0.00002235976,0.8218315,0.00029027995,0.17504501,0.0017084747,0.00004851969],"about_ca_topic_score_codex":0.009372252,"about_ca_topic_score_gemma":0.0073773433,"teacher_disagreement_score":0.021800194,"about_ca_system_score_codex":0.0024004574,"about_ca_system_score_gemma":0.0028765243,"threshold_uncertainty_score":0.115291774},"labels":[],"label_agreement":null},{"id":"W2072845842","doi":"10.1002/sim.5852","title":"Concordance measures in shared frailty models: application to clustered data in cancer prognosis","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Frailty in Older Adults","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Institute of Cancer Research; Institut National Du Cancer; Association pour la Recherche sur le Cancer","keywords":"Concordance; Censoring (clinical trials); Frailty Index; Nonparametric statistics; Statistics; Index (typography); Covariate; Medicine; Econometrics; Computer science; Internal medicine; Mathematics","score_opus":0.11797337068013256,"score_gpt":0.3889841790892407,"score_spread":0.2710108084091082,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2072845842","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.41588056,0.0028508017,0.577127,0.001112037,0.00015203918,0.0004580093,0.00071596424,0.00028306062,0.0014204911],"genre_scores_gemma":[0.8919802,0.00030172936,0.1064233,0.00013320574,0.00009080041,0.00032995318,0.00040115984,0.000049897542,0.00028979222],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9596451,0.032995388,0.0013058379,0.0028577542,0.002638138,0.0005578159],"domain_scores_gemma":[0.71348107,0.2492679,0.013433026,0.01786746,0.004274132,0.0016764421],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.075103745,0.0010092503,0.0022669123,0.003928742,0.0012544128,0.0022017611,0.0026714175,0.0017816535,0.0008850721],"category_scores_gemma":[0.22072889,0.0004993846,0.0029816746,0.0042706444,0.0029117186,0.0021794704,0.004029191,0.003295265,0.00011927377],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013911095,0.000331217,0.13003089,0.0006758932,0.0031752053,0.0009125677,0.0014267388,0.6417384,0.0009939276,0.085002884,0.0019027111,0.13241848],"study_design_scores_gemma":[0.00015112196,0.00036616973,0.019540416,0.00014044545,0.00027085285,0.00025894324,0.00013406001,0.88570935,0.0005838482,0.091963954,0.0007814683,0.0000993012],"about_ca_topic_score_codex":0.005996372,"about_ca_topic_score_gemma":0.0043405127,"teacher_disagreement_score":0.075103745,"about_ca_system_score_codex":0.002290016,"about_ca_system_score_gemma":0.0022317192,"threshold_uncertainty_score":0.3971913},"labels":[],"label_agreement":null},{"id":"W2072863153","doi":"10.1002/sim.2108","title":"Application of hidden Markov models to multiple sclerosis lesion count data","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":44,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; Actua; Simon Fraser University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Markov chain; Markov model; Multiple sclerosis; Perspective (graphical); Count data; Clinical trial; Artificial intelligence; Machine learning; Hidden Markov model; Data mining; Medicine; Statistics; Mathematics; Pathology","score_opus":0.10554507787332336,"score_gpt":0.3455094780797473,"score_spread":0.23996440020642396,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2072863153","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011402912,0.00089917204,0.98519343,0.0014102871,0.000050791954,0.00004103545,0.00012193761,0.00019031542,0.00069016835],"genre_scores_gemma":[0.47284356,0.003419861,0.5193449,0.0005324581,0.00037588255,0.00039495694,0.00081630534,0.00013541804,0.0021366316],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9958895,0.0029190306,0.0001731165,0.0003550223,0.00053551834,0.00012788118],"domain_scores_gemma":[0.9515418,0.045073364,0.001181951,0.0011948985,0.0007492289,0.00025867778],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008971609,0.0007380149,0.0014942859,0.0019013236,0.00077075313,0.0015730903,0.0019200743,0.0020518613,0.0014616618],"category_scores_gemma":[0.045603957,0.0011054735,0.0015639651,0.002506443,0.0011818688,0.0024143322,0.00195364,0.0030828493,0.00034096837],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00014316391,0.000090722715,0.005002865,0.00020178317,0.00027147148,0.00041768482,0.0005526263,0.6655761,0.0009254634,0.21781172,0.0016655768,0.107340775],"study_design_scores_gemma":[0.000020058911,0.000016580694,0.0004541622,0.00002996523,0.000019620667,0.00006680618,0.000025276104,0.8270733,0.00021414747,0.17101665,0.0010395306,0.000023861176],"about_ca_topic_score_codex":0.007833807,"about_ca_topic_score_gemma":0.008853065,"teacher_disagreement_score":0.008971609,"about_ca_system_score_codex":0.0016029929,"about_ca_system_score_gemma":0.0016145478,"threshold_uncertainty_score":0.047447026},"labels":[],"label_agreement":null},{"id":"W2075167302","doi":"10.1002/sim.3642","title":"A geometric confidence ellipse approach to the estimation of the ratio of two variables. <i>Statistics in Medicine</i> 2008; <b>27</b>:5956–5974.","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Scientific Measurement and Uncertainty Evaluation","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Ellipse; Term (time); Medical statistics; Applied mathematics; Statistics; Carry (investment); Mathematics; Confidence interval; Calculus (dental); Computer science; Geometry; Medicine; Physics","score_opus":0.16376878348943286,"score_gpt":0.42243540449526507,"score_spread":0.2586666210058322,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2075167302","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0047286903,0.004376588,0.9837601,0.001220926,0.00037921048,0.00017600013,0.00032590627,0.00041446366,0.0046181083],"genre_scores_gemma":[0.29284275,0.0046241125,0.6923686,0.0007458219,0.0015165159,0.0013840136,0.0015988066,0.0005925028,0.004326955],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8989984,0.08245967,0.00285504,0.006794355,0.008098198,0.0007942409],"domain_scores_gemma":[0.62972075,0.33434328,0.01156882,0.012583407,0.010242003,0.0015418362],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08159995,0.002219036,0.0034571392,0.012135605,0.0015004505,0.0076089003,0.005714088,0.0039020013,0.009992227],"category_scores_gemma":[0.3656635,0.0015978136,0.0036682733,0.0074389945,0.0076239295,0.005614078,0.006892552,0.005775094,0.0024766312],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004453944,0.0001414673,0.014127939,0.00085681234,0.0014351822,0.00058276905,0.0010004624,0.19634713,0.0003457331,0.43153492,0.025058875,0.3281234],"study_design_scores_gemma":[0.0001264326,0.00029976742,0.006107157,0.00072763395,0.00039417908,0.00088950555,0.0004929431,0.5795125,0.0006591852,0.38220364,0.028399892,0.00018718273],"about_ca_topic_score_codex":0.006452176,"about_ca_topic_score_gemma":0.0023945896,"teacher_disagreement_score":0.08159995,"about_ca_system_score_codex":0.0028593687,"about_ca_system_score_gemma":0.002545121,"threshold_uncertainty_score":0.43154693},"labels":[],"label_agreement":null},{"id":"W2075875797","doi":"10.1002/sim.6198","title":"Small sample GEE estimation of regression parameters for longitudinal data","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":50,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Trent University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Statistics; Generalized estimating equation; Estimator; Mathematics; Marginal model; Sample size determination; Estimating equations; Regression; Regression analysis; Gee; Confidence interval; Standard error; Econometrics","score_opus":0.2614241119052628,"score_gpt":0.46521650086800426,"score_spread":0.20379238896274143,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2075875797","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0036354153,0.00038618938,0.99532676,0.00010832654,0.00001957329,0.000030361418,0.00008535136,0.00017752907,0.00023056804],"genre_scores_gemma":[0.18733059,0.002039765,0.80527395,0.00034448013,0.000111584326,0.0007527985,0.0010667368,0.000522271,0.002557891],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9952851,0.0033682145,0.00019512766,0.00055914506,0.00046864527,0.00012372849],"domain_scores_gemma":[0.9805818,0.015904449,0.001084885,0.0013769417,0.0009321457,0.00011970716],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010318259,0.0011712203,0.0019864456,0.0015944897,0.0003237067,0.00096594956,0.0017081667,0.0012410708,0.002470299],"category_scores_gemma":[0.05662339,0.0006898761,0.0016585309,0.0016143295,0.0009310184,0.0023807157,0.0014106635,0.0024458682,0.00058710465],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00045972943,0.00019517094,0.015494712,0.0017059078,0.00179797,0.0009801184,0.00061848824,0.40311104,0.00846296,0.19033414,0.0096458,0.36719388],"study_design_scores_gemma":[0.00010337679,0.00011845472,0.0045083296,0.00013514672,0.00025750426,0.0006350667,0.00009860313,0.7955953,0.0024581917,0.18622075,0.009769417,0.00009983987],"about_ca_topic_score_codex":0.0034394048,"about_ca_topic_score_gemma":0.003314829,"teacher_disagreement_score":0.010318259,"about_ca_system_score_codex":0.0005477831,"about_ca_system_score_gemma":0.0014888898,"threshold_uncertainty_score":0.054568827},"labels":[],"label_agreement":null},{"id":"W2078983319","doi":"10.1002/sim.6226","title":"A random set approach to confidence regions with applications to the effective dose with combinations of agents","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"Engineering and Physical Sciences Research Council; York University","keywords":"Set (abstract data type); Computer science; Confidence interval; Mathematics; Statistics","score_opus":0.08333612306527494,"score_gpt":0.434442795328112,"score_spread":0.35110667226283704,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2078983319","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0015356047,0.0005300353,0.9971386,0.00019499428,0.00004120864,0.00006893098,0.00005402282,0.00009862218,0.00033808104],"genre_scores_gemma":[0.18834925,0.0019555774,0.8039268,0.00042492244,0.0006461417,0.0016440559,0.0006981669,0.0003201719,0.0020349415],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9448981,0.045551337,0.0014825272,0.0034853935,0.0038114106,0.00077113707],"domain_scores_gemma":[0.49339083,0.47855595,0.008613517,0.009794581,0.008341296,0.0013038515],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.079090424,0.00275009,0.005278887,0.007693893,0.0015434207,0.004673817,0.0076655843,0.004363673,0.005633371],"category_scores_gemma":[0.25713155,0.0021861014,0.0053188787,0.0051490823,0.007066612,0.004843607,0.0061279135,0.008018545,0.00080058584],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00036502964,0.0001521989,0.003497717,0.00056324963,0.0009887734,0.00061648973,0.00051841955,0.49128667,0.00090091245,0.44220552,0.0018950533,0.05701007],"study_design_scores_gemma":[0.00008257098,0.00015787265,0.00046634563,0.0001373687,0.00012737744,0.00017190521,0.000063879575,0.848941,0.0006494735,0.14683363,0.0023002867,0.00006822557],"about_ca_topic_score_codex":0.0051839924,"about_ca_topic_score_gemma":0.0025542134,"teacher_disagreement_score":0.079090424,"about_ca_system_score_codex":0.0028319373,"about_ca_system_score_gemma":0.002354067,"threshold_uncertainty_score":0.41827512},"labels":[],"label_agreement":null},{"id":"W2079160353","doi":"10.1002/sim.1432","title":"Estimating the distribution of lag in the effect of short‐term exposures and interventions: adaptation of a non‐parametric regression spline model","year":2003,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; McGill University; Montreal General Hospital; Institut National de la Recherche Scientifique","funders":"","keywords":"Statistics; Lag; Econometrics; Parametric statistics; Covariate; Regression analysis; Framingham Heart Study; Mathematics; Computer science; Medicine; Framingham Risk Score; Disease; Internal medicine","score_opus":0.25325483171826113,"score_gpt":0.4727333543971716,"score_spread":0.2194785226789105,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2079160353","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017485205,0.00029912617,0.98097813,0.00027201482,0.000056599485,0.00011517064,0.00019207386,0.00027366422,0.000328044],"genre_scores_gemma":[0.59680605,0.002334889,0.38860527,0.0003480504,0.00028842533,0.0016246571,0.001583686,0.00035523783,0.008053708],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9898968,0.0074835136,0.00040539878,0.0009593194,0.0007329292,0.0005221672],"domain_scores_gemma":[0.91639745,0.07382279,0.0025524083,0.004380562,0.0023239995,0.0005227536],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.046006184,0.0014428378,0.0043986635,0.0023912406,0.00048291692,0.0021953492,0.0050767963,0.0031656825,0.0037117272],"category_scores_gemma":[0.07875312,0.0015297796,0.004397849,0.0032494226,0.002078458,0.002761664,0.0022416934,0.0051026545,0.0008514662],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006770719,0.0002131238,0.008276539,0.00026926474,0.0005781704,0.00033546687,0.000318785,0.84913576,0.0010008618,0.07400376,0.0009793518,0.064211875],"study_design_scores_gemma":[0.00009804246,0.00014955507,0.0011455036,0.000030775613,0.00011873824,0.00006431272,0.00002874951,0.9669513,0.00023318204,0.030284023,0.0008606013,0.000035192737],"about_ca_topic_score_codex":0.010397515,"about_ca_topic_score_gemma":0.0051987236,"teacher_disagreement_score":0.046006184,"about_ca_system_score_codex":0.0012268429,"about_ca_system_score_gemma":0.0027458433,"threshold_uncertainty_score":0.24330688},"labels":[],"label_agreement":null},{"id":"W2079566573","doi":"10.1002/sim.2814","title":"Semi‐parametric regression models for cost‐effectiveness analysis: improving the efficiency of estimation from censored data","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hospital for Sick Children; University of Toronto; University of Waterloo","funders":"","keywords":"Estimator; Statistics; Weighting; Covariate; Inverse probability weighting; Contrast (vision); Parametric statistics; Regression; Linear regression; Regression analysis; Mathematics; Variance (accounting); Econometrics; Computer science; Medicine; Artificial intelligence","score_opus":0.3306744055215149,"score_gpt":0.49142923597086385,"score_spread":0.16075483044934896,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2079566573","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.001106343,0.00044404477,0.9977422,0.00021203916,0.000024889276,0.000053495092,0.000040367893,0.00010798672,0.00026864992],"genre_scores_gemma":[0.13694508,0.0037827138,0.85316026,0.00059806893,0.00040892558,0.0016185665,0.0006658163,0.00043479886,0.002385705],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.94129825,0.05260253,0.0010378743,0.0016207735,0.0029758646,0.000464794],"domain_scores_gemma":[0.80373776,0.17952883,0.0064171464,0.00664957,0.003339226,0.0003274324],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0584243,0.0023072164,0.004113018,0.0026555336,0.00042698174,0.002588532,0.0049037854,0.0025189465,0.0040661],"category_scores_gemma":[0.1789823,0.0016268526,0.003955926,0.003429231,0.0018474329,0.005432691,0.0023567427,0.007186494,0.0015111177],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00031763272,0.00019519633,0.0021595482,0.0007949392,0.0007893564,0.0002892687,0.00040346343,0.5899057,0.0009622416,0.24976192,0.0038608888,0.15055987],"study_design_scores_gemma":[0.00008219074,0.00012498586,0.000501647,0.00012171226,0.00009520191,0.00010965921,0.00002884693,0.8907363,0.0005836442,0.10428318,0.003263019,0.00006967386],"about_ca_topic_score_codex":0.0022916398,"about_ca_topic_score_gemma":0.002003216,"teacher_disagreement_score":0.0584243,"about_ca_system_score_codex":0.0015069403,"about_ca_system_score_gemma":0.0019087156,"threshold_uncertainty_score":0.30898094},"labels":[],"label_agreement":null},{"id":"W2080813654","doi":"10.1002/sim.1173","title":"Number needed to treat (NNT): estimation of a measure of clinical benefit","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":81,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; McMaster University Medical Centre","funders":"","keywords":"Number needed to treat; Estimator; Context (archaeology); Event (particle physics); Medicine; Randomized controlled trial; Statistics; Absolute risk reduction; Clinical trial; Confidence interval; Econometrics; Computer science; Relative risk; Mathematics; Surgery; Internal medicine","score_opus":0.5337564789862494,"score_gpt":0.6268482031556304,"score_spread":0.09309172416938094,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2080813654","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017492,0.15575282,0.761569,0.01721848,0.015400743,0.013236046,0.006309848,0.0010890635,0.011932033],"genre_scores_gemma":[0.36596352,0.04870826,0.49818823,0.019373076,0.0065942206,0.052070666,0.0037025588,0.0006521079,0.004747273],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7831093,0.17503943,0.012537027,0.008525112,0.020149052,0.0006400527],"domain_scores_gemma":[0.6751333,0.302113,0.010924946,0.0065026376,0.0042219716,0.0011042546],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.14809483,0.002585815,0.013638433,0.004387716,0.0008595958,0.0033566214,0.0036367245,0.0067605367,0.0057116775],"category_scores_gemma":[0.37093803,0.00064150017,0.00615368,0.0035181055,0.004594005,0.0059357663,0.0031313717,0.008134321,0.0012097481],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.012455651,0.0009632795,0.020744743,0.04512102,0.057066895,0.0006736637,0.0010605516,0.07617289,0.0017399708,0.15941891,0.057303052,0.56727934],"study_design_scores_gemma":[0.00478714,0.011909386,0.010974376,0.014635739,0.02676404,0.0017245295,0.00063665485,0.12509646,0.004483172,0.6531008,0.14519599,0.0006918387],"about_ca_topic_score_codex":0.0011830211,"about_ca_topic_score_gemma":0.0008052694,"teacher_disagreement_score":0.85190517,"about_ca_system_score_codex":0.0041303732,"about_ca_system_score_gemma":0.004161833,"threshold_uncertainty_score":0.7832097},"labels":[],"label_agreement":null},{"id":"W2082625612","doi":"10.1002/sim.4037","title":"Bayesian sample size for diagnostic test studies in the absence of a gold standard: Comparing identifiable with non‐identifiable models","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre; Royal Victoria Regional Health Centre; McGill University; Royal Victoria Hospital","funders":"","keywords":"Sample size determination; Statistics; Sample (material); Bayesian probability; Conditional independence; Imperfect; Statistical hypothesis testing; Computer science; Gold standard (test); Econometrics; Independence (probability theory); Mathematics","score_opus":0.08947318568387154,"score_gpt":0.4160162086473315,"score_spread":0.32654302296346,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2082625612","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.05109348,0.0047315275,0.93187135,0.005361651,0.0004937171,0.0015645686,0.00051067513,0.00024477864,0.004128265],"genre_scores_gemma":[0.6578589,0.0017442699,0.3317109,0.0016137059,0.0004430245,0.004685239,0.000712264,0.0001559398,0.0010757957],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7101726,0.2569321,0.006894593,0.012205544,0.012239662,0.0015554716],"domain_scores_gemma":[0.15735207,0.8008322,0.013261362,0.021723425,0.0053203134,0.0015106146],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.34879985,0.0016621089,0.003961384,0.0034654054,0.0015276617,0.004679186,0.0063168122,0.0066301217,0.0031643708],"category_scores_gemma":[0.6523741,0.0018703741,0.0045986148,0.0016982422,0.009758399,0.009040403,0.006581864,0.0070006093,0.0003260703],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0065529807,0.00044173707,0.048962798,0.003306294,0.0047971397,0.0010647351,0.0034050473,0.22851986,0.0016474387,0.52333987,0.0047555882,0.17320654],"study_design_scores_gemma":[0.0014417695,0.0014249698,0.010420397,0.0010755517,0.0013857867,0.00037733425,0.00042755206,0.4299618,0.001411271,0.54560536,0.0062854863,0.00018269329],"about_ca_topic_score_codex":0.0030981677,"about_ca_topic_score_gemma":0.0021584178,"teacher_disagreement_score":0.6512002,"about_ca_system_score_codex":0.0036968677,"about_ca_system_score_gemma":0.0034445783,"threshold_uncertainty_score":0.80304575},"labels":[],"label_agreement":null},{"id":"W2083675429","doi":"10.1002/1097-0258(20000730)19:14<1952::aid-sim474>3.0.co;2-k","title":"The theory of dispersion models. Bent J�rgensen, Chapman and Hall, 1997. No. of pages: 237. Price: �39.95. ISBN 0-412-99718-8","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Citation; Library science; Mathematical economics; Statistics; Computer science; Mathematics","score_opus":0.04923718532831025,"score_gpt":0.3494584815210401,"score_spread":0.30022129619272986,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2083675429","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0025656624,0.47120774,0.45950446,0.011988222,0.00593922,0.00010448477,0.0036248537,0.0026233422,0.042442143],"genre_scores_gemma":[0.10666155,0.5395831,0.26178393,0.0029910882,0.007969433,0.0008174042,0.005449884,0.0031441182,0.071599476],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9963862,0.0013475703,0.0003559189,0.0005539114,0.0012434489,0.00011304506],"domain_scores_gemma":[0.9884179,0.008371359,0.00097550446,0.0009502761,0.0011055926,0.00017936327],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004376135,0.0039780755,0.0055186166,0.0042004464,0.0009404437,0.00482525,0.0024934914,0.0029482532,0.0253118],"category_scores_gemma":[0.018539414,0.0036301075,0.0019898536,0.0055832253,0.0033746907,0.007616008,0.0021693066,0.0063393237,0.013398427],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019743726,0.00009660039,0.0011529144,0.0036228988,0.0007199424,0.0003487001,0.00078051246,0.025447411,0.0016780557,0.23589614,0.3064924,0.4235671],"study_design_scores_gemma":[0.000040085913,0.00008000001,0.0021307312,0.0010743262,0.00027504613,0.00075352174,0.00019062645,0.017697794,0.0007066447,0.7242235,0.25264952,0.0001782207],"about_ca_topic_score_codex":0.0076062363,"about_ca_topic_score_gemma":0.00815415,"teacher_disagreement_score":0.0253118,"about_ca_system_score_codex":0.0028970537,"about_ca_system_score_gemma":0.0018217397,"threshold_uncertainty_score":0.084676385},"labels":[],"label_agreement":null},{"id":"W2084292194","doi":"10.1002/sim.2731","title":"Developments in cluster randomized trials and <i>Statistics in Medicine</i>","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":248,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"Royal Society","keywords":"Sample size determination; Medical statistics; Statistics; Cluster (spacecraft); Cluster analysis; Randomized controlled trial; Computer science; Population; Psychological intervention; Research design; Medicine; Econometrics; Data science; Mathematics; Surgery","score_opus":0.08618382227126765,"score_gpt":0.4329159356107219,"score_spread":0.34673211333945425,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2084292194","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0005791133,0.16980775,0.76735836,0.038826525,0.010225883,0.0006341304,0.00025730074,0.00044331964,0.011867657],"genre_scores_gemma":[0.019045558,0.13632669,0.798959,0.019458594,0.020031631,0.0032775514,0.0002347149,0.00053586386,0.002130437],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.7996178,0.16822897,0.008494695,0.0059692506,0.01669221,0.0009970459],"domain_scores_gemma":[0.5907113,0.36931247,0.009268369,0.016011726,0.012838517,0.0018576778],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.14576566,0.002188136,0.0059994,0.0070496453,0.0013983446,0.007007676,0.005191343,0.007548613,0.0051487964],"category_scores_gemma":[0.22903499,0.0019018756,0.0038881295,0.012975643,0.014395131,0.0073122797,0.0043004085,0.022175634,0.0023449145],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00018967903,0.00007970778,0.00047483935,0.0042249397,0.0005055555,0.0001243827,0.00049707334,0.004075204,0.00021460681,0.7620524,0.027495692,0.200066],"study_design_scores_gemma":[0.00017777782,0.00031173913,0.000605182,0.002757224,0.00018297978,0.0004756189,0.00013642598,0.009123604,0.000529565,0.8313168,0.15427335,0.00010966915],"about_ca_topic_score_codex":0.0025706894,"about_ca_topic_score_gemma":0.0013399113,"teacher_disagreement_score":0.85423434,"about_ca_system_score_codex":0.0059290593,"about_ca_system_score_gemma":0.010841206,"threshold_uncertainty_score":0.7708917},"labels":[],"label_agreement":null},{"id":"W2085249160","doi":"10.1002/sim.3448","title":"Statistical issues in the design and analysis of expertise‐based randomized clinical trials","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Hip disorders and treatments","field":"Medicine","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"Canadian Institutes of Health Research","keywords":"Randomized controlled trial; Research design; Confounding; Medicine; Clinical study design; Medical physics; Intervention (counseling); Completely randomized design; Computer science; Randomized experiment; Clinical trial; Physical therapy; Statistics; Surgery; Mathematics; Nursing","score_opus":0.17718784971561388,"score_gpt":0.5005786867762901,"score_spread":0.3233908370606763,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2085249160","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005345693,0.006956682,0.93505514,0.009562451,0.009033327,0.027496407,0.00081428053,0.0013662822,0.0043697543],"genre_scores_gemma":[0.104348786,0.0015894569,0.7690758,0.008565741,0.0025271394,0.111577325,0.0002852997,0.00043463742,0.0015957976],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.07017191,0.8686401,0.02757805,0.008055682,0.02441829,0.0011359586],"domain_scores_gemma":[0.087452754,0.8231581,0.022859165,0.054246552,0.011127417,0.0011559982],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7233786,0.0037042827,0.010894227,0.007408384,0.00299844,0.0095409695,0.0077139735,0.012411809,0.0067291376],"category_scores_gemma":[0.8736344,0.0032471872,0.009351021,0.013620875,0.018741528,0.0070759295,0.0059628803,0.01807384,0.0016886661],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.02769976,0.0011641942,0.01013973,0.026439728,0.023469837,0.0021053571,0.0068043345,0.03224362,0.0019517877,0.4431083,0.05721639,0.367657],"study_design_scores_gemma":[0.012709066,0.009309919,0.009199779,0.010284904,0.004397797,0.0011977492,0.0010385687,0.18893039,0.005647543,0.6570092,0.099648885,0.0006261621],"about_ca_topic_score_codex":0.0018457869,"about_ca_topic_score_gemma":0.0013131573,"teacher_disagreement_score":0.2766214,"about_ca_system_score_codex":0.008118202,"about_ca_system_score_gemma":0.010965694,"threshold_uncertainty_score":0.34112346},"labels":[],"label_agreement":null},{"id":"W2087047720","doi":"10.1002/(sici)1097-0258(20000315)19:5<715::aid-sim342>3.0.co;2-t","title":"GEE Analysis of negatively correlated binary responses: a caution","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Royal Victoria Hospital","funders":"National Cancer Institute","keywords":"Generalized estimating equation; Bounding overwatch; Binary number; Independence (probability theory); Statistics; Correlation; Binary data; Gee; Mathematics; Binomial (polynomial); Negative binomial distribution; Applied mathematics; Econometrics; Computer science","score_opus":0.05795689346548086,"score_gpt":0.41620766905050866,"score_spread":0.3582507755850278,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2087047720","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.024311937,0.02506432,0.65492314,0.25745505,0.023113523,0.00094542006,0.0015632055,0.003523549,0.009099924],"genre_scores_gemma":[0.24284877,0.013934533,0.45102954,0.24782963,0.017391307,0.0024431453,0.000551916,0.002204539,0.021766663],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.85711086,0.108587794,0.011160031,0.008990794,0.013552793,0.0005977634],"domain_scores_gemma":[0.5587799,0.3669347,0.0090837255,0.038097207,0.024131278,0.0029731926],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.22383264,0.0027186933,0.0061689215,0.004354571,0.0018019566,0.0043788617,0.008742268,0.004351543,0.0026944908],"category_scores_gemma":[0.50540155,0.0013327713,0.0035771804,0.004771564,0.012846834,0.008093519,0.0050327694,0.023409907,0.0028241016],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020402195,0.0002989086,0.054006297,0.00783798,0.005245081,0.008132012,0.014573235,0.005487606,0.00629622,0.13895664,0.42766225,0.3294636],"study_design_scores_gemma":[0.000505916,0.00069250073,0.04242561,0.009590694,0.0016457891,0.008455878,0.0061798315,0.039617658,0.0065996386,0.6708314,0.21246213,0.000992952],"about_ca_topic_score_codex":0.009328613,"about_ca_topic_score_gemma":0.0151192155,"teacher_disagreement_score":0.22383264,"about_ca_system_score_codex":0.0015392269,"about_ca_system_score_gemma":0.002241218,"threshold_uncertainty_score":0.95715255},"labels":[],"label_agreement":null},{"id":"W2087811140","doi":"10.1002/sim.3830","title":"A copula‐based mixed Poisson model for bivariate recurrent events under event‐dependent censoring","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":33,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Canadian Institutes of Health Research; Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs; GlaxoSmithKline","keywords":"Bivariate analysis; Copula (linguistics); Censoring (clinical trials); Estimator; Econometrics; Poisson distribution; Statistics; Marginal model; Random effects model; Marginal distribution; Negative binomial distribution; Parametric statistics; Mathematics; Regression analysis; Random variable; Medicine; Internal medicine; Meta-analysis","score_opus":0.1449655822639636,"score_gpt":0.4545331163864958,"score_spread":0.3095675341225322,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2087811140","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0050066994,0.0006730449,0.9926515,0.00035215155,0.00006349964,0.00007718429,0.00017926643,0.00013506379,0.00086155743],"genre_scores_gemma":[0.39895272,0.005210177,0.5760287,0.0009367597,0.00074165233,0.0021822127,0.0016946979,0.00044849497,0.013804622],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99379784,0.0042038565,0.00022903155,0.0007671765,0.00067628716,0.00032581048],"domain_scores_gemma":[0.9843114,0.011986879,0.0014992196,0.0010957652,0.000792308,0.00031445673],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014413408,0.0013665218,0.002348795,0.001777409,0.000732834,0.0020972656,0.004634168,0.0022721966,0.0037908815],"category_scores_gemma":[0.038295887,0.0010264276,0.0022927674,0.00264133,0.001697038,0.0027277002,0.0017114695,0.0032412908,0.0013936231],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013249058,0.00010234779,0.004173988,0.0003801605,0.00025194066,0.00082297093,0.0003949343,0.34090298,0.0010111658,0.60576993,0.0050529903,0.041004106],"study_design_scores_gemma":[0.00003894258,0.000091590984,0.0010093171,0.000060297392,0.000093682385,0.00032454057,0.00005013582,0.83094555,0.00019058678,0.16228348,0.004858786,0.000053122818],"about_ca_topic_score_codex":0.0052856016,"about_ca_topic_score_gemma":0.0038802775,"teacher_disagreement_score":0.014413408,"about_ca_system_score_codex":0.001598419,"about_ca_system_score_gemma":0.00217517,"threshold_uncertainty_score":0.076226234},"labels":[],"label_agreement":null},{"id":"W2089365578","doi":"10.1002/sim.3295","title":"The power of testing a semi‐parametric shared gamma frailty parameter in failure time data","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Insurance, Mortality, Demography, Risk Management","field":"Social Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"","keywords":"Estimator; Nonparametric statistics; Statistics; Parametric statistics; Mathematics; Hazard; Power (physics); Parametric model; Function (biology); Econometrics; Computer science; Physics","score_opus":0.08350817376501589,"score_gpt":0.3597535548972013,"score_spread":0.2762453811321854,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2089365578","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.45698887,0.0011324735,0.53385025,0.0018627613,0.00013592733,0.00025938387,0.00041454248,0.00024648328,0.005109299],"genre_scores_gemma":[0.973545,0.00019594183,0.025184283,0.00022861129,0.00007957755,0.00021610862,0.00022334256,0.000042720538,0.00028447772],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95322275,0.0372512,0.00146093,0.0035799239,0.0036185237,0.0008665962],"domain_scores_gemma":[0.35020456,0.6264243,0.00691948,0.013491618,0.002163079,0.0007969029],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11421949,0.00084977056,0.002668709,0.0023108942,0.0006795767,0.0023593013,0.0024335934,0.003894262,0.0030155987],"category_scores_gemma":[0.38327876,0.0005401176,0.002782798,0.0017623261,0.0059980806,0.005070969,0.0033381712,0.0024763707,0.00042843525],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008041472,0.00084257394,0.24260192,0.0015026027,0.0041773613,0.002130589,0.002497312,0.26192334,0.010744009,0.107345834,0.0031829474,0.35501003],"study_design_scores_gemma":[0.00093260186,0.003695101,0.08242059,0.00052930583,0.0009306676,0.002340564,0.0009705944,0.5963546,0.008900999,0.3004875,0.0022070748,0.00023053265],"about_ca_topic_score_codex":0.000686525,"about_ca_topic_score_gemma":0.00024133333,"teacher_disagreement_score":0.11421949,"about_ca_system_score_codex":0.0008699342,"about_ca_system_score_gemma":0.0013017607,"threshold_uncertainty_score":0.6040576},"labels":[],"label_agreement":null},{"id":"W2089415104","doi":"10.1002/sim.1687","title":"Inflation of the type I error rate when a continuous confounding variable is categorized in logistic regression analyses","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":248,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Confounding; Statistics; Categorical variable; Type I and type II errors; Logistic regression; Mathematics; Inflation (cosmology); Sample size determination; Econometrics; Variable (mathematics); Regression analysis","score_opus":0.5624346742081207,"score_gpt":0.6001798666323802,"score_spread":0.037745192424259555,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2089415104","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.043008465,0.0044373004,0.94084066,0.003113846,0.0019764658,0.0021341168,0.0003753471,0.0009323945,0.0031813278],"genre_scores_gemma":[0.57599986,0.0017063883,0.40785536,0.0048865518,0.0010135343,0.005482162,0.00057582976,0.00071142655,0.0017689009],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.3503839,0.5251655,0.04071745,0.029611545,0.051153705,0.0029679022],"domain_scores_gemma":[0.08561659,0.82280946,0.037958458,0.0398222,0.012853602,0.000939805],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.47830343,0.002276254,0.0037026203,0.004241432,0.0019057855,0.0043992708,0.0034504135,0.0057501295,0.0019696052],"category_scores_gemma":[0.7635874,0.0016108197,0.00482435,0.005101701,0.009339102,0.005247748,0.00556312,0.01027712,0.0007312943],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.014898176,0.001153811,0.32425106,0.0064837513,0.01559317,0.007480071,0.014842256,0.082832485,0.007979672,0.14336869,0.015065902,0.36605084],"study_design_scores_gemma":[0.0018071209,0.0060629854,0.14468504,0.0064665307,0.0085175205,0.010127123,0.003422387,0.4521603,0.03348206,0.30835268,0.023161888,0.0017543964],"about_ca_topic_score_codex":0.002364938,"about_ca_topic_score_gemma":0.0023084779,"teacher_disagreement_score":0.52169657,"about_ca_system_score_codex":0.0044993344,"about_ca_system_score_gemma":0.0039416905,"threshold_uncertainty_score":0.64334476},"labels":[],"label_agreement":null},{"id":"W2090906954","doi":"10.1002/sim.2791","title":"A likelihood approach to estimating sensitivity and specificity for binocular data: application in ophthalmology","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Royal Alexandra Hospital; University of Alberta; University of Calgary","funders":"Natural Sciences and Engineering Research Council of Canada; University of Calgary","keywords":"Sensitivity (control systems); Extension (predicate logic); Maximum likelihood; Computer science; Statistics; Binary data; Optometry; Mathematics; Artificial intelligence; Binary number; Medicine","score_opus":0.11626529237192867,"score_gpt":0.4428071937723671,"score_spread":0.32654190140043843,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2090906954","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0019115248,0.0004970583,0.9966125,0.00041905756,0.00002244469,0.0000311205,0.000041540705,0.00007350738,0.000391194],"genre_scores_gemma":[0.11488029,0.0015612566,0.88112175,0.00027589028,0.0003158546,0.00040058637,0.00020698491,0.00014065385,0.001096698],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97703856,0.019549506,0.000594205,0.0010745912,0.0015601948,0.00018290387],"domain_scores_gemma":[0.8379876,0.15281783,0.0032885142,0.003286108,0.002141108,0.00047890097],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.030381044,0.0014430323,0.0022706068,0.0049372097,0.00089154165,0.0024815898,0.0027244878,0.0031710577,0.002312196],"category_scores_gemma":[0.18607795,0.0010263489,0.0017753367,0.004081405,0.0035848592,0.002748648,0.003122682,0.003971617,0.00073935237],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002920967,0.00017357529,0.017656177,0.00070507743,0.0007358354,0.001215877,0.0010086311,0.30370763,0.0022177817,0.3546186,0.00506342,0.3126054],"study_design_scores_gemma":[0.000057278336,0.00011961329,0.0034447953,0.00014543039,0.00007847431,0.0011577159,0.00015739343,0.6095952,0.00071610935,0.3808684,0.0035426882,0.00011695382],"about_ca_topic_score_codex":0.0028062994,"about_ca_topic_score_gemma":0.0023556948,"teacher_disagreement_score":0.030381044,"about_ca_system_score_codex":0.0013330461,"about_ca_system_score_gemma":0.0019040139,"threshold_uncertainty_score":0.16067219},"labels":[],"label_agreement":null},{"id":"W2091019309","doi":"10.1002/sim.5466","title":"Sample size formulas for estimating intraclass correlation coefficients with precision and assurance","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":467,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"","keywords":"Intraclass correlation; Confidence interval; Reliability (semiconductor); Sample size determination; Statistics; Interval (graph theory); Mathematics; Coverage probability; Limit (mathematics); Interval estimation; Sample (material); Correlation coefficient; Correlation; Computer science; Reproducibility; Power (physics); Mathematical analysis; Combinatorics; Physics","score_opus":0.0854428886232476,"score_gpt":0.3958068923931848,"score_spread":0.31036400376993717,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2091019309","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0008068777,0.0004975127,0.99730325,0.00014509096,0.00006766779,0.00020569534,0.00004702564,0.00017677087,0.00075002055],"genre_scores_gemma":[0.030111043,0.0009770368,0.9642659,0.00034582763,0.00024811778,0.0029390205,0.0001731038,0.00020343851,0.0007365174],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.94550496,0.036181286,0.0031412449,0.0024060952,0.012356334,0.00041001724],"domain_scores_gemma":[0.69354427,0.2722119,0.008609248,0.011498303,0.0137368115,0.00039952886],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07010303,0.0019456067,0.0020435213,0.0065801954,0.0006737425,0.0018395318,0.0038502063,0.0028834587,0.0053610657],"category_scores_gemma":[0.43800405,0.0010259735,0.0014521312,0.0035972155,0.0021462587,0.004309688,0.0029592952,0.0047382195,0.0020703352],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00029597152,0.00021597376,0.0047391746,0.0010860753,0.00043763578,0.0003353592,0.0012418716,0.055469878,0.002294069,0.29321906,0.016821232,0.6238437],"study_design_scores_gemma":[0.00055290014,0.0007522691,0.0062887687,0.0016255159,0.00045909264,0.0011454031,0.00024713058,0.43748534,0.006104487,0.51413274,0.030988911,0.00021738854],"about_ca_topic_score_codex":0.0013412487,"about_ca_topic_score_gemma":0.0015935632,"teacher_disagreement_score":0.07010303,"about_ca_system_score_codex":0.0014701582,"about_ca_system_score_gemma":0.0018922131,"threshold_uncertainty_score":0.37074465},"labels":[],"label_agreement":null},{"id":"W2091501852","doi":"10.1002/sim.3289","title":"Interval estimation of risk difference for data sampled from clusters","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Windsor","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Estimator; Statistics; Confidence interval; Mathematics; Cluster (spacecraft); Variance (accounting); Cluster sampling; Sample size determination; Coverage probability; Interval estimation; Sampling (signal processing); Computer science; Econometrics; Population; Demography","score_opus":0.16970249555560965,"score_gpt":0.4414141727607394,"score_spread":0.2717116772051298,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2091501852","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009676342,0.00092425075,0.9876386,0.00013378379,0.00009980622,0.00023641226,0.00027063265,0.00033393377,0.0006861515],"genre_scores_gemma":[0.20170195,0.0011637306,0.7912554,0.00033701115,0.00020915682,0.0022946487,0.0018860879,0.00020130974,0.00095062074],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.94736177,0.038741317,0.0024445339,0.0057279565,0.005236355,0.0004880214],"domain_scores_gemma":[0.7441729,0.22535656,0.010337118,0.014883391,0.0046536154,0.0005964853],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04882107,0.001123504,0.0029259876,0.0039155544,0.00050297106,0.0018345733,0.0040088613,0.0025857917,0.0031050073],"category_scores_gemma":[0.2610552,0.00064689683,0.0027200498,0.0035264937,0.0017829593,0.0023095577,0.0027633207,0.003543293,0.0006845441],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0030702525,0.00040009915,0.04407735,0.0024498575,0.0046999934,0.0009063825,0.0019151053,0.18110842,0.00392383,0.22019853,0.0072465497,0.53000367],"study_design_scores_gemma":[0.00044045981,0.0010956716,0.018944684,0.00065494335,0.0008611293,0.0011723951,0.00035108766,0.7141925,0.0055160252,0.2435688,0.012928205,0.0002740718],"about_ca_topic_score_codex":0.0016179511,"about_ca_topic_score_gemma":0.00055938296,"teacher_disagreement_score":0.04882107,"about_ca_system_score_codex":0.001068263,"about_ca_system_score_gemma":0.0011511062,"threshold_uncertainty_score":0.2581936},"labels":[],"label_agreement":null},{"id":"W2092848201","doi":"10.1002/sim.2680","title":"Modelling smoking history using a comprehensive smoking index: application to lung cancer","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Smoking Behavior and Cessation","field":"Medicine","cited_by":139,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Montreal General Hospital; Université de Montréal; McGill University; Queen's University","funders":"","keywords":"Lung cancer; Statistics; Smoking history; Covariate; Econometrics; Regression analysis; Medicine; Index (typography); Demography; Computer science; Mathematics; Oncology; Internal medicine","score_opus":0.05194383189967641,"score_gpt":0.3575085325930135,"score_spread":0.3055647006933371,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2092848201","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.40099543,0.0015092018,0.59376585,0.0006976315,0.00005160148,0.00026427075,0.0008078245,0.00055610016,0.0013519938],"genre_scores_gemma":[0.87187505,0.0008171824,0.12475988,0.00008972429,0.0000601579,0.00018129422,0.0007613877,0.000056076005,0.0013992754],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9980253,0.0011860202,0.00009613569,0.00034833784,0.00022815606,0.000116078285],"domain_scores_gemma":[0.9823292,0.014718911,0.001147011,0.0008168314,0.00065592484,0.00033219642],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008474787,0.0011712043,0.0019143879,0.0020783194,0.00050278293,0.0011423968,0.0014897372,0.00089169585,0.0008179024],"category_scores_gemma":[0.026677582,0.0004933874,0.0019033228,0.003069831,0.00077283755,0.00092676526,0.0012237914,0.0010420996,0.00012521686],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004633338,0.00015672219,0.111315295,0.00017489294,0.0012096626,0.0003200606,0.00044672104,0.7811418,0.0013236488,0.0119560445,0.00069773017,0.09079408],"study_design_scores_gemma":[0.000029283636,0.00018806224,0.017425124,0.000014577915,0.00011395494,0.0000836171,0.000042737138,0.9740849,0.000260228,0.007134887,0.0005819264,0.00004064591],"about_ca_topic_score_codex":0.06432156,"about_ca_topic_score_gemma":0.042713374,"teacher_disagreement_score":0.06432156,"about_ca_system_score_codex":0.0018453862,"about_ca_system_score_gemma":0.0020817511,"threshold_uncertainty_score":0.12789434},"labels":[],"label_agreement":null},{"id":"W2095752327","doi":"10.1002/(sici)1097-0258(20000315)19:5<723::aid-sim379>3.0.co;2-a","title":"Interval estimation for Cohen's kappa as a measure of agreement","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":217,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Confidence interval; Statistics; Mathematics; Measure (data warehouse); Kappa; Statistic; Variance (accounting); Delta method; Coverage probability; Asymptotic analysis; Computation; Asymptotic distribution; Cohen's kappa; Applied mathematics; Computer science; Algorithm","score_opus":0.11948518704376823,"score_gpt":0.4260948892710158,"score_spread":0.3066097022272476,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2095752327","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01449865,0.004029885,0.9684323,0.00034121433,0.00042138563,0.0004896483,0.0006398733,0.00079025637,0.010356778],"genre_scores_gemma":[0.23013839,0.002170677,0.7607723,0.00027090928,0.0004029033,0.0034911733,0.0010694223,0.00043467685,0.0012495762],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.88243437,0.073471084,0.0093514845,0.008111298,0.025526503,0.0011053241],"domain_scores_gemma":[0.72677356,0.22011895,0.014268459,0.011604728,0.026369905,0.0008644462],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0671707,0.00091949623,0.002020541,0.010514451,0.0012728136,0.0031269183,0.0040423907,0.001764458,0.0056109834],"category_scores_gemma":[0.35187203,0.00045548257,0.0021520834,0.0076470324,0.0021750913,0.0037535569,0.0038512198,0.0030729293,0.0018018271],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013625259,0.00025201068,0.020504374,0.004183941,0.0017819274,0.00032242743,0.0044105654,0.02588647,0.0031588431,0.19698578,0.026440132,0.714711],"study_design_scores_gemma":[0.00033591362,0.0025551491,0.058203224,0.004742139,0.0016647425,0.0034833292,0.0034158004,0.26082444,0.01144719,0.5630474,0.089386486,0.0008941182],"about_ca_topic_score_codex":0.0015303632,"about_ca_topic_score_gemma":0.0008842369,"teacher_disagreement_score":0.0671707,"about_ca_system_score_codex":0.0015670634,"about_ca_system_score_gemma":0.002102059,"threshold_uncertainty_score":0.3552369},"labels":[],"label_agreement":null},{"id":"W2095973044","doi":"10.1002/(sici)1097-0258(20000415)19:7<975::aid-sim381>3.0.co;2-9","title":"A robust mixed linear model analysis for longitudinal data","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":62,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; Okanagan University College; Okanagan College","funders":"Natural Sciences and Engineering Research Council of Canada; University of British Columbia; McMaster University","keywords":"Random effects model; Mixed model; Generalized linear mixed model; Autocorrelation; Bayesian probability; Longitudinal data; Computer science; Statistics; Linear model; Econometrics; Linear regression; Robust regression; Subject (documents); Mathematics; Data mining; Medicine","score_opus":0.3914669943620445,"score_gpt":0.473223682916968,"score_spread":0.08175668855492346,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2095973044","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00014131554,0.00039531625,0.9983821,0.000103899576,0.00006184326,0.00007609905,0.00019376483,0.00044055254,0.0002050509],"genre_scores_gemma":[0.0060678376,0.00084081164,0.9897226,0.0001538027,0.00020817186,0.0012740256,0.0006819498,0.00029029392,0.0007604316],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9706069,0.023070341,0.00096158544,0.001978688,0.0031164866,0.000265936],"domain_scores_gemma":[0.97316325,0.02056283,0.001752199,0.0026433517,0.0015710646,0.00030731247],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02239689,0.0023036047,0.0030570186,0.003252677,0.00086552935,0.002518362,0.0038209192,0.0019694816,0.007592058],"category_scores_gemma":[0.065907754,0.0011694896,0.004402798,0.0036100685,0.0011150414,0.001988256,0.0032758308,0.0041188565,0.004206924],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00036937578,0.00020924976,0.0019992818,0.001948472,0.0029131535,0.0006794405,0.00044738248,0.08374695,0.004761789,0.32681066,0.026383443,0.5497308],"study_design_scores_gemma":[0.0002759142,0.00096293923,0.0021332377,0.00080652133,0.001046002,0.0011579103,0.0001246596,0.4941424,0.0040283175,0.3656603,0.12930687,0.00035488073],"about_ca_topic_score_codex":0.0017723368,"about_ca_topic_score_gemma":0.0019620229,"teacher_disagreement_score":0.02239689,"about_ca_system_score_codex":0.0010115871,"about_ca_system_score_gemma":0.0034934562,"threshold_uncertainty_score":0.11844748},"labels":[],"label_agreement":null},{"id":"W2096521417","doi":"10.1002/sim.5317","title":"A simulation study of finite‐sample properties of marginal structural Cox proportional hazards models","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":31,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Eunice Kennedy Shriver National Institute of Child Health and Human Development; National Center for Research Resources; National Institute of Allergy and Infectious Diseases; National Institute on Alcohol Abuse and Alcoholism; National Institutes of Health","keywords":"Statistics; Confidence interval; Hazard ratio; Proportional hazards model; Mathematics; Estimator; Standard error; Confounding; Mean squared error; Sample size determination; Marginal structural model; Range (aeronautics); Nominal level; Econometrics; Coverage probability","score_opus":0.21699590011067962,"score_gpt":0.45071684087391395,"score_spread":0.23372094076323432,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2096521417","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.56112224,0.000984355,0.42927277,0.001456739,0.0001215727,0.0004333277,0.0010917956,0.00036097385,0.005156279],"genre_scores_gemma":[0.9434386,0.00026266198,0.05411174,0.00020909587,0.000027142121,0.00052993436,0.00061520643,0.000048761372,0.00075685215],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9893328,0.009016928,0.00025374055,0.0006355422,0.00045087494,0.00031023874],"domain_scores_gemma":[0.62804174,0.3522179,0.0068807737,0.0072418563,0.004629756,0.0009878979],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0356744,0.00062752346,0.0012069696,0.00088605133,0.0005803345,0.0010067491,0.0021970742,0.0013101758,0.002973535],"category_scores_gemma":[0.14191005,0.00054351526,0.001705669,0.001015332,0.00149626,0.0015368558,0.0010387142,0.0021765418,0.00017669392],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00057885324,0.00016118106,0.019396221,0.0002238025,0.000328884,0.00024746754,0.00041054614,0.92724353,0.0004476966,0.041859586,0.0009200516,0.008182231],"study_design_scores_gemma":[0.00010766812,0.00015725239,0.0015220435,0.000048174024,0.000065183995,0.00005592806,0.000088821,0.98442185,0.00025752338,0.0128598055,0.0003938986,0.00002172005],"about_ca_topic_score_codex":0.0102083795,"about_ca_topic_score_gemma":0.006847843,"teacher_disagreement_score":0.9643256,"about_ca_system_score_codex":0.0014166428,"about_ca_system_score_gemma":0.0016628823,"threshold_uncertainty_score":0.18866658},"labels":[],"label_agreement":null},{"id":"W2096649070","doi":"10.1002/sim.3387","title":"Bivariate modeling of interobserver agreement coefficients","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Bivariate analysis; Anxiety; Outcome (game theory); Trait; Psychology; Statistics; Depression (economics); Clinical psychology; Medicine; Psychiatry; Mathematics; Computer science","score_opus":0.2636804934739743,"score_gpt":0.41067999707973896,"score_spread":0.14699950360576464,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2096649070","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.10067422,0.00081602903,0.89156836,0.0010570196,0.00018451977,0.0006945487,0.0011183533,0.00073228014,0.0031546606],"genre_scores_gemma":[0.8322732,0.00087629375,0.1542054,0.00020179027,0.00018912021,0.0024012488,0.0016690813,0.00033194796,0.007852002],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9545941,0.032668274,0.0017663501,0.0057805334,0.003502819,0.0016879949],"domain_scores_gemma":[0.8081056,0.1537402,0.010938187,0.014369826,0.011858133,0.0009881046],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.057258803,0.0019209599,0.0021890556,0.0053053764,0.0014437409,0.0038049493,0.0037541864,0.0025094538,0.0074192197],"category_scores_gemma":[0.18984991,0.001381787,0.0033089437,0.0057984046,0.002711539,0.003005923,0.003140839,0.0037974687,0.00300504],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00079279294,0.00046229135,0.23564301,0.000577741,0.002740091,0.0016233231,0.01159733,0.29718032,0.0019522439,0.29398975,0.00870193,0.14473921],"study_design_scores_gemma":[0.00004292633,0.00028351895,0.028593685,0.0002030999,0.00032881842,0.0005960018,0.0008423489,0.85890394,0.0009327226,0.1032761,0.0058724647,0.00012433282],"about_ca_topic_score_codex":0.012089817,"about_ca_topic_score_gemma":0.0054071,"teacher_disagreement_score":0.057258803,"about_ca_system_score_codex":0.0021847507,"about_ca_system_score_gemma":0.0021600195,"threshold_uncertainty_score":0.3028171},"labels":[],"label_agreement":null},{"id":"W2096726864","doi":"10.1002/sim.1028","title":"Regression models for allele sharing: analysis of accumulating data in affected sib pair studies","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic Associations and Epidemiology","field":"Biochemistry, Genetics and Molecular Biology","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill Genome Centre; McGill University Health Centre; Lunenfeld-Tanenbaum Research Institute; University of Toronto; Mount Sinai Hospital","funders":"","keywords":"Genetics; Allele; Biology; Microsatellite; Genotype; Genetic marker; Genetic association; Disease; Gene; Single-nucleotide polymorphism; Medicine","score_opus":0.17005254369451553,"score_gpt":0.4320606346360941,"score_spread":0.2620080909415786,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2096726864","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.265933,0.0070118583,0.71590877,0.0019573234,0.00036879204,0.00094360585,0.0044557443,0.0013026069,0.002118247],"genre_scores_gemma":[0.7839325,0.0030017672,0.1927614,0.00060259906,0.00048910995,0.0029685616,0.008842345,0.0004953967,0.0069061946],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.90873784,0.07595341,0.0025845307,0.008688711,0.002545748,0.0014897215],"domain_scores_gemma":[0.52252823,0.43401617,0.019214068,0.01852911,0.0040194276,0.0016930323],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.15939802,0.0026680543,0.0055227764,0.008421968,0.0016380473,0.0039038216,0.009653827,0.0050006206,0.007523153],"category_scores_gemma":[0.25898647,0.0021311925,0.0057689343,0.00921004,0.0034500642,0.004156185,0.0040250025,0.0055685383,0.0015210949],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002277448,0.0007811135,0.39172235,0.0013247156,0.016095689,0.004678422,0.0027372048,0.40874302,0.0011740024,0.06531519,0.0071445564,0.0980063],"study_design_scores_gemma":[0.00023003435,0.0004956942,0.043012846,0.00034587877,0.0017942637,0.0010525489,0.0005388359,0.88986576,0.0003674042,0.056627005,0.005513392,0.0001563384],"about_ca_topic_score_codex":0.013008914,"about_ca_topic_score_gemma":0.0059112064,"teacher_disagreement_score":0.15939802,"about_ca_system_score_codex":0.0016344656,"about_ca_system_score_gemma":0.0016544557,"threshold_uncertainty_score":0.84298736},"labels":[],"label_agreement":null},{"id":"W2096911929","doi":"10.1002/sim.1290","title":"Clinical research and trials in developing countries","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Blood Pressure and Hypertension Studies","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Developing country; Clinical trial; Medicine; Economic growth; Economics; Pathology","score_opus":0.47626321956660006,"score_gpt":0.5355790470014814,"score_spread":0.05931582743488134,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2096911929","genre_codex":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014398989,0.56060594,0.032211483,0.15386559,0.03064909,0.07896211,0.009305722,0.00058528886,0.119415835],"genre_scores_gemma":[0.1692261,0.33708462,0.10408738,0.07753279,0.026574863,0.26895803,0.005122119,0.00032514698,0.011088986],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5273394,0.4170572,0.037708975,0.0048905304,0.010577227,0.0024267198],"domain_scores_gemma":[0.510678,0.28210983,0.08765446,0.05127469,0.04915287,0.019130217],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.25827026,0.0011930639,0.0046440563,0.0064228545,0.0029389174,0.00825186,0.0019467301,0.0060347747,0.021073723],"category_scores_gemma":[0.38377768,0.0014804722,0.0018002825,0.016312996,0.006231318,0.005379473,0.005968889,0.0055507855,0.005323068],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.02161595,0.0015450952,0.015168218,0.07454343,0.004496992,0.00091482775,0.0026452364,0.0025808048,0.0011541166,0.11982788,0.2337984,0.521709],"study_design_scores_gemma":[0.035264283,0.008446781,0.02145015,0.088138744,0.0022898451,0.0011903398,0.0020463718,0.0012474458,0.0008434831,0.06324907,0.7756438,0.00018962403],"about_ca_topic_score_codex":0.0026311607,"about_ca_topic_score_gemma":0.0030275946,"teacher_disagreement_score":0.74172974,"about_ca_system_score_codex":0.0074470043,"about_ca_system_score_gemma":0.05063206,"threshold_uncertainty_score":0.91468483},"labels":[],"label_agreement":null},{"id":"W2097027506","doi":"10.1002/1097-0258(20000730)19:14<1849::aid-sim506>3.0.co;2-1","title":"Estimating treatment effects in randomized clinical trials in the presence of non-compliance","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":159,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"Universiteit Leiden","keywords":"Confounding; Randomization; Randomized controlled trial; Outcome (game theory); Clinical trial; Medicine; Placebo; Estimator; Statistics; Instrumental variable; Causal inference; Econometrics; Internal medicine; Mathematics","score_opus":0.28974091353382553,"score_gpt":0.5720931192616544,"score_spread":0.2823522057278289,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2097027506","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008835229,0.012252669,0.968828,0.0032339583,0.0008413474,0.0038735308,0.00035655257,0.00043881044,0.0013399214],"genre_scores_gemma":[0.22968785,0.007883799,0.7375632,0.0033997083,0.0011301116,0.018172627,0.0007097764,0.00016056298,0.0012923252],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.29246777,0.66639924,0.018153146,0.009995091,0.011806282,0.0011784156],"domain_scores_gemma":[0.15420438,0.80438566,0.023548776,0.014657194,0.002617696,0.00058625307],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.46716672,0.0040790155,0.01291666,0.0075416393,0.0013071134,0.0062311734,0.005770726,0.011145333,0.0043017524],"category_scores_gemma":[0.7364189,0.003865177,0.010118636,0.008388656,0.009720949,0.008364313,0.005010463,0.009223339,0.0010201372],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009097969,0.0007565379,0.018481435,0.024998214,0.027722666,0.0017028404,0.0024193702,0.18664436,0.0011695323,0.41532207,0.0063449885,0.30533996],"study_design_scores_gemma":[0.0045931824,0.002670834,0.0031925966,0.002905105,0.005243473,0.0005121393,0.0001967368,0.16590329,0.0013341713,0.80432224,0.008910054,0.0002161842],"about_ca_topic_score_codex":0.0020743595,"about_ca_topic_score_gemma":0.0013228984,"teacher_disagreement_score":0.5328333,"about_ca_system_score_codex":0.0046092924,"about_ca_system_score_gemma":0.007826042,"threshold_uncertainty_score":0.6570783},"labels":[],"label_agreement":null},{"id":"W2097223884","doi":"10.1002/sim.2813","title":"A comparison of the statistical power of different methods for the analysis of cluster randomization trials with binary outcomes","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":58,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Statistics; Wilcoxon signed-rank test; Statistical power; Mathematics; Generalized estimating equation; Cluster randomised controlled trial; Statistical hypothesis testing; Randomization; Intraclass correlation; Type I and type II errors; Restricted randomization; Randomized controlled trial; Sample size determination; Resampling; Random effects model; Meta-analysis; Medicine; Mann–Whitney U test","score_opus":0.13128335709531094,"score_gpt":0.5424346180488774,"score_spread":0.41115126095356647,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2097223884","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.078758545,0.014784016,0.8825285,0.0026471382,0.0013731921,0.010467044,0.0012523135,0.00090863224,0.0072806617],"genre_scores_gemma":[0.36240628,0.0034483753,0.6047288,0.0013067516,0.00024614786,0.025523944,0.00084412284,0.00060505985,0.0008904858],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.4697768,0.48251575,0.014968529,0.011355676,0.020088788,0.0012944853],"domain_scores_gemma":[0.15701371,0.79582334,0.016003648,0.019888008,0.010511135,0.00076020765],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.40307885,0.0022854668,0.004512257,0.0051955953,0.0013564728,0.0040361737,0.0034853464,0.004551568,0.00626107],"category_scores_gemma":[0.6913324,0.0015252176,0.009927823,0.0048005814,0.004973448,0.005252983,0.0037382944,0.0050061215,0.0007107089],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.05198953,0.0019183157,0.03597588,0.021739682,0.05326297,0.0006442325,0.005863421,0.11614381,0.004893466,0.14507817,0.01374225,0.5487483],"study_design_scores_gemma":[0.03272177,0.029076185,0.06501802,0.017200468,0.020960914,0.0019223887,0.0018590257,0.4426036,0.016227242,0.32109627,0.049202833,0.0021113933],"about_ca_topic_score_codex":0.0011881974,"about_ca_topic_score_gemma":0.0009100504,"teacher_disagreement_score":0.59692115,"about_ca_system_score_codex":0.0029243103,"about_ca_system_score_gemma":0.0042667673,"threshold_uncertainty_score":0.7361101},"labels":[],"label_agreement":null},{"id":"W2098062766","doi":"10.1002/sim.822","title":"Simultaneous modelling of operative mortality and long‐term survival after coronary artery bypass surgery","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Insurance, Mortality, Demography, Risk Management","field":"Social Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"BC Cancer Agency; Simon Fraser University; University of Alberta","funders":"Division of Mathematical Sciences; Heart and Stroke Foundation of Canada","keywords":"Covariate; Poisson regression; Coronary artery bypass surgery; Proportional hazards model; Medicine; Survival analysis; Bypass surgery; Accelerated failure time model; Artery; Poisson distribution; Regression analysis; Term (time); Surgery; Cardiology; Internal medicine; Statistics; Mathematics; Population","score_opus":0.05708445331418451,"score_gpt":0.3571871711612406,"score_spread":0.3001027178470561,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2098062766","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6422361,0.000854877,0.34807676,0.0025842402,0.00011017334,0.00015118152,0.0014959337,0.0002999045,0.0041907313],"genre_scores_gemma":[0.9821181,0.00043056748,0.010180166,0.000056263445,0.00005004246,0.00024627981,0.0005478645,0.00002880823,0.006341817],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9978054,0.0011940178,0.00009392429,0.0003466712,0.00021279791,0.00034717095],"domain_scores_gemma":[0.9910863,0.006610396,0.0011788104,0.0003655627,0.00035455753,0.000404285],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00417192,0.0010003996,0.0011260202,0.00084956293,0.0005201431,0.0016471173,0.0021884057,0.0020112263,0.0023379682],"category_scores_gemma":[0.01574837,0.00087472313,0.0019147323,0.0011256128,0.0013964808,0.0015608149,0.0028963138,0.0019385038,0.0003916712],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005565972,0.00023113929,0.035214413,0.00014298866,0.00035627352,0.00058423285,0.001126419,0.89828545,0.0019893534,0.04442648,0.00063869834,0.016448002],"study_design_scores_gemma":[0.000051704046,0.00028100648,0.014098812,0.000028400085,0.00012011994,0.00013028282,0.00014759738,0.9518325,0.0004886496,0.031725008,0.0010545684,0.000041414747],"about_ca_topic_score_codex":0.009488487,"about_ca_topic_score_gemma":0.009606617,"teacher_disagreement_score":0.009488487,"about_ca_system_score_codex":0.0014234993,"about_ca_system_score_gemma":0.0023247502,"threshold_uncertainty_score":0.022063494},"labels":[],"label_agreement":null},{"id":"W2098196266","doi":"10.1002/sim.6115","title":"Statistics for quantifying heterogeneity in univariate and bivariate meta-analyses of binary data: The case of meta-analyses of diagnostic accuracy","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":215,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Canadian Institutes of Health Research","keywords":"Bivariate analysis; Univariate; Statistics; Variance (accounting); Estimator; Meta-analysis; Econometrics; Statistic; Bivariate data; Mathematics; Multivariate statistics; Medicine","score_opus":0.9168580576152456,"score_gpt":0.6467614652057476,"score_spread":0.270096592409498,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2098196266","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.004810434,0.027185867,0.9632841,0.0015236547,0.0004548693,0.0005473736,0.001023821,0.00060226605,0.0005676493],"genre_scores_gemma":[0.237032,0.011524543,0.74172056,0.0017130708,0.0010374221,0.004821503,0.0012419341,0.0006039456,0.00030502392],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.59528494,0.35124078,0.023676956,0.011619976,0.017431915,0.00074539194],"domain_scores_gemma":[0.27317768,0.6699879,0.022549054,0.028596438,0.005335911,0.00035295976],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.29157823,0.0031181357,0.009074728,0.015776526,0.00069269945,0.0047295094,0.006194029,0.0056785145,0.0013859391],"category_scores_gemma":[0.5997665,0.0014654832,0.015256893,0.015568791,0.004140815,0.00508566,0.0046351026,0.00800042,0.00041953608],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020266715,0.00015196702,0.033622485,0.04316075,0.18558489,0.0018336318,0.0019118529,0.15761358,0.0021851887,0.18405052,0.018124059,0.36973453],"study_design_scores_gemma":[0.0008933693,0.0009877877,0.012333795,0.007532652,0.03864757,0.0022751847,0.00045715805,0.2165527,0.0044018202,0.6976382,0.01755423,0.0007255357],"about_ca_topic_score_codex":0.001248588,"about_ca_topic_score_gemma":0.0009457388,"teacher_disagreement_score":0.70842177,"about_ca_system_score_codex":0.002628387,"about_ca_system_score_gemma":0.0023649796,"threshold_uncertainty_score":0.87361014},"labels":[],"label_agreement":null},{"id":"W2098373960","doi":"10.1002/sim.1858","title":"Methods for modelling change in cluster randomization trials","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":63,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Guelph; Cancer Care Ontario","funders":"National Institute on Drug Abuse","keywords":"Cluster randomised controlled trial; Sample size determination; Cluster (spacecraft); Randomization; Analysis of covariance; Intervention (counseling); Computer science; Statistics; Covariance; Statistical power; Outcome (game theory); Random effects model; Econometrics; Randomized controlled trial; Medicine; Mathematics; Meta-analysis","score_opus":0.3197844407598542,"score_gpt":0.5497557916040277,"score_spread":0.22997135084417347,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2098373960","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0004779922,0.001107817,0.9942971,0.0007431931,0.00023545255,0.0018118384,0.00028854492,0.0003252156,0.0007129252],"genre_scores_gemma":[0.028218852,0.001921655,0.9378795,0.0009509808,0.00031667075,0.028285025,0.0005359174,0.00021287374,0.0016785705],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7100932,0.26211658,0.007395058,0.007888422,0.011345756,0.0011609243],"domain_scores_gemma":[0.57672364,0.3758335,0.018754242,0.019644257,0.008085459,0.0009589265],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.24552618,0.003362334,0.006151138,0.0066169035,0.00118443,0.0045064525,0.008615368,0.0077457125,0.011185549],"category_scores_gemma":[0.45565516,0.0025321709,0.006336267,0.007738261,0.00511469,0.004994622,0.0041475887,0.00956134,0.003088278],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012674235,0.00020085114,0.0031528107,0.00414362,0.0030184584,0.00025606004,0.0011168729,0.15660752,0.00030557273,0.6419577,0.01413231,0.17384084],"study_design_scores_gemma":[0.0011950537,0.0005373169,0.0007693913,0.0012026241,0.0007620887,0.0001603134,0.0000907893,0.2707694,0.00045670886,0.7021411,0.021760702,0.00015454329],"about_ca_topic_score_codex":0.0052103098,"about_ca_topic_score_gemma":0.0039540473,"teacher_disagreement_score":0.7544738,"about_ca_system_score_codex":0.0047178203,"about_ca_system_score_gemma":0.0061156936,"threshold_uncertainty_score":0.93040055},"labels":[],"label_agreement":null},{"id":"W2098378469","doi":"10.1002/sim.6740","title":"Flexible estimation of survival curves conditional on non‐linear and time‐dependent predictor effects","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Royal Victoria Hospital; McGill University","funders":"","keywords":"Covariate; Statistics; Proportional hazards model; Mathematics; Survival analysis; Accelerated failure time model; Survival function; Hazard ratio; Econometrics; Confidence interval","score_opus":0.2149952634266843,"score_gpt":0.4461013914400345,"score_spread":0.23110612801335023,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2098378469","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.037606835,0.00018043608,0.96048117,0.00029535254,0.000018224737,0.000118696364,0.00050056505,0.0003757782,0.00042290596],"genre_scores_gemma":[0.7513868,0.00045736376,0.24297272,0.0003132854,0.000047244965,0.00071368215,0.0020951335,0.00033311738,0.0016806942],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98970896,0.006978043,0.0005801243,0.001126625,0.0010422211,0.0005640215],"domain_scores_gemma":[0.92536706,0.057329427,0.0048011662,0.010242975,0.0018338917,0.0004254938],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03257642,0.001148465,0.0017983862,0.0020881689,0.00042061252,0.0018511526,0.002250943,0.0013910892,0.0027656015],"category_scores_gemma":[0.106736735,0.00079766015,0.0031385976,0.0027571176,0.0017535222,0.003113984,0.0027860366,0.003567506,0.0005461341],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00055674015,0.0001370383,0.0601239,0.00037318,0.0010552063,0.00059173815,0.00072871445,0.66822505,0.0033238789,0.07336955,0.002415191,0.18909986],"study_design_scores_gemma":[0.00012435962,0.00021143652,0.028238706,0.000120832934,0.00017742653,0.0002615341,0.00011080506,0.8048374,0.002365953,0.16058943,0.0028391331,0.00012298241],"about_ca_topic_score_codex":0.005606356,"about_ca_topic_score_gemma":0.004267991,"teacher_disagreement_score":0.03257642,"about_ca_system_score_codex":0.001162447,"about_ca_system_score_gemma":0.0020661897,"threshold_uncertainty_score":0.1722827},"labels":[],"label_agreement":null},{"id":"W2098998213","doi":"10.1002/sim.4228","title":"A flexible genome‐wide bootstrap method that accounts for rankingand threshold‐selection bias in GWAS interpretation and replication study design","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic Associations and Epidemiology","field":"Biochemistry, Genetics and Molecular Biology","cited_by":42,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Lunenfeld-Tanenbaum Research Institute; Public Health Ontario; University of Toronto; Mount Sinai Hospital","funders":"Canadian Institutes of Health Research; Wellcome Trust","keywords":"Replication (statistics); Genome-wide association study; Selection (genetic algorithm); Computer science; Interpretation (philosophy); Selection bias; Computational biology; Biology; Statistics; Genetics; Artificial intelligence; Mathematics; Single-nucleotide polymorphism; Gene","score_opus":0.14032837438079812,"score_gpt":0.39995353483916085,"score_spread":0.2596251604583627,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2098998213","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0016623511,0.000108386084,0.99520457,0.000090615256,0.00008115381,0.0003717502,0.0002949863,0.0018681512,0.0003180802],"genre_scores_gemma":[0.027607242,0.00011482538,0.9664115,0.00021697502,0.00013048873,0.002554798,0.0008848942,0.0011062154,0.00097313256],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97799087,0.015764767,0.0012695251,0.0018230238,0.002836297,0.0003155021],"domain_scores_gemma":[0.9509104,0.02948873,0.0027947994,0.01180562,0.0043495586,0.000650932],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.035523683,0.0011340135,0.0019827555,0.0025185703,0.000890074,0.0014740307,0.0030313074,0.0020169003,0.011526352],"category_scores_gemma":[0.11535223,0.0010214424,0.0027649149,0.00306565,0.0010174244,0.0011001492,0.0025762059,0.0028126275,0.0052731866],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00097625016,0.0002638807,0.010029374,0.0010684703,0.0014732884,0.0007403392,0.00079935323,0.03277447,0.014639306,0.036049906,0.028187763,0.8729975],"study_design_scores_gemma":[0.0023271702,0.0011083377,0.026511233,0.0006896391,0.0014269566,0.0031316245,0.0002220612,0.6336176,0.022761941,0.17650975,0.13107927,0.0006143952],"about_ca_topic_score_codex":0.0012874337,"about_ca_topic_score_gemma":0.0019855024,"teacher_disagreement_score":0.035523683,"about_ca_system_score_codex":0.00048237017,"about_ca_system_score_gemma":0.003186425,"threshold_uncertainty_score":0.18786943},"labels":[],"label_agreement":null},{"id":"W2099320460","doi":"10.1002/sim.6523","title":"Adaptive sampling in two‐phase designs: a biomarker study for progression in arthritis","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":33,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Estimator; Computer science; Sampling (signal processing); Exploit; Adaptive sampling; Optimal design; Sample size determination; Phase (matter); Biomarker; Resource allocation; Statistics; Data mining; Machine learning; Mathematics; Monte Carlo method","score_opus":0.5332255921247048,"score_gpt":0.6223730082861888,"score_spread":0.08914741616148403,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2099320460","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04723827,0.00053769257,0.9494162,0.0005154008,0.00018249494,0.0011922498,0.00003879651,0.00008029085,0.00079857593],"genre_scores_gemma":[0.44776958,0.00044226492,0.5466694,0.00040852313,0.0001371501,0.003619292,0.00005923477,0.000023193972,0.0008713572],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.94425803,0.05155675,0.0005255619,0.0016458187,0.0017228436,0.00029107137],"domain_scores_gemma":[0.9181011,0.07045214,0.0037327213,0.004854066,0.0021760545,0.0006839235],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.048335806,0.0011457043,0.0014957471,0.00082375965,0.000621972,0.0010856666,0.0017256646,0.0022205715,0.0022012168],"category_scores_gemma":[0.10776352,0.0006418145,0.001704307,0.00076570845,0.0025568036,0.0012166116,0.0016134768,0.0019528423,0.00023070857],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.019730628,0.0031273714,0.015033323,0.0020976965,0.0014782021,0.000508571,0.0012710196,0.20835947,0.014456812,0.4039982,0.0022854952,0.32765323],"study_design_scores_gemma":[0.00652541,0.02579275,0.005794747,0.0003390126,0.0008386475,0.00026894364,0.0001730419,0.7286752,0.005841655,0.21846639,0.007048763,0.00023538478],"about_ca_topic_score_codex":0.00045066405,"about_ca_topic_score_gemma":0.00046264284,"teacher_disagreement_score":0.048335806,"about_ca_system_score_codex":0.0007510737,"about_ca_system_score_gemma":0.0017756892,"threshold_uncertainty_score":0.25562721},"labels":[],"label_agreement":null},{"id":"W2099844674","doi":"10.1002/sim.1115","title":"Current and future challenges in the design and analysis of cluster randomization trials","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":172,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; Cancer Care Ontario","funders":"","keywords":"Randomization; Cluster (spacecraft); Cluster randomised controlled trial; Restricted randomization; Psychological intervention; Randomized controlled trial; Research design; Computer science; Medicine; Statistics; Mathematics; Psychiatry","score_opus":0.6291099124678506,"score_gpt":0.5903951121051839,"score_spread":0.038714800362666724,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2099844674","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.002299597,0.10123846,0.47134817,0.40370303,0.01170131,0.00298208,0.0008653067,0.00096927315,0.0048926985],"genre_scores_gemma":[0.030714927,0.034824442,0.8564803,0.046935137,0.015745407,0.013093883,0.000527184,0.000587109,0.0010917232],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.2266357,0.6802186,0.041205887,0.013429029,0.03689036,0.0016204164],"domain_scores_gemma":[0.043062665,0.89244413,0.012048683,0.019699,0.029671872,0.003073598],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.75922245,0.0031897922,0.014422854,0.007825679,0.0035463045,0.015032389,0.016851114,0.01387118,0.008687127],"category_scores_gemma":[0.8520404,0.0034950867,0.0085371155,0.0133167375,0.025259404,0.020816393,0.008705618,0.024877474,0.003584854],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0031786973,0.0004592709,0.0050256173,0.019550093,0.003112853,0.00028452178,0.003198725,0.012532891,0.00034478723,0.29302388,0.062434483,0.5968542],"study_design_scores_gemma":[0.002261056,0.0014870996,0.004102049,0.018826101,0.0013555972,0.00076183287,0.0016647086,0.029526487,0.00046601187,0.83292305,0.106019296,0.0006067169],"about_ca_topic_score_codex":0.010311721,"about_ca_topic_score_gemma":0.010948822,"teacher_disagreement_score":0.24077755,"about_ca_system_score_codex":0.013101595,"about_ca_system_score_gemma":0.041003786,"threshold_uncertainty_score":0.29692167},"labels":[],"label_agreement":null},{"id":"W2100071619","doi":"10.1002/sim.1021","title":"Spatio‐temporal modelling of rates for the construction of disease maps","year":2002,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Data-Driven Disease Surveillance","field":"Medicine","cited_by":69,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Simon Fraser University; University of British Columbia; Providence Health Care","funders":"","keywords":"Popularity; Generalized linear mixed model; Geography; Disease; Computer science; Econometrics; Statistics; Regional science; Medicine; Mathematics","score_opus":0.09536913431964626,"score_gpt":0.3928062166486516,"score_spread":0.29743708232900534,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2100071619","genre_codex":"methods","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0025545708,0.0619532,0.9222136,0.0019474874,0.00054465444,0.00022707648,0.0019375272,0.00072882953,0.00789305],"genre_scores_gemma":[0.10618394,0.17888695,0.70223314,0.0004508006,0.00054162147,0.0015458602,0.0032418622,0.00035432505,0.006561566],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99710375,0.0016445393,0.0002936878,0.00034087885,0.00055589847,0.000061138715],"domain_scores_gemma":[0.9917841,0.0053190202,0.0010713461,0.00078994466,0.000964157,0.000071377355],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0056823283,0.0012143314,0.0015746265,0.005310482,0.00042341527,0.0025912516,0.0031405122,0.0015547193,0.0032489887],"category_scores_gemma":[0.019706089,0.001073538,0.0025434012,0.0076885317,0.001012794,0.0022530432,0.0011860969,0.0018609468,0.0025171787],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000043372136,0.000062988656,0.008248734,0.0037804896,0.00058712816,0.0002598394,0.000718593,0.17908502,0.0007907698,0.24504898,0.014709019,0.54666513],"study_design_scores_gemma":[0.00004045418,0.00010335341,0.009032641,0.0029645357,0.0005072168,0.001227179,0.00056050264,0.40340266,0.001924182,0.23376267,0.34622374,0.00025086696],"about_ca_topic_score_codex":0.019313283,"about_ca_topic_score_gemma":0.015570215,"teacher_disagreement_score":0.019313283,"about_ca_system_score_codex":0.0025118506,"about_ca_system_score_gemma":0.0022006743,"threshold_uncertainty_score":0.038401723},"labels":[],"label_agreement":null},{"id":"W2101735855","doi":"10.1002/sim.1285","title":"The Cochrane Collaboration – how is it progressing?","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Medical Research Council Canada","keywords":"Cochrane collaboration; Systematic review; Quality (philosophy); Randomized controlled trial; Medicine; MEDLINE; Alternative medicine; Cochrane Library; Family medicine; Political science; Pathology","score_opus":0.5756978853050545,"score_gpt":0.5595249099652397,"score_spread":0.016172975339814766,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2101735855","genre_codex":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00017797895,0.59354085,0.0061531286,0.3421188,0.05114053,0.00055360736,0.00023614825,0.00039666123,0.005682281],"genre_scores_gemma":[0.0066389,0.8048519,0.053289738,0.09805616,0.030142162,0.0028359813,0.0008809842,0.0004861097,0.0028180808],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.69779074,0.16570511,0.05180891,0.01664781,0.06313411,0.004913355],"domain_scores_gemma":[0.31880572,0.35202873,0.021803927,0.030863224,0.25071615,0.025782207],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.34979203,0.005196815,0.014264645,0.030225193,0.0062885187,0.032152504,0.010700091,0.033363327,0.0136974575],"category_scores_gemma":[0.60308844,0.004913046,0.007281236,0.034845266,0.017396266,0.043124985,0.010976787,0.042813126,0.010753393],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006289373,0.00012288091,0.00088669214,0.034263995,0.0014394686,0.00012807801,0.00066020555,0.001084665,0.00016612,0.037758816,0.3609232,0.56193686],"study_design_scores_gemma":[0.00070033147,0.00025506204,0.0010247073,0.124546304,0.0013765439,0.0002287718,0.0006247084,0.0013406753,0.00018729681,0.06845386,0.8009514,0.00031030612],"about_ca_topic_score_codex":0.035699304,"about_ca_topic_score_gemma":0.020178638,"teacher_disagreement_score":0.650208,"about_ca_system_score_codex":0.022483548,"about_ca_system_score_gemma":0.10205074,"threshold_uncertainty_score":0.8018222},"labels":[],"label_agreement":null},{"id":"W2102520493","doi":"10.1002/sim.2328","title":"A comparison of propensity score methods: a case‐study estimating the effectiveness of post‐AMI statin use","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":565,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Propensity score matching; Statin; Statistics; Medicine; Internal medicine; Computer science; Mathematics","score_opus":0.31956570844757465,"score_gpt":0.5520056064708857,"score_spread":0.23243989802331105,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2102520493","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6975621,0.0377024,0.24866512,0.005035453,0.0010272039,0.0038723464,0.0014047244,0.00017060655,0.0045599616],"genre_scores_gemma":[0.93658614,0.004597637,0.055002306,0.0003704702,0.0002762365,0.0020230748,0.00049367937,0.00004849537,0.0006019714],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.7997295,0.18372197,0.0059872465,0.0035064,0.0062885564,0.00076621916],"domain_scores_gemma":[0.5786208,0.38148198,0.014774143,0.01982217,0.004505003,0.00079589646],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.16027315,0.0013869968,0.0029620416,0.0039393306,0.0009784806,0.0025346568,0.0025756927,0.004596793,0.0041961144],"category_scores_gemma":[0.31711867,0.0011565369,0.007926463,0.0028993094,0.0021425162,0.0027391997,0.0025308717,0.0020612099,0.00043433707],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.12084996,0.0068618553,0.31445304,0.006946105,0.07888829,0.0022134779,0.0024510033,0.06738662,0.00223261,0.07985225,0.005979885,0.31188497],"study_design_scores_gemma":[0.04950507,0.05221142,0.1831572,0.003986435,0.06093164,0.0064549493,0.0034745326,0.5229113,0.005480115,0.090080224,0.0206837,0.0011235134],"about_ca_topic_score_codex":0.002435663,"about_ca_topic_score_gemma":0.0013692948,"teacher_disagreement_score":0.16027315,"about_ca_system_score_codex":0.0017199396,"about_ca_system_score_gemma":0.0017386375,"threshold_uncertainty_score":0.84761554},"labels":[],"label_agreement":null},{"id":"W2102524079","doi":"10.1002/sim.4056","title":"The analysis of case cohort design in the presence of competing risks with application to estimate the risk of delayed cardiac toxicity among Hodgkin Lymphoma survivors","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Princess Margaret Cancer Centre; Centre for Advancing Health Outcomes; Public Health Ontario; University of Toronto; University Health Network","funders":"Canadian Institutes of Health Research","keywords":"Jackknife resampling; Statistics; Medicine; Cohort; Econometrics; Disease; Proportional hazards model; Hazard ratio; Lymphoma; Mathematics; Risk analysis (engineering); Internal medicine; Confidence interval","score_opus":0.04083622230011155,"score_gpt":0.40275318208260447,"score_spread":0.36191695978249294,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2102524079","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.026586253,0.00045288337,0.971923,0.00034018783,0.0000666179,0.0002516592,0.000074176874,0.00006763401,0.00023758636],"genre_scores_gemma":[0.37722775,0.00103832,0.6177544,0.00032865064,0.00018913396,0.0018164878,0.00021023332,0.000063044565,0.0013719641],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9467976,0.04874721,0.00064781756,0.0015203228,0.0019552512,0.00033177494],"domain_scores_gemma":[0.79923385,0.18501484,0.0058281724,0.0073217344,0.0019076931,0.00069370354],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08782921,0.0009064599,0.0019072753,0.001749414,0.0010132673,0.0013305176,0.0022904226,0.0019713608,0.0022736683],"category_scores_gemma":[0.14369254,0.0006815344,0.002669867,0.0017074236,0.0018001298,0.0013528244,0.0022176027,0.0024838736,0.0001494577],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018363012,0.00056653365,0.06315224,0.000836794,0.003672026,0.0026849515,0.001696047,0.44935122,0.0029304852,0.2833282,0.002314637,0.18763046],"study_design_scores_gemma":[0.00024860416,0.00083585957,0.005449587,0.00006217077,0.00037331932,0.0006684511,0.00013210949,0.8964298,0.00077685056,0.092487015,0.0024453327,0.00009088442],"about_ca_topic_score_codex":0.0030841655,"about_ca_topic_score_gemma":0.002557329,"teacher_disagreement_score":0.08782921,"about_ca_system_score_codex":0.0011122349,"about_ca_system_score_gemma":0.0022080368,"threshold_uncertainty_score":0.46449083},"labels":[],"label_agreement":null},{"id":"W2103338349","doi":"10.1002/sim.5683","title":"Ties between event times and jump times in the Cox model","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Guelph","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Covariate; Event (particle physics); Computer science; Proportional hazards model; Software; Jump; Statistics; Binary data; Event data; Econometrics; Binary number; Mathematics; Machine learning","score_opus":0.09058773425645432,"score_gpt":0.42891557197893215,"score_spread":0.3383278377224778,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2103338349","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015110779,0.00056548894,0.9814626,0.00076427293,0.00010197149,0.000101884005,0.00018801812,0.00027882814,0.0014262575],"genre_scores_gemma":[0.6340096,0.0022964196,0.3451422,0.00085805985,0.001049791,0.0015602628,0.0012278077,0.0005701337,0.01328562],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98027724,0.011467171,0.0012186333,0.0035760878,0.0026841897,0.0007766686],"domain_scores_gemma":[0.82515496,0.1547672,0.006912835,0.010549956,0.0014068125,0.0012081364],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.048803847,0.0009537321,0.0018385015,0.0026630256,0.0019962604,0.0041880626,0.0041463245,0.0032881282,0.010254857],"category_scores_gemma":[0.2082643,0.001792928,0.0029116066,0.002993182,0.0040033115,0.011025404,0.00506032,0.007200295,0.0012168335],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00074920245,0.00011017083,0.019909957,0.00040954826,0.00041288603,0.001071753,0.0020933107,0.10003149,0.0007827768,0.7603968,0.003405917,0.11062633],"study_design_scores_gemma":[0.00012653926,0.0001123918,0.002791739,0.00012782286,0.000235329,0.00048309178,0.00014754375,0.3151777,0.0007078845,0.6743975,0.0055960743,0.00009636612],"about_ca_topic_score_codex":0.0022610785,"about_ca_topic_score_gemma":0.0018936448,"teacher_disagreement_score":0.048803847,"about_ca_system_score_codex":0.0016373285,"about_ca_system_score_gemma":0.0015703457,"threshold_uncertainty_score":0.25810254},"labels":[],"label_agreement":null},{"id":"W2103371988","doi":"10.1002/sim.5705","title":"The performance of different propensity score methods for estimating marginal hazard ratios","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":995,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Work & Health; Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Covariate; Statistics; Odds ratio; Estimator; Matching (statistics); Medicine; Mathematics","score_opus":0.22020753118718697,"score_gpt":0.4904911441960751,"score_spread":0.27028361300888815,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2103371988","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01159006,0.0024820564,0.982255,0.00056120637,0.00014961678,0.00073941465,0.0005586121,0.0005693061,0.0010946308],"genre_scores_gemma":[0.14935502,0.003587149,0.8403804,0.0004836649,0.0002539513,0.0024482338,0.0017602328,0.00059194735,0.0011394751],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9445329,0.0427882,0.0035351135,0.0037981137,0.0048376773,0.00050803693],"domain_scores_gemma":[0.8561339,0.113713674,0.00880006,0.012644682,0.008081779,0.0006259437],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08822343,0.0021359937,0.0026043802,0.006001812,0.0010824611,0.0032505195,0.003723077,0.0019756246,0.0049142726],"category_scores_gemma":[0.25330463,0.0012780768,0.0054498767,0.0072197267,0.0017110875,0.0040718545,0.0029290218,0.003688971,0.0011512208],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012503407,0.00030038226,0.061745383,0.0020094574,0.006442712,0.00027854647,0.001276182,0.20269658,0.0014115591,0.109685615,0.010446728,0.60245657],"study_design_scores_gemma":[0.00076576904,0.00059238414,0.026055261,0.0010103391,0.0015148267,0.000585917,0.0003417928,0.788509,0.003236327,0.15512212,0.021806056,0.00046029844],"about_ca_topic_score_codex":0.008023952,"about_ca_topic_score_gemma":0.006251296,"teacher_disagreement_score":0.91177654,"about_ca_system_score_codex":0.0019360386,"about_ca_system_score_gemma":0.0034830058,"threshold_uncertainty_score":0.46657568},"labels":[],"label_agreement":null},{"id":"W2104787905","doi":"10.1002/sim.3447","title":"Empirical study of the dependence of the results of multivariable flexible survival analyses on model selection strategy","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Colorectal Cancer Screening and Detection","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"McGill University","keywords":"Covariate; Parametric statistics; Proportional hazards model; Econometrics; Statistics; Selection (genetic algorithm); Model selection; Log-linear model; Semiparametric model; Parametric model; Computer science; Mathematics; Linear model; Artificial intelligence","score_opus":0.1659354445021612,"score_gpt":0.43033108846376195,"score_spread":0.26439564396160076,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2104787905","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8307216,0.001844317,0.16335176,0.00085344946,0.000035507477,0.00023931678,0.00041187197,0.0003674432,0.0021746252],"genre_scores_gemma":[0.9826402,0.00020759404,0.016186425,0.000071657036,0.000016471155,0.00009374717,0.0004919783,0.000101952566,0.00018990875],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8944495,0.09300955,0.0030617393,0.0038842345,0.004550429,0.0010445405],"domain_scores_gemma":[0.1592481,0.81329685,0.00911846,0.014179529,0.0035370963,0.00062005204],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.18706037,0.0013074091,0.0011175595,0.0019234675,0.0006177627,0.0016755867,0.001608742,0.0016464392,0.001487193],"category_scores_gemma":[0.47026882,0.00043549936,0.00247003,0.0022469864,0.0019338883,0.00256826,0.0021608472,0.0022988587,0.0002826079],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0025177395,0.00047261312,0.57175815,0.0005895226,0.004585254,0.0010672759,0.0022133754,0.2805451,0.002893954,0.0075203106,0.0014379582,0.12439881],"study_design_scores_gemma":[0.00023385054,0.0026874687,0.20268357,0.00030289954,0.0010531473,0.0009914039,0.0007764807,0.7751597,0.004459621,0.009986676,0.0015287295,0.00013642099],"about_ca_topic_score_codex":0.0015030075,"about_ca_topic_score_gemma":0.00094586547,"teacher_disagreement_score":0.18706037,"about_ca_system_score_codex":0.00076993246,"about_ca_system_score_gemma":0.0010026882,"threshold_uncertainty_score":0.9892816},"labels":[],"label_agreement":null},{"id":"W2105005539","doi":"10.1002/sim.2341","title":"Curious phenomena in Bayesian adjustment for exposure misclassification","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Intuition; Bayesian probability; Confounding; Econometrics; Bayes' theorem; Computer science; Statistics; Psychology; Artificial intelligence; Mathematics","score_opus":0.06258520134647907,"score_gpt":0.3989627422616326,"score_spread":0.3363775409151535,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2105005539","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0033293383,0.0031752652,0.96646434,0.02121236,0.0006724476,0.00012307789,0.0000887749,0.00025263627,0.004681724],"genre_scores_gemma":[0.2963923,0.005924355,0.6668184,0.017117579,0.006371098,0.0017824138,0.00022152215,0.0006895207,0.0046827816],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8680648,0.100589916,0.005375737,0.010747364,0.013655375,0.0015666757],"domain_scores_gemma":[0.4749541,0.4708757,0.015241773,0.029658092,0.008316566,0.0009537268],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2595974,0.0018116082,0.0040250695,0.0037422387,0.0031734942,0.0058316337,0.0059406585,0.007617914,0.003956963],"category_scores_gemma":[0.60774523,0.0019480758,0.0032309531,0.0040618484,0.020874297,0.018987106,0.007434063,0.02078804,0.000928116],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000060184873,0.00001687137,0.0012910748,0.00022775352,0.0001697386,0.00018646546,0.00089297007,0.006865543,0.000112918555,0.9575711,0.0046805963,0.027924834],"study_design_scores_gemma":[0.000030436093,0.000016233294,0.00030830994,0.0000977642,0.000025443358,0.000104921586,0.000035516066,0.01261108,0.00012344628,0.98347175,0.0031490289,0.000026176893],"about_ca_topic_score_codex":0.0039798347,"about_ca_topic_score_gemma":0.0018028239,"teacher_disagreement_score":0.2595974,"about_ca_system_score_codex":0.0037487298,"about_ca_system_score_gemma":0.0025479815,"threshold_uncertainty_score":0.91304827},"labels":[],"label_agreement":null},{"id":"W2105385152","doi":"10.1002/sim.5676","title":"Comparing diagnostic tests: trials in people with discordant test results","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Women's Health Research Institute","funders":"National Institutes of Health; Universiteit van Amsterdam; National Institute for Health and Care Research","keywords":"Statistics; Test (biology); Gold standard (test); Statistical hypothesis testing; Computer science; Medicine; Econometrics; Mathematics","score_opus":0.5460293344991152,"score_gpt":0.5771006422312617,"score_spread":0.031071307732146503,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2105385152","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.052680038,0.08198036,0.7960043,0.013363783,0.008108266,0.03228301,0.0015177871,0.00087446,0.013187952],"genre_scores_gemma":[0.42410976,0.014699395,0.49072412,0.008703706,0.0022157957,0.05660402,0.00082150125,0.00028752885,0.0018341924],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.38728717,0.5572218,0.028948842,0.00972114,0.016011136,0.000809915],"domain_scores_gemma":[0.28833264,0.65018475,0.031721726,0.022947913,0.005231275,0.0015817746],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.31341723,0.0023247346,0.0078798635,0.004662482,0.0010544248,0.006023927,0.004303317,0.010215141,0.005702175],"category_scores_gemma":[0.6904366,0.0013721981,0.0064015095,0.0046940795,0.006992099,0.00963651,0.0047070743,0.0050739627,0.0011377698],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.055560935,0.001442747,0.014646282,0.06922781,0.044084895,0.0012871241,0.0029749621,0.015978944,0.0024611119,0.3544712,0.013244764,0.4246192],"study_design_scores_gemma":[0.034621038,0.01577806,0.0058064125,0.027854107,0.020458952,0.0027113853,0.0008652907,0.050069902,0.0066242353,0.7821897,0.05239617,0.000624693],"about_ca_topic_score_codex":0.0002889247,"about_ca_topic_score_gemma":0.00022120411,"teacher_disagreement_score":0.6865828,"about_ca_system_score_codex":0.0026009479,"about_ca_system_score_gemma":0.0038806447,"threshold_uncertainty_score":0.8466788},"labels":[],"label_agreement":null},{"id":"W2105409485","doi":"10.1002/sim.5592","title":"Generalization of Youden index for multiple‐class classification problems applied to the assessment of externally validated cognition in Parkinson disease screening","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Parkinson's Disease Mechanisms and Treatments","field":"Medicine","cited_by":40,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Generalization; Youden's J statistic; Cognition; Dementia; Class (philosophy); Index (typography); Disease; Medicine; Psychology; Artificial intelligence; Computer science; Mathematics; Machine learning; Psychiatry; Pathology; Receiver operating characteristic","score_opus":0.05750702138733923,"score_gpt":0.35622437236500176,"score_spread":0.29871735097766255,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2105409485","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.044616926,0.000980457,0.9516799,0.00038534467,0.00014757963,0.00026297322,0.0002091172,0.00016864252,0.0015491258],"genre_scores_gemma":[0.404039,0.0010308519,0.5908504,0.00024372365,0.00040409647,0.0011661359,0.0008155643,0.00012892994,0.001321177],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9821317,0.008811913,0.0020736943,0.0021909259,0.0042965366,0.00049516227],"domain_scores_gemma":[0.91829115,0.062401444,0.005559106,0.0067860996,0.0060536386,0.00090857595],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03689119,0.0011792773,0.0020017293,0.00458967,0.0009963372,0.003155299,0.002198208,0.0021512315,0.0015661654],"category_scores_gemma":[0.120231174,0.00049468246,0.0018693579,0.0037561455,0.0034145182,0.004824295,0.0034930992,0.0036013077,0.00046316042],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00087634206,0.00034393178,0.05220046,0.00074971805,0.0010511613,0.000715777,0.0016222723,0.21444176,0.0056792325,0.31372002,0.0076182955,0.40098098],"study_design_scores_gemma":[0.000058896854,0.00041046718,0.017101105,0.00017149371,0.000101290105,0.0006726399,0.0002716866,0.75303304,0.0019439681,0.2213849,0.0046763364,0.0001742226],"about_ca_topic_score_codex":0.0019334668,"about_ca_topic_score_gemma":0.0013802876,"teacher_disagreement_score":0.03689119,"about_ca_system_score_codex":0.0016865154,"about_ca_system_score_gemma":0.0018042,"threshold_uncertainty_score":0.19510162},"labels":[],"label_agreement":null},{"id":"W2105683190","doi":"10.1002/sim.1936","title":"Statistical methods for multivariate interval‐censored recurrent events","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":49,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Covariate; Marginal model; Statistics; Multivariate statistics; Inference; Covariance; Event (particle physics); Econometrics; Computer science; Regression; Regression analysis; Mathematics; Artificial intelligence","score_opus":0.151138328609492,"score_gpt":0.5335458073367135,"score_spread":0.38240747872722153,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2105683190","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00039535903,0.00033370592,0.9984054,0.00008279643,0.00005313667,0.000113909075,0.00014887349,0.00028091343,0.00018589551],"genre_scores_gemma":[0.04229425,0.0018277126,0.948253,0.0003026949,0.00048825305,0.0034681777,0.0013173997,0.00044247377,0.0016059916],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9663363,0.025364978,0.0015103755,0.00270683,0.0036914358,0.00039011054],"domain_scores_gemma":[0.8164375,0.16057462,0.0073905853,0.01044838,0.0045520915,0.0005967683],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03820614,0.0019840866,0.0031125867,0.005222283,0.00076759985,0.0026798244,0.0057863602,0.00232633,0.007991381],"category_scores_gemma":[0.1401899,0.0011525545,0.0036143865,0.0059676585,0.0026505827,0.0036368282,0.0033051008,0.0076368926,0.0016780244],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002860756,0.00019845058,0.005857529,0.0013184804,0.001380264,0.0005175094,0.00048709044,0.20912907,0.0014634224,0.5071949,0.010644179,0.26152307],"study_design_scores_gemma":[0.00012121736,0.00016816158,0.0018093098,0.00026254586,0.00014516606,0.00026614274,0.000078524645,0.5996788,0.0006130856,0.3841939,0.012557318,0.00010585388],"about_ca_topic_score_codex":0.002837118,"about_ca_topic_score_gemma":0.0017477284,"teacher_disagreement_score":0.03820614,"about_ca_system_score_codex":0.001682828,"about_ca_system_score_gemma":0.0030555094,"threshold_uncertainty_score":0.20205581},"labels":[],"label_agreement":null},{"id":"W2105829241","doi":"10.1002/sim.2157","title":"Variance estimation for clustered recurrent event data with a small number of clusters","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"McMaster University","keywords":"Jackknife resampling; Estimator; Statistics; Variance (accounting); Mathematics; Parametric statistics; Econometrics; Computer science","score_opus":0.14738161384038687,"score_gpt":0.4531564777925362,"score_spread":0.30577486395214937,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2105829241","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0061222212,0.00018383504,0.9931744,0.00007558496,0.00003032646,0.00007210503,0.000066499604,0.00015334145,0.00012168357],"genre_scores_gemma":[0.2916442,0.00046793924,0.7040198,0.00023727944,0.0001676387,0.0010113061,0.0011129562,0.00021368542,0.001125143],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97178835,0.02090256,0.0013065725,0.003767789,0.0018285682,0.00040618514],"domain_scores_gemma":[0.8377626,0.1329664,0.007527637,0.016966462,0.004307688,0.00046924708],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.052996535,0.0010029691,0.0024791886,0.0026073821,0.0007292496,0.0015865624,0.0037044834,0.0020672155,0.0014473695],"category_scores_gemma":[0.20345916,0.0008040077,0.0022084173,0.002654512,0.0021812308,0.0022934251,0.0019912706,0.0024555111,0.000420253],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011003086,0.00024943415,0.034902174,0.0011563667,0.0028605447,0.000864143,0.0015407614,0.36269093,0.0046077496,0.25138283,0.0065419786,0.33210272],"study_design_scores_gemma":[0.00015836734,0.00030968964,0.008077697,0.00015885093,0.00025541685,0.0003480246,0.00015761172,0.791499,0.002609071,0.19214042,0.0041789147,0.000106986285],"about_ca_topic_score_codex":0.002685116,"about_ca_topic_score_gemma":0.0020620546,"teacher_disagreement_score":0.052996535,"about_ca_system_score_codex":0.0009158328,"about_ca_system_score_gemma":0.0015084924,"threshold_uncertainty_score":0.28027582},"labels":[],"label_agreement":null},{"id":"W2105884242","doi":"10.1002/sim.5687","title":"A semiparametric marginal mixture cure model for clustered survival data","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Statistics; Semiparametric regression; Expectation–maximization algorithm; Proportional hazards model; Mathematics; Estimating equations; Regression; Marginal model; Generalized estimating equation; Mixture model; Econometrics; Maximization; Semiparametric model; Data set; Regression analysis; Computer science; Applied mathematics; Mathematical optimization; Nonparametric statistics; Estimator; Maximum likelihood","score_opus":0.2795852382993958,"score_gpt":0.4776737204319478,"score_spread":0.19808848213255198,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2105884242","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011688107,0.00014551078,0.9873219,0.00019510361,0.00001795805,0.000041441825,0.00015665866,0.00016061694,0.00027268156],"genre_scores_gemma":[0.597548,0.0007925073,0.3921796,0.0003602833,0.00021431444,0.0009634066,0.001670429,0.00025569677,0.0060157413],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9967062,0.0018880923,0.00011606978,0.0005843991,0.00050302956,0.00020216382],"domain_scores_gemma":[0.9911266,0.005748182,0.001054861,0.001137618,0.00070093566,0.0002318146],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008018852,0.0009956848,0.0017729257,0.0019554612,0.00047535176,0.0014784898,0.0033599937,0.0015183656,0.0033931986],"category_scores_gemma":[0.021326514,0.0007451469,0.0024084672,0.0017247752,0.0015931899,0.001916822,0.002013447,0.002437935,0.00089647324],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000290762,0.0001842204,0.010368783,0.00032528135,0.00045580763,0.0003865332,0.00073493185,0.5910221,0.003028069,0.30653465,0.0034138337,0.08325503],"study_design_scores_gemma":[0.000020620686,0.00007167277,0.0012466377,0.000022982329,0.000058409252,0.00012614884,0.00003690901,0.9328151,0.00032518254,0.06359933,0.0016355666,0.00004152955],"about_ca_topic_score_codex":0.0028166038,"about_ca_topic_score_gemma":0.0026325176,"teacher_disagreement_score":0.008018852,"about_ca_system_score_codex":0.00086321874,"about_ca_system_score_gemma":0.0011424937,"threshold_uncertainty_score":0.042408228},"labels":[],"label_agreement":null},{"id":"W2106169817","doi":"10.1002/sim.4394","title":"A longitudinal model for magnetic resonance imaging lesion count data in multiple sclerosis patients","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Multiple Sclerosis Research Studies","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; Simon Fraser University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Magnetic resonance imaging; Nonparametric statistics; Computer science; Parametric statistics; Statistical power; Markov chain; Statistics; Data mining; Medical physics; Artificial intelligence; Pattern recognition (psychology); Machine learning; Medicine; Mathematics; Radiology","score_opus":0.2889380930924304,"score_gpt":0.3821583364372672,"score_spread":0.0932202433448368,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2106169817","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3945425,0.0013886517,0.5795272,0.00976317,0.0003355221,0.0012420353,0.008606188,0.00086874247,0.003725994],"genre_scores_gemma":[0.9030835,0.001143851,0.07364209,0.00090528093,0.00024283509,0.002855285,0.0043844776,0.000073610354,0.013669021],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.993487,0.00442368,0.00023723459,0.00095197896,0.00042531325,0.00047484285],"domain_scores_gemma":[0.9685005,0.025863882,0.0025662642,0.0012487344,0.0012212747,0.0005992461],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.021302408,0.0012538902,0.0020315952,0.0018707188,0.00070593384,0.0020494654,0.0031046972,0.0034597341,0.0093735885],"category_scores_gemma":[0.030043356,0.0008165182,0.0020743643,0.0016717424,0.0017532828,0.002317676,0.0014564104,0.0031879188,0.0013334157],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0026678292,0.00059381657,0.053363632,0.00030407854,0.00067533937,0.0010355781,0.0011699338,0.78167504,0.0016513421,0.120756164,0.0046732663,0.03143399],"study_design_scores_gemma":[0.00050664385,0.00047382383,0.0057583326,0.000051464907,0.00016076674,0.00017833871,0.000087806555,0.96444285,0.00021961177,0.026416294,0.0016297206,0.00007439242],"about_ca_topic_score_codex":0.013072372,"about_ca_topic_score_gemma":0.009596387,"teacher_disagreement_score":0.021302408,"about_ca_system_score_codex":0.0021192147,"about_ca_system_score_gemma":0.0025929457,"threshold_uncertainty_score":0.112659216},"labels":[],"label_agreement":null},{"id":"W2107476951","doi":"10.1002/sim.1764","title":"Test of treatment effect in pre‐drug and post‐drug count data with zero‐inflation","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Windsor","funders":"National Cancer Institute","keywords":"Likelihood-ratio test; Score test; Statistics; Statistic; Mathematics; Test statistic; Null hypothesis; Medicine; Statistical hypothesis testing","score_opus":0.21690093327361057,"score_gpt":0.5248629691353153,"score_spread":0.3079620358617047,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2107476951","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1612053,0.0009169151,0.8231345,0.0011650998,0.00039931602,0.0051909783,0.002775261,0.0011325121,0.004080126],"genre_scores_gemma":[0.69726,0.00033036925,0.28398493,0.00072283356,0.00025020895,0.011789129,0.0029937269,0.00022401566,0.0024446964],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.82740813,0.13945986,0.0061491877,0.0062206667,0.01926454,0.0014976187],"domain_scores_gemma":[0.44678593,0.51549625,0.01216238,0.018905675,0.005255198,0.0013945169],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.13216422,0.001342929,0.0042784484,0.0038065694,0.0007690185,0.0014165269,0.0024123553,0.0029854309,0.009959976],"category_scores_gemma":[0.36761215,0.0006541599,0.0055483514,0.0034117685,0.0035095334,0.002332519,0.0022871818,0.0038625435,0.001158051],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.03854946,0.0025809768,0.1271246,0.0033752902,0.017550377,0.0016394034,0.0018766767,0.054512855,0.011994647,0.088618055,0.009923463,0.6422542],"study_design_scores_gemma":[0.004803573,0.05005716,0.22719048,0.00048772828,0.0034633814,0.0021991218,0.00074470043,0.5292982,0.018379265,0.1432902,0.019471372,0.0006148095],"about_ca_topic_score_codex":0.0006826131,"about_ca_topic_score_gemma":0.00041980876,"teacher_disagreement_score":0.86783576,"about_ca_system_score_codex":0.0008722441,"about_ca_system_score_gemma":0.0026179687,"threshold_uncertainty_score":0.6989596},"labels":[],"label_agreement":null},{"id":"W2107724629","doi":"10.1002/sim.1095","title":"Detecting and eliminating erroneous gestational ages: a normal mixture model","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Birth, Development, and Health","field":"Medicine","cited_by":61,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Canada; Dalhousie University; McGill University","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Gestational age; Birth weight; Medicine; Obstetrics; Population; Small for gestational age; Gestation; Pregnancy","score_opus":0.030349878838301425,"score_gpt":0.3359374601667724,"score_spread":0.30558758132847097,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2107724629","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017084494,0.00038898352,0.98086643,0.00047336423,0.00008864481,0.000050831182,0.00015114927,0.0005854166,0.00031074983],"genre_scores_gemma":[0.42647088,0.0011324439,0.56342,0.0006312992,0.00040954677,0.0004826418,0.0017119717,0.00034949323,0.005391756],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99475056,0.0027628965,0.00027562497,0.0013180268,0.00055106054,0.00034187126],"domain_scores_gemma":[0.9773377,0.018450916,0.00100375,0.0013840888,0.0016254509,0.00019815382],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01500227,0.0018506426,0.002995451,0.0028316192,0.0012108393,0.002227119,0.004718076,0.0033912868,0.001287734],"category_scores_gemma":[0.042997934,0.0014802903,0.0024116314,0.0023884566,0.002793015,0.0035234538,0.0027986052,0.0037555117,0.0011030713],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00088241516,0.00024588103,0.012399235,0.0001508125,0.00046452167,0.00025913992,0.00048117945,0.7067861,0.0011858894,0.033542585,0.004000938,0.23960134],"study_design_scores_gemma":[0.000045025357,0.000033119828,0.0012204217,0.000018273298,0.00005707251,0.00007568086,0.000018794903,0.97981215,0.00038600405,0.017566944,0.00072779943,0.000038720784],"about_ca_topic_score_codex":0.018753877,"about_ca_topic_score_gemma":0.011613465,"teacher_disagreement_score":0.018753877,"about_ca_system_score_codex":0.0013882244,"about_ca_system_score_gemma":0.0018110701,"threshold_uncertainty_score":0.07934052},"labels":[],"label_agreement":null},{"id":"W2108268423","doi":"10.1002/sim.745","title":"Logistic discrimination of mixtures of M. tuberculosis and non‐specific tuberculin reactions","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"International Union Against Tuberculosis and Lung Disease","keywords":"Homogeneity (statistics); Tuberculin; Tuberculosis; Logistic regression; Goodness of fit; Tuberculin test; Medicine; Statistics; Mathematics; Pathology","score_opus":0.03422861188760234,"score_gpt":0.32867853000775166,"score_spread":0.2944499181201493,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2108268423","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.40948817,0.0008657936,0.58682656,0.00036817184,0.00006348448,0.00014291829,0.00017597953,0.000616014,0.0014529345],"genre_scores_gemma":[0.9323055,0.00037715214,0.06547045,0.000058246107,0.000065886336,0.000103606224,0.000398139,0.000064466665,0.0011565388],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9952742,0.0030569073,0.00024505053,0.0006107532,0.0005266275,0.00028642372],"domain_scores_gemma":[0.9585288,0.03496301,0.0034715475,0.0016182362,0.0010076362,0.0004108359],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012576946,0.0009800354,0.0014410918,0.002822229,0.0004587139,0.0018031825,0.001596961,0.0013160269,0.0015816507],"category_scores_gemma":[0.056105547,0.0011450432,0.001444828,0.0013512475,0.0013791445,0.0018962937,0.0022955202,0.001538692,0.00066946656],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.005270719,0.00042679632,0.3409592,0.0006073602,0.0010369816,0.0021776827,0.0018944357,0.22195931,0.016350525,0.059806403,0.0030202447,0.34649032],"study_design_scores_gemma":[0.000070238064,0.0001418162,0.030040521,0.000042044387,0.000115661926,0.001105159,0.00015061643,0.92429465,0.0026787939,0.04023002,0.001033097,0.000097490694],"about_ca_topic_score_codex":0.0010999837,"about_ca_topic_score_gemma":0.0011043508,"teacher_disagreement_score":0.012576946,"about_ca_system_score_codex":0.00063535414,"about_ca_system_score_gemma":0.0004733724,"threshold_uncertainty_score":0.066514075},"labels":[],"label_agreement":null},{"id":"W2108304037","doi":"10.1002/sim.3335","title":"Nonparametric statistical inference method for partial areas under receiver operating characteristic curves, with application to genomic studies","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic and phenotypic traits in livestock","field":"Biochemistry, Genetics and Molecular Biology","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto","funders":"","keywords":"Receiver operating characteristic; Nonparametric statistics; Confidence interval; Statistics; Statistical inference; Inference; Statistical hypothesis testing; Asymptotic distribution; Statistic; False positive paradox; Sample size determination; Mathematics; Computer science; Normality; Artificial intelligence; Estimator","score_opus":0.03394799462728353,"score_gpt":0.3616748351351232,"score_spread":0.3277268405078397,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2108304037","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00084370654,0.00026805318,0.9978967,0.00008280177,0.000041735882,0.00005353625,0.00007432382,0.0004609709,0.00027824708],"genre_scores_gemma":[0.06604593,0.0007046777,0.92991006,0.0002317272,0.00027817953,0.0010229844,0.0004515765,0.00039426395,0.0009605627],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97870296,0.014827352,0.000984919,0.0022767643,0.002978974,0.00022899377],"domain_scores_gemma":[0.9221841,0.06349982,0.0040882933,0.0047964067,0.0050733592,0.00035796122],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026238412,0.0010626748,0.0018097674,0.0054909354,0.0006871855,0.0019080598,0.0021262446,0.0016432562,0.0032859582],"category_scores_gemma":[0.11611523,0.00059035723,0.0020465828,0.0037530595,0.002022132,0.0022242914,0.0016332904,0.0037035986,0.0012873117],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003207304,0.00016890574,0.00822063,0.0014761544,0.00097222277,0.00061539584,0.00081429497,0.06724455,0.0059194653,0.16009073,0.013799945,0.74035686],"study_design_scores_gemma":[0.00013744147,0.0005623375,0.010594741,0.00036149344,0.0003295039,0.0021433828,0.00021562325,0.69673884,0.0076455036,0.24808349,0.03291696,0.00027070462],"about_ca_topic_score_codex":0.0013414787,"about_ca_topic_score_gemma":0.00089624693,"teacher_disagreement_score":0.026238412,"about_ca_system_score_codex":0.0011040022,"about_ca_system_score_gemma":0.002282967,"threshold_uncertainty_score":0.13876367},"labels":[],"label_agreement":null},{"id":"W2108885350","doi":"10.1002/sim.1107","title":"A new approach to training back‐propagation artificial neural networks: empirical evaluation on ten data sets from clinical studies","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Neural Networks and Applications","field":"Computer Science","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Initialization; Artificial neural network; Backpropagation; Computer science; Logistic regression; Artificial intelligence; Deviance (statistics); Training set; Machine learning; Regularization (linguistics); Data mining; Statistics; Mathematics","score_opus":0.5224149863755513,"score_gpt":0.4928430757635104,"score_spread":0.029571910612040897,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2108885350","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.78252465,0.003924978,0.20811352,0.0008795253,0.00029689813,0.0005227211,0.0009282022,0.00096468715,0.0018448049],"genre_scores_gemma":[0.84646493,0.00091805664,0.1493164,0.00018249896,0.00009137378,0.0004433646,0.0018006951,0.00008534763,0.0006973937],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.994079,0.0040643034,0.00047804415,0.00040536813,0.000880519,0.00009279257],"domain_scores_gemma":[0.96191174,0.03081494,0.001098186,0.0021255217,0.0037106755,0.00033894533],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020385373,0.0013889015,0.0009974227,0.0016562784,0.0003815106,0.0009108424,0.0012974358,0.0013343933,0.0007042006],"category_scores_gemma":[0.048876483,0.00035051594,0.0006648679,0.0013995516,0.00075167627,0.00092510815,0.0010902415,0.0015181986,0.00029812468],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002137031,0.0013423132,0.10580238,0.001115985,0.0015739929,0.00045254387,0.0005068971,0.3384116,0.0038190803,0.0009664296,0.004092784,0.53977895],"study_design_scores_gemma":[0.0002068375,0.0013037889,0.022324188,0.00018374485,0.00027745334,0.0003064122,0.00015453137,0.96767676,0.0045333593,0.0014640228,0.0015043982,0.00006449472],"about_ca_topic_score_codex":0.0031320103,"about_ca_topic_score_gemma":0.004541339,"teacher_disagreement_score":0.020385373,"about_ca_system_score_codex":0.00051598373,"about_ca_system_score_gemma":0.00074477954,"threshold_uncertainty_score":0.107809424},"labels":[],"label_agreement":null},{"id":"W2109920122","doi":"10.1002/sim.1261","title":"Exact unconditional tests for testing non‐inferiority in matched‐pairs design","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Women's Health Research Institute","funders":"","keywords":"Nuisance parameter; Exact statistics; Mathematics; Confidence interval; Sample size determination; Statistics; Monotonic function; Statistical hypothesis testing; Nominal level; Exact test; Score test; Applied mathematics; Estimator; Mathematical analysis","score_opus":0.6511081763640518,"score_gpt":0.5669361148639607,"score_spread":0.08417206150009116,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2109920122","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.020614024,0.0012273465,0.9722477,0.00051792286,0.00028631755,0.0010487824,0.00026259324,0.00033453712,0.003460665],"genre_scores_gemma":[0.3869149,0.001032697,0.60238564,0.0008970987,0.00057587266,0.006592469,0.0004329683,0.0000942023,0.0010741137],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.765193,0.19683912,0.0068146316,0.006494014,0.023349961,0.0013092842],"domain_scores_gemma":[0.46359286,0.49623114,0.014297937,0.018856114,0.006048385,0.00097356184],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.13398875,0.001201972,0.0050151143,0.0037622524,0.0010287919,0.002014786,0.003722904,0.0029761798,0.009897077],"category_scores_gemma":[0.44775963,0.00072854076,0.0022164688,0.0028191041,0.0044922107,0.0048954166,0.0033342692,0.0037380827,0.0010321961],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0053741187,0.0006501884,0.016952759,0.0033214109,0.0025873873,0.0010897228,0.0009687844,0.03406586,0.0022970994,0.41205245,0.006507371,0.5141328],"study_design_scores_gemma":[0.0018200463,0.0077371523,0.013908837,0.00068693794,0.0009233108,0.0020701142,0.00038163,0.24217437,0.0058492892,0.71376294,0.01039236,0.00029310441],"about_ca_topic_score_codex":0.00020289542,"about_ca_topic_score_gemma":0.00021308636,"teacher_disagreement_score":0.13398875,"about_ca_system_score_codex":0.0013560414,"about_ca_system_score_gemma":0.0027074842,"threshold_uncertainty_score":0.70860875},"labels":[],"label_agreement":null},{"id":"W2110136166","doi":"10.1002/sim.6563","title":"Testing for treatment‐biomarker interaction based on local partial‐likelihood","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada; Compute Canada","keywords":"Biomarker; Computer science; Statistics; Econometrics; Mathematics; Biology; Genetics","score_opus":0.2507980355314263,"score_gpt":0.46878566211708883,"score_spread":0.2179876265856625,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2110136166","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.21748453,0.0015107127,0.77540076,0.0016783309,0.00008316962,0.000425445,0.0008631645,0.00069833925,0.001855591],"genre_scores_gemma":[0.9428748,0.00028330582,0.05422861,0.00043721616,0.000106895706,0.0008442943,0.0006981003,0.00007745556,0.00044933154],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.93008196,0.058842953,0.0016899937,0.00458773,0.003904821,0.00089265074],"domain_scores_gemma":[0.5380873,0.4349751,0.01079684,0.012248457,0.0023996113,0.0014927415],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06502932,0.0010854525,0.004074334,0.0024944071,0.00082911184,0.0017732658,0.0026222416,0.0021883661,0.0039081634],"category_scores_gemma":[0.24836579,0.0006140472,0.0035027862,0.0022759107,0.0050616455,0.0034228878,0.0029424573,0.0029766227,0.00042442817],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009867716,0.0011382096,0.32828325,0.002018902,0.012515771,0.0030507809,0.0016388969,0.1900263,0.006387065,0.11221757,0.005111309,0.32774425],"study_design_scores_gemma":[0.00083258195,0.004206471,0.0621848,0.00018336943,0.0012271679,0.0012872881,0.00041394448,0.7003065,0.0048964247,0.22105779,0.003205111,0.0001985813],"about_ca_topic_score_codex":0.00090243074,"about_ca_topic_score_gemma":0.00055583904,"teacher_disagreement_score":0.06502932,"about_ca_system_score_codex":0.0008167089,"about_ca_system_score_gemma":0.0015837845,"threshold_uncertainty_score":0.34391207},"labels":[],"label_agreement":null},{"id":"W2110147390","doi":"10.1002/sim.1266","title":"Binary partitioning for continuous longitudinal data: categorizing a prognostic variable","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":67,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; Hospital for Sick Children; Population Health Research Institute; University of Toronto","funders":"","keywords":"Permutation (music); Statistics; Confidence interval; Statistic; Binary data; Binary number; Continuous variable; Variable (mathematics); Test statistic; Mathematics; Interval (graph theory); Set (abstract data type); Computer science; Data set; Statistical hypothesis testing; Combinatorics","score_opus":0.0849101194140717,"score_gpt":0.3404597345364128,"score_spread":0.2555496151223411,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2110147390","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.020269137,0.0006243,0.9771212,0.0005020038,0.00007882177,0.0002957831,0.00015175105,0.00018578324,0.00077110354],"genre_scores_gemma":[0.14024505,0.0004512775,0.85677916,0.00021628842,0.00013640668,0.0007353002,0.0006510678,0.00014252364,0.0006428546],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9861653,0.009091106,0.0005962022,0.0013592729,0.002384281,0.0004039908],"domain_scores_gemma":[0.9650263,0.02554611,0.003113193,0.003297059,0.002456916,0.0005604003],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02161017,0.0012121092,0.0015611737,0.0035388428,0.0012879794,0.0024248415,0.0029172504,0.0022676585,0.003138932],"category_scores_gemma":[0.081857644,0.0005248166,0.0012789194,0.0029912302,0.0021551638,0.0026700294,0.002155256,0.0021896814,0.0013794322],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013075335,0.0005825351,0.049634036,0.0011560273,0.00050268165,0.00034360396,0.001776875,0.06374829,0.008949335,0.15997459,0.0053534,0.7066712],"study_design_scores_gemma":[0.0003394743,0.001146005,0.019510267,0.0005072661,0.00020485702,0.0016794037,0.00071105006,0.55801094,0.009504715,0.39539158,0.012764049,0.00023038032],"about_ca_topic_score_codex":0.0011521061,"about_ca_topic_score_gemma":0.0011292155,"teacher_disagreement_score":0.02161017,"about_ca_system_score_codex":0.000881447,"about_ca_system_score_gemma":0.0019090935,"threshold_uncertainty_score":0.1142869},"labels":[],"label_agreement":null},{"id":"W2110182275","doi":"10.1002/sim.4087","title":"Copula‐based regression models for a bivariate mixed discrete and continuous outcome","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":132,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Copula (linguistics); Bivariate analysis; Covariate; Econometrics; Joint probability distribution; Marginal distribution; Marginal model; Regression; Mathematics; Statistics; Regression analysis; Computer science; Random variable","score_opus":0.08217114540050523,"score_gpt":0.43601078137216187,"score_spread":0.35383963597165663,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2110182275","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0037896873,0.0004864786,0.99407554,0.0003290734,0.00002791634,0.000044175056,0.00022147835,0.00016166101,0.00086402806],"genre_scores_gemma":[0.32430902,0.004573247,0.6531082,0.0006582709,0.00037651038,0.001664007,0.00182839,0.0005433173,0.012939053],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9925689,0.005068508,0.000233339,0.0011166921,0.0006553766,0.00035718112],"domain_scores_gemma":[0.98192364,0.013881222,0.0019859183,0.001149313,0.00083858665,0.0002212942],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013864906,0.0023888848,0.0030661311,0.0025644177,0.00075591396,0.0032552397,0.004616938,0.0025088857,0.00645716],"category_scores_gemma":[0.034969468,0.0014956689,0.0032981858,0.0038141408,0.0020368476,0.0041129086,0.0022050028,0.004532021,0.002203673],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006676352,0.00008600857,0.0028465292,0.00026780416,0.00038281942,0.00028804265,0.00042000084,0.35383767,0.0005931672,0.607613,0.0038178235,0.029780392],"study_design_scores_gemma":[0.00001827028,0.000037096936,0.0007433318,0.000048804268,0.00006897795,0.000105897045,0.000038115413,0.8478377,0.00014108683,0.14816687,0.0027577977,0.000036097306],"about_ca_topic_score_codex":0.008963105,"about_ca_topic_score_gemma":0.0063107396,"teacher_disagreement_score":0.013864906,"about_ca_system_score_codex":0.002066891,"about_ca_system_score_gemma":0.0017589587,"threshold_uncertainty_score":0.073325515},"labels":[],"label_agreement":null},{"id":"W2110786877","doi":"10.1002/sim.3444","title":"A Bayesian multilevel model for estimating the diet/disease relationship in a multicenter study with exposures measured with error: The EPIC study","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Nutritional Studies and Diet","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Cancer Institute","keywords":"European Prospective Investigation into Cancer and Nutrition; Breast cancer; Medicine; Gibbs sampling; Statistics; Multilevel model; Cancer; Incidence (geometry); Prospective cohort study; Disease; Bayesian probability; Demography; Oncology; Internal medicine; Mathematics","score_opus":0.0986514510617566,"score_gpt":0.3617802300035223,"score_spread":0.2631287789417657,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2110786877","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04164214,0.0012459005,0.95295775,0.0016163588,0.00010650443,0.0003270462,0.0010044008,0.0001807761,0.00091908785],"genre_scores_gemma":[0.4408959,0.0020123546,0.54560935,0.0010197961,0.00034994681,0.0029556053,0.002838936,0.00013458458,0.0041835126],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9803718,0.015623063,0.0007398914,0.0018414577,0.0009421032,0.00048167852],"domain_scores_gemma":[0.96875775,0.024322996,0.0026959316,0.0022345667,0.0014969616,0.0004918769],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.032702073,0.0012669928,0.0033693942,0.002396431,0.0011282412,0.0027264978,0.0061823023,0.0035451876,0.0032907783],"category_scores_gemma":[0.053784963,0.0016272182,0.0039743874,0.0026898333,0.0018848457,0.0024039447,0.0037600817,0.003924143,0.0005945601],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015196438,0.00029663127,0.07897966,0.0007498956,0.004363555,0.0013495398,0.0021377609,0.50874454,0.0021821829,0.30761153,0.005536802,0.08652824],"study_design_scores_gemma":[0.00042249876,0.00036380792,0.009282859,0.00025785793,0.0009295603,0.00025767623,0.00017850295,0.83667934,0.00025206318,0.14668761,0.004558911,0.00012928473],"about_ca_topic_score_codex":0.024391815,"about_ca_topic_score_gemma":0.019177662,"teacher_disagreement_score":0.032702073,"about_ca_system_score_codex":0.0017726144,"about_ca_system_score_gemma":0.0022392045,"threshold_uncertainty_score":0.17294717},"labels":[],"label_agreement":null},{"id":"W2110818436","doi":"10.1002/sim.6607","title":"Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies","year":2015,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":4309,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; National Institute of Mental Health; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Propensity score matching; Covariate; Inverse probability weighting; Observational study; Inverse probability; Weighting; Average treatment effect; Statistics; Treatment and control groups; Econometrics; Sample size determination; Baseline (sea); Mathematics; Selection bias; Computer science; Posterior probability; Medicine; Bayesian probability","score_opus":0.6765608631638398,"score_gpt":0.5702584500397571,"score_spread":0.10630241312408273,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2110818436","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011510972,0.011566644,0.9593448,0.024299862,0.0008635535,0.00033289206,0.00014521787,0.0001886445,0.0021072277],"genre_scores_gemma":[0.038703844,0.01061026,0.93793243,0.008251778,0.0014695745,0.0022595103,0.00017426923,0.0001940757,0.00040422703],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.5464783,0.39535573,0.020927593,0.012663006,0.023506425,0.0010689197],"domain_scores_gemma":[0.41138324,0.51226383,0.024449145,0.035749022,0.014978078,0.001176717],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3219493,0.0019577893,0.0045696697,0.009172669,0.001790489,0.010723759,0.007313782,0.0075355354,0.0028489665],"category_scores_gemma":[0.5747454,0.0020863882,0.0039077513,0.010450662,0.022694346,0.01663251,0.010227887,0.018560374,0.001250197],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009943253,0.00008329565,0.0029052375,0.0061633694,0.001226492,0.00017664388,0.0031357645,0.009738759,0.00032214788,0.7230092,0.010978963,0.24216083],"study_design_scores_gemma":[0.000079399935,0.00007738317,0.00053198554,0.0032035762,0.00013955709,0.0001094208,0.00023244694,0.008052526,0.00039390815,0.96894157,0.018168384,0.000069756825],"about_ca_topic_score_codex":0.0044023506,"about_ca_topic_score_gemma":0.003131632,"teacher_disagreement_score":0.6780507,"about_ca_system_score_codex":0.00563736,"about_ca_system_score_gemma":0.011039628,"threshold_uncertainty_score":0.8361572},"labels":[],"label_agreement":null},{"id":"W2110863526","doi":"10.1002/sim.2375","title":"Generalized gamma frailty model","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Frailty in Older Adults","field":"Medicine","cited_by":103,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Memorial University of Newfoundland; McMaster University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Weibull distribution; Angina; Isosorbide dinitrate; Mathematics; Monte Carlo method; Generalization; Computer science; Statistics; Applied mathematics; Medicine; Cardiology; Myocardial infarction","score_opus":0.04814974984037414,"score_gpt":0.3575446066813403,"score_spread":0.3093948568409662,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2110863526","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.047035825,0.002488545,0.9293673,0.002582813,0.0005481303,0.00029407963,0.0015183444,0.0006108319,0.015554073],"genre_scores_gemma":[0.83374006,0.0066451766,0.11101112,0.0013998975,0.000795245,0.001020885,0.0021814115,0.00021433858,0.042991903],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9981159,0.0006057531,0.00008541294,0.00043892764,0.0003739821,0.000379999],"domain_scores_gemma":[0.9971853,0.0012145463,0.00041202418,0.00035710962,0.000679005,0.00015195458],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0034272515,0.001674868,0.0019254393,0.002189593,0.0010589311,0.0022468132,0.003810209,0.0028026388,0.008394135],"category_scores_gemma":[0.009870586,0.0005684215,0.0025790182,0.002491847,0.0019118177,0.002986331,0.0020860506,0.0031574906,0.0026166902],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00018407014,0.00010825581,0.014050489,0.00033677588,0.00036082446,0.002310736,0.0008407165,0.28318092,0.0021069972,0.61357576,0.016972221,0.0659722],"study_design_scores_gemma":[0.00013372929,0.00017868183,0.0049617304,0.00017435133,0.00026553706,0.0021408913,0.0002644919,0.5611236,0.0005414776,0.40876493,0.021296678,0.00015385689],"about_ca_topic_score_codex":0.014025957,"about_ca_topic_score_gemma":0.0077338917,"teacher_disagreement_score":0.014025957,"about_ca_system_score_codex":0.0019140106,"about_ca_system_score_gemma":0.0022181852,"threshold_uncertainty_score":0.028081119},"labels":[],"label_agreement":null},{"id":"W2111018099","doi":"10.1002/sim.1851","title":"The use of quantile regression in health care research: a case study examining gender differences in the timeliness of thrombolytic therapy","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":94,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; Sunnybrook Health Science Centre; Institute for Clinical Evaluative Sciences; Women's College Hospital; Toronto General Hospital; University of Toronto","funders":"","keywords":"Quantile regression; Quantile; Thrombolysis; Medicine; Regression analysis; Statistics; Regression; Linear regression; Econometrics; Proportional hazards model; Myocardial infarction; Mathematics; Internal medicine","score_opus":0.8464695569015244,"score_gpt":0.5826981290513751,"score_spread":0.2637714278501493,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2111018099","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8368339,0.008103047,0.111756526,0.027549151,0.00023029471,0.00040336268,0.00048203368,0.00009279952,0.014548892],"genre_scores_gemma":[0.96598303,0.002395876,0.029995985,0.0006233815,0.00008524183,0.00017790712,0.000060689497,0.00002734504,0.00065058435],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.96369785,0.033354834,0.00055710407,0.00051530235,0.0011503977,0.00072452094],"domain_scores_gemma":[0.864196,0.123516135,0.005434097,0.0032480676,0.0026928568,0.0009128419],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.044399425,0.0003695114,0.00087801984,0.0015575895,0.0017923862,0.001522605,0.0010107601,0.002403228,0.0017324341],"category_scores_gemma":[0.08791648,0.00037171502,0.001228697,0.004472656,0.002551103,0.0017699407,0.0020256785,0.0023258496,0.00017739195],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001700102,0.0017317168,0.5008596,0.00072165014,0.0005844374,0.020043358,0.02708735,0.034668382,0.0012715616,0.15721685,0.013657063,0.24045798],"study_design_scores_gemma":[0.0006817013,0.0043167314,0.338239,0.0014100806,0.0009123991,0.020522851,0.061323293,0.32167253,0.0045495844,0.1954153,0.050369237,0.00058734167],"about_ca_topic_score_codex":0.014682296,"about_ca_topic_score_gemma":0.012560198,"teacher_disagreement_score":0.95560056,"about_ca_system_score_codex":0.0026220046,"about_ca_system_score_gemma":0.0022223133,"threshold_uncertainty_score":0.2348094},"labels":[],"label_agreement":null},{"id":"W2111584093","doi":"10.1002/sim.4444","title":"Effect of dependent errors in the assessment of diagnostic or screening test accuracy when the reference standard is imperfect","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Hemodynamic Monitoring and Therapy","field":"Medicine","cited_by":41,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"National Medical Research Council; National Health and Medical Research Council","keywords":"Computer science; Statistics; Imperfect; Test (biology); Diagnostic accuracy; Medicine; Mathematics; Internal medicine","score_opus":0.03475759916229025,"score_gpt":0.4109700804001137,"score_spread":0.3762124812378234,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2111584093","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5277974,0.004438103,0.4435955,0.0030651353,0.00027722862,0.0005331469,0.0007969809,0.000769363,0.018727237],"genre_scores_gemma":[0.9510181,0.00037513004,0.046699002,0.00036503453,0.000045839213,0.00010195132,0.00014518353,0.00011020108,0.0011396016],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.94818836,0.035695985,0.0028216864,0.004056488,0.008386401,0.00085123326],"domain_scores_gemma":[0.48447114,0.44778514,0.024218587,0.03458499,0.008109681,0.0008304535],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.070574604,0.0012838555,0.00095823884,0.002007433,0.0005653188,0.002365684,0.0018262542,0.0023789983,0.0018052841],"category_scores_gemma":[0.34876978,0.0009111636,0.0016942703,0.0015026504,0.004048406,0.002810621,0.0041319937,0.0019894284,0.00043540576],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0023213807,0.0005114478,0.3410267,0.00075868197,0.0024735963,0.002328135,0.001870355,0.41742045,0.009184026,0.09509277,0.0023345372,0.124678016],"study_design_scores_gemma":[0.00037448987,0.001455385,0.22581047,0.0008131862,0.0015208352,0.003339381,0.00048275586,0.5971673,0.030841598,0.13351944,0.0041861017,0.0004890766],"about_ca_topic_score_codex":0.0037324117,"about_ca_topic_score_gemma":0.0028351613,"teacher_disagreement_score":0.9294254,"about_ca_system_score_codex":0.0017007143,"about_ca_system_score_gemma":0.0011445426,"threshold_uncertainty_score":0.37323862},"labels":[],"label_agreement":null},{"id":"W2111616613","doi":"10.1002/sim.5827","title":"Inverse probability weighted estimating equations for randomized trials in transfusion medicine","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Confounding; Randomized controlled trial; Inverse probability; Clinical trial; Platelet transfusion; Intensive care medicine; Platelet; Emergency medicine; Internal medicine; Statistics; Mathematics","score_opus":0.5703369182139181,"score_gpt":0.5825705942847677,"score_spread":0.012233676070849575,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2111616613","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0014045185,0.006615351,0.9808539,0.0029280344,0.0017555476,0.0040192115,0.0007281423,0.00063971226,0.0010556611],"genre_scores_gemma":[0.036001597,0.0060557295,0.9158813,0.002250954,0.0013122289,0.03339432,0.0013847968,0.0002715455,0.003447561],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5833477,0.39048597,0.009236477,0.008155696,0.0076199668,0.0011541374],"domain_scores_gemma":[0.3390643,0.620731,0.018226175,0.013233662,0.007894892,0.00084996386],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.26356378,0.0048646117,0.013739788,0.0069553265,0.0010468863,0.006140344,0.0075463876,0.007861797,0.015637614],"category_scores_gemma":[0.5260875,0.0028539884,0.0073850504,0.008306798,0.004413569,0.005481907,0.0045860773,0.014796129,0.003758671],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.003105717,0.0004848335,0.006738797,0.0127221905,0.013735009,0.0007679528,0.00084209215,0.16707917,0.0005236683,0.48660144,0.036492426,0.2709067],"study_design_scores_gemma":[0.0024685496,0.00082401955,0.0014842165,0.0029795482,0.002233123,0.00028371674,0.00013299016,0.43610495,0.000511145,0.5253173,0.0274171,0.00024331853],"about_ca_topic_score_codex":0.0050396943,"about_ca_topic_score_gemma":0.0035951221,"teacher_disagreement_score":0.26356378,"about_ca_system_score_codex":0.005849331,"about_ca_system_score_gemma":0.008378008,"threshold_uncertainty_score":0.908157},"labels":[],"label_agreement":null},{"id":"W2114217608","doi":"10.1002/sim.4192","title":"Comparison of two populations of curves with an application in neuronal data analysis","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Control Systems and Identification","field":"Engineering","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Carnegie Mellon University; University of Pittsburgh","keywords":"Nonparametric statistics; Statistic; Test statistic; Statistical hypothesis testing; Parametric statistics; Bayesian probability; Statistics; Mathematics; Functional data analysis; Histogram; Akaike information criterion; Computer science; Pattern recognition (psychology); Artificial intelligence","score_opus":0.09554432008486988,"score_gpt":0.37469032354282217,"score_spread":0.27914600345795226,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2114217608","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03342042,0.00019111049,0.96467465,0.00015966235,0.00004475278,0.000113499504,0.00007237933,0.00073715614,0.0005864378],"genre_scores_gemma":[0.29204652,0.00016452212,0.7059536,0.000092730894,0.000074334996,0.00048800808,0.00027274576,0.0003126014,0.00059492985],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99013734,0.0057168608,0.00051222224,0.0013282691,0.0020288676,0.0002763161],"domain_scores_gemma":[0.9019673,0.083524205,0.0037175766,0.0064770863,0.0032326526,0.0010811421],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020216346,0.00095735,0.0016263245,0.0061464813,0.000960713,0.0017494181,0.0016698312,0.0027153639,0.0034709629],"category_scores_gemma":[0.12099988,0.0005842675,0.0021977674,0.0034197238,0.0039802543,0.0030608897,0.0026994685,0.0025967935,0.0006627369],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0022082774,0.0005942488,0.02317767,0.0007343433,0.00071628497,0.00095084624,0.0019020461,0.16802093,0.031539258,0.14465675,0.0028342956,0.6226651],"study_design_scores_gemma":[0.00013611413,0.0007971321,0.010780924,0.00008973863,0.00007493717,0.0006857891,0.00026687802,0.8708848,0.010762002,0.10094592,0.0044133253,0.00016230979],"about_ca_topic_score_codex":0.00069930416,"about_ca_topic_score_gemma":0.00032608554,"teacher_disagreement_score":0.020216346,"about_ca_system_score_codex":0.00066210906,"about_ca_system_score_gemma":0.00078855176,"threshold_uncertainty_score":0.10691553},"labels":[],"label_agreement":null},{"id":"W2115003020","doi":"10.1002/sim.6178","title":"Impact of the model‐building strategy on inference about nonlinear and time‐dependent covariate effects in survival analysis","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":40,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Royal Victoria Hospital; McGill University","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Covariate; Spurious relationship; Econometrics; Proportional hazards model; Inference; Multivariable calculus; Context (archaeology); Statistics; Computer science; Mathematics; Artificial intelligence","score_opus":0.06310715120089137,"score_gpt":0.43557922383711273,"score_spread":0.37247207263622134,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2115003020","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.047729503,0.00075252383,0.947987,0.0016546,0.000076231016,0.00025782766,0.00017621853,0.0005623199,0.00080372504],"genre_scores_gemma":[0.47715273,0.00047452055,0.51970696,0.0011347821,0.00005879925,0.00051683723,0.0003187982,0.00028994246,0.000346554],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9222485,0.070913106,0.0018741392,0.0019282096,0.0025617925,0.00047426592],"domain_scores_gemma":[0.6654358,0.31540188,0.0044003488,0.010018185,0.0039475327,0.00079627073],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11612801,0.0015282102,0.0018680203,0.00087835494,0.0013969118,0.0021173824,0.0021027084,0.0014676404,0.0010084785],"category_scores_gemma":[0.26060015,0.0010160092,0.0023502873,0.0012596138,0.0020832643,0.0024862564,0.0028512732,0.0051940572,0.00024858039],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012872387,0.0001880882,0.027949607,0.0005509914,0.0018021846,0.00062343874,0.00085917715,0.813678,0.0031077384,0.0461961,0.001966538,0.10179082],"study_design_scores_gemma":[0.00022139265,0.0002638303,0.002858714,0.00016911149,0.00031969504,0.00012643104,0.00009710144,0.9503356,0.0020138114,0.042276517,0.0012393546,0.000078417215],"about_ca_topic_score_codex":0.019225297,"about_ca_topic_score_gemma":0.019539526,"teacher_disagreement_score":0.11612801,"about_ca_system_score_codex":0.001863691,"about_ca_system_score_gemma":0.0075976816,"threshold_uncertainty_score":0.614151},"labels":[],"label_agreement":null},{"id":"W2115556810","doi":"10.1002/sim.6602","title":"Optimal full matching for survival outcomes: a method that merits more widespread use","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":106,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; National Institute of Mental Health; Ontario Ministry of Health and Long-Term Care; Heart and Stroke Foundation of Canada","keywords":"Matching (statistics); Computer science; Statistics; Survival analysis; Propensity score matching; Econometrics; Mathematics","score_opus":0.33550950521288825,"score_gpt":0.5206100027822268,"score_spread":0.18510049756933855,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2115556810","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0031386863,0.0005170628,0.99447495,0.0007597556,0.000049882063,0.0001312121,0.000069486596,0.00021686904,0.0006421075],"genre_scores_gemma":[0.11996162,0.0010078885,0.87577707,0.0007731022,0.0002688373,0.0006382907,0.00033513867,0.00019346384,0.0010446187],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9799875,0.014626402,0.0007781136,0.0018269477,0.0024996898,0.000281289],"domain_scores_gemma":[0.9493768,0.03707037,0.002627945,0.008395829,0.002106033,0.00042298323],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.037296895,0.0008866053,0.0030719172,0.0029200325,0.0009354927,0.0018835436,0.002335986,0.0018868813,0.004646351],"category_scores_gemma":[0.119907856,0.0008743598,0.0024556874,0.0034841842,0.0021427607,0.004479817,0.0039888364,0.0028414151,0.0008914427],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00051815953,0.00025459589,0.008028397,0.0006080171,0.00068386627,0.00013323591,0.00045507113,0.10409128,0.0011664206,0.25619632,0.008159012,0.6197057],"study_design_scores_gemma":[0.00027843504,0.00024449776,0.0028613282,0.00019482439,0.0001520443,0.0002707527,0.00008017912,0.45860323,0.0011211189,0.5235429,0.012561686,0.000088996574],"about_ca_topic_score_codex":0.0028551463,"about_ca_topic_score_gemma":0.0016902856,"teacher_disagreement_score":0.037296895,"about_ca_system_score_codex":0.0014027682,"about_ca_system_score_gemma":0.0044154283,"threshold_uncertainty_score":0.19724715},"labels":[],"label_agreement":null},{"id":"W2115650046","doi":"10.1002/sim.1674","title":"Evaluation of Cox's model and logistic regression for matched case‐control data with time‐dependent covariates: a simulation study","year":2003,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":48,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; McGill University; Institut National de la Recherche Scientifique; Montreal General Hospital","funders":"Canadian Institutes of Health Research","keywords":"Covariate; Statistics; Proportional hazards model; Logistic regression; Regression analysis; Econometrics; Population; Regression; Mathematics; Computer science; Demography","score_opus":0.30995005806529696,"score_gpt":0.507983957605554,"score_spread":0.19803389954025707,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2115650046","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5252165,0.005179725,0.45982572,0.0024425765,0.00030367394,0.0014598941,0.0011875954,0.00046667916,0.0039176247],"genre_scores_gemma":[0.8882248,0.0013416231,0.10739328,0.00025770476,0.00007645351,0.0009349791,0.00073380984,0.00007527201,0.00096214295],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.960275,0.035853773,0.0007312038,0.0012707381,0.0013761424,0.00049310294],"domain_scores_gemma":[0.60926557,0.373026,0.0059264842,0.005587251,0.0049841814,0.0012105417],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08751033,0.0011966269,0.002359857,0.0020097394,0.0006686246,0.0017505243,0.0029235026,0.0026916387,0.0026023162],"category_scores_gemma":[0.15989162,0.0010001132,0.0030702003,0.0021933466,0.001253076,0.0018911482,0.0017500444,0.002143292,0.00023890748],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002397304,0.00037947303,0.027300194,0.00042390442,0.0009634122,0.00048348427,0.0003024188,0.92792344,0.00026214635,0.019488933,0.0009818664,0.01909334],"study_design_scores_gemma":[0.0002418137,0.00040572966,0.0015624092,0.000064169464,0.00011854213,0.00011097031,0.00008306229,0.99282104,0.00016223444,0.0039028728,0.00049455574,0.000032626333],"about_ca_topic_score_codex":0.014063575,"about_ca_topic_score_gemma":0.0052493936,"teacher_disagreement_score":0.91248965,"about_ca_system_score_codex":0.0028042838,"about_ca_system_score_gemma":0.002709469,"threshold_uncertainty_score":0.46280438},"labels":[],"label_agreement":null},{"id":"W2115955076","doi":"10.1002/sim.6223","title":"A multiple imputation strategy for sequential multiple assignment randomized trials","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":62,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"National Institute of Biomedical Imaging and Bioengineering; National Institute of Mental Health; National Institute on Drug Abuse; National Institutes of Natural Sciences; Natural Sciences and Engineering Research Council of Canada; National Cancer Institute; University of North Carolina at Chapel Hill; National Institutes of Health; Canadian Institutes of Health Research; Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada","keywords":"Missing data; Imputation (statistics); Computer science; Inference; Randomized controlled trial; Machine learning; Data mining; Artificial intelligence; Medicine","score_opus":0.17336014046715387,"score_gpt":0.4659763089343173,"score_spread":0.29261616846716343,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2115955076","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00013609625,0.00019498979,0.99828404,0.00038941187,0.00007398931,0.00027024446,0.00009479906,0.00011970909,0.00043657274],"genre_scores_gemma":[0.016055582,0.0006376914,0.9761874,0.0006075831,0.00026693038,0.0038385622,0.00031682535,0.00013427684,0.0019551609],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9035568,0.08406589,0.0031001463,0.003146007,0.0055070263,0.00062403374],"domain_scores_gemma":[0.91819376,0.06745299,0.004002636,0.005524675,0.0041188253,0.0007070861],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09010335,0.0020960653,0.004434889,0.003669938,0.0011361787,0.0031885188,0.006103229,0.0041231755,0.01366069],"category_scores_gemma":[0.15185446,0.0017105516,0.0045341705,0.0066143903,0.0019796442,0.003346652,0.0036406089,0.006836162,0.0045978706],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00048761498,0.00016335394,0.0012277329,0.0013583759,0.0018236339,0.0004752923,0.00052206544,0.08268206,0.0007401021,0.5813754,0.023429029,0.3057153],"study_design_scores_gemma":[0.00052775483,0.00033292195,0.00035233045,0.00044614906,0.0003792343,0.00031970532,0.000066286855,0.36472347,0.0006187421,0.6038574,0.028275426,0.00010055476],"about_ca_topic_score_codex":0.0019178148,"about_ca_topic_score_gemma":0.0023829762,"teacher_disagreement_score":0.9098967,"about_ca_system_score_codex":0.0020955577,"about_ca_system_score_gemma":0.006548816,"threshold_uncertainty_score":0.47651774},"labels":[],"label_agreement":null},{"id":"W2117578563","doi":"10.1002/sim.6185","title":"Propensity score methods for estimating relative risks in cluster randomized trials with low‐incidence binary outcomes and selection bias","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":38,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"Agence Nationale de la Recherche","keywords":"Propensity score matching; Covariate; Selection bias; Statistics; Inverse probability weighting; Weighting; Randomization; Regression; Randomized controlled trial; Selection (genetic algorithm); Relative risk; Cluster (spacecraft); Medicine; Econometrics; Mathematics; Computer science; Internal medicine; Artificial intelligence; Confidence interval","score_opus":0.39037839286313425,"score_gpt":0.5488097609618341,"score_spread":0.15843136809869984,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2117578563","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0032508317,0.0018724389,0.9927209,0.00039741636,0.00009973563,0.0008924893,0.00013212912,0.00016903311,0.00046482714],"genre_scores_gemma":[0.12615655,0.002686744,0.86288697,0.0005971826,0.0003036494,0.0061701154,0.0003812289,0.00014506334,0.0006725953],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.85052896,0.13808306,0.003412385,0.002946146,0.0046033007,0.0004262206],"domain_scores_gemma":[0.7509046,0.22179084,0.012453481,0.010129171,0.004143762,0.0005781698],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.12713994,0.0016245208,0.0031811285,0.00470197,0.0005857755,0.002132429,0.0027527567,0.0020421972,0.0043922197],"category_scores_gemma":[0.28488266,0.00095118576,0.0038216019,0.005318604,0.0019199519,0.0026569702,0.002940772,0.0043701907,0.00064741896],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020126652,0.00026775038,0.0089411745,0.004996541,0.006773013,0.00026981797,0.00086711015,0.1818671,0.0008861265,0.44969153,0.0077448105,0.33568233],"study_design_scores_gemma":[0.0021451192,0.0007759674,0.0031814598,0.0009036648,0.0015413935,0.0002375041,0.000101187994,0.39901263,0.0010282354,0.5812407,0.009708297,0.00012385464],"about_ca_topic_score_codex":0.0017262995,"about_ca_topic_score_gemma":0.0013198892,"teacher_disagreement_score":0.8728601,"about_ca_system_score_codex":0.0016386494,"about_ca_system_score_gemma":0.004232546,"threshold_uncertainty_score":0.6723883},"labels":[],"label_agreement":null},{"id":"W2117590223","doi":"10.1002/sim.4392","title":"Relative survival multistate Markov model","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Canadian Institutes of Health Research","keywords":"Relative survival; Markov model; Proportional hazards model; Survival analysis; Markov chain; Statistics; Hazard ratio; Hazard; Relative risk; Medicine; Econometrics; Computer science; Mathematics; Cancer; Cancer registry; Internal medicine; Biology; Confidence interval","score_opus":0.1961461722704344,"score_gpt":0.42473856033198565,"score_spread":0.22859238806155124,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2117590223","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.020377286,0.00071139756,0.9700558,0.0013389137,0.00016360676,0.00017985489,0.00207903,0.00060729746,0.0044867303],"genre_scores_gemma":[0.76672214,0.0020765015,0.20080552,0.0007137722,0.000343296,0.0017292458,0.004657374,0.00015835585,0.02279377],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9972391,0.0015135539,0.00012772773,0.000540996,0.00031449454,0.00026415396],"domain_scores_gemma":[0.9920495,0.00597714,0.0006257179,0.0005132326,0.0006400368,0.00019440617],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0063719787,0.0011094311,0.0019013825,0.0014132599,0.0006509455,0.0017084009,0.0033582635,0.0019794435,0.0110372435],"category_scores_gemma":[0.012313219,0.0006639109,0.0016985226,0.0015949217,0.0012028344,0.0023002997,0.0015087497,0.0028516974,0.0019478776],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002450704,0.00008228881,0.004236382,0.00021727818,0.0001603697,0.00028537493,0.00027846295,0.65029794,0.0006036041,0.31444776,0.0039201505,0.025225308],"study_design_scores_gemma":[0.000055605127,0.000055188877,0.00043371675,0.000022905135,0.00005175983,0.00008409627,0.000026217476,0.9064733,0.0001493292,0.08991074,0.0027113461,0.00002585553],"about_ca_topic_score_codex":0.006701079,"about_ca_topic_score_gemma":0.004531111,"teacher_disagreement_score":0.0110372435,"about_ca_system_score_codex":0.0017351857,"about_ca_system_score_gemma":0.0017975867,"threshold_uncertainty_score":0.03692323},"labels":[],"label_agreement":null},{"id":"W2117746627","doi":"10.1002/sim.5598","title":"Predictive accuracy of risk factors and markers: a simulation study of the effect of novel markers on different performance measures for logistic regression models","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":48,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Logistic regression; Statistics; Brier score; Statistic; Context (archaeology); Econometrics; Risk factor; Mathematics; Computer science; Medicine; Internal medicine; Biology","score_opus":0.1122183754170676,"score_gpt":0.4199947875634422,"score_spread":0.30777641214637463,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2117746627","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92960465,0.0018170706,0.06551755,0.0009093871,0.00006527399,0.00022015837,0.00039126203,0.00017840364,0.001296214],"genre_scores_gemma":[0.9811172,0.0002711119,0.01786615,0.000096868,0.000017026588,0.00012750461,0.0002774136,0.000028106146,0.00019862609],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.97934777,0.017432796,0.00057707797,0.0011647616,0.00096965587,0.00050791894],"domain_scores_gemma":[0.54427904,0.42974538,0.009193383,0.01019159,0.0052778632,0.0013127621],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05825346,0.0013680416,0.0017189435,0.0015331156,0.00077028695,0.0016372871,0.0016075539,0.0022618596,0.0007587098],"category_scores_gemma":[0.17630641,0.00062836317,0.002448845,0.0014867159,0.0021114864,0.002306999,0.0017881979,0.0032698307,0.00013710443],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0032074118,0.00046234127,0.06366641,0.00020265067,0.000844394,0.00026266923,0.00033188894,0.91017926,0.000973915,0.006056252,0.0006000989,0.013212688],"study_design_scores_gemma":[0.0002131875,0.0009781847,0.0076921084,0.000085693995,0.00024661154,0.00012273202,0.00008366531,0.9833059,0.001239713,0.0056892415,0.00028225788,0.000060689516],"about_ca_topic_score_codex":0.007983718,"about_ca_topic_score_gemma":0.0051717875,"teacher_disagreement_score":0.94174653,"about_ca_system_score_codex":0.002131586,"about_ca_system_score_gemma":0.0017223574,"threshold_uncertainty_score":0.3080774},"labels":[],"label_agreement":null},{"id":"W2118529247","doi":"10.1002/sim.5923","title":"Multiscale analysis of neural spike trains","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Neural dynamics and brain function","field":"Neuroscience","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Spike train; Spike (software development); Computer science; Train; Point process; Neural coding; Artificial intelligence; Algorithm; Pattern recognition (psychology); Statistics; Mathematics","score_opus":0.03014703907680518,"score_gpt":0.309619638389451,"score_spread":0.27947259931264584,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2118529247","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04437713,0.00032884706,0.95362973,0.00015869176,0.000024801677,0.000014446189,0.000100629644,0.00020723339,0.001158466],"genre_scores_gemma":[0.8382695,0.0008239618,0.15827695,0.00011271211,0.00014304197,0.00006299789,0.00027325266,0.00023090481,0.0018068522],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9996388,0.00009252449,0.000020629279,0.00007620945,0.00013499698,0.000036841695],"domain_scores_gemma":[0.9982476,0.0010287355,0.0003132326,0.00015557135,0.00017227225,0.000082520244],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009762997,0.00035027473,0.000436082,0.0013844142,0.00020619115,0.0007269152,0.00056419004,0.00050725974,0.0012531315],"category_scores_gemma":[0.005260905,0.00021218011,0.00074954034,0.00083081785,0.00076919363,0.0010630722,0.00075393385,0.0006681844,0.0002063538],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000077411634,0.00003617534,0.00445558,0.00022793972,0.00014307149,0.0003341535,0.0002828801,0.5773543,0.051782034,0.28597018,0.0012137902,0.07812244],"study_design_scores_gemma":[0.0000030040787,0.00001688785,0.002434642,0.000010452029,0.000010566867,0.000076310265,0.000016721531,0.9384231,0.0018667086,0.0560718,0.0010517387,0.000018101438],"about_ca_topic_score_codex":0.0010267269,"about_ca_topic_score_gemma":0.0006319239,"teacher_disagreement_score":0.0013844142,"about_ca_system_score_codex":0.00041936035,"about_ca_system_score_gemma":0.0003394545,"threshold_uncertainty_score":0.0051631927},"labels":[],"label_agreement":null},{"id":"W2118670046","doi":"10.1002/sim.2518","title":"Confidence intervals for multinomial logistic regression in sparse data","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":54,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Lunenfeld-Tanenbaum Research Institute; University of Toronto; Mount Sinai Hospital","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Multinomial logistic regression; Statistics; Logistic regression; Mathematics; Covariate; Confidence interval; Likelihood function; Binary data; Multinomial distribution; Econometrics; Wald test; Maximum likelihood; Binary number; Statistical hypothesis testing","score_opus":0.2664078337446776,"score_gpt":0.4945271978202856,"score_spread":0.228119364075608,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2118670046","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.018239489,0.004447138,0.9726778,0.0008716058,0.00012252462,0.00011278503,0.00047325136,0.0006581369,0.0023972362],"genre_scores_gemma":[0.5197155,0.0035464722,0.47092944,0.0005690887,0.00040964442,0.0011762455,0.0023317933,0.0003409926,0.0009807787],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9628756,0.02480062,0.0022109668,0.0033976194,0.0061724246,0.0005428099],"domain_scores_gemma":[0.47336295,0.47689793,0.01956654,0.016864043,0.011684511,0.0016239858],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.057452593,0.001018041,0.0022484825,0.0058980705,0.0007125,0.0030829709,0.0037267066,0.0029183533,0.0042086397],"category_scores_gemma":[0.46963584,0.0006826512,0.0017408783,0.0046881037,0.0025459828,0.00452929,0.0035822222,0.0041006105,0.00076435605],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010933193,0.00015078217,0.031006448,0.0021112482,0.0011860316,0.00080170785,0.0011301462,0.157902,0.0013835436,0.45765257,0.009322716,0.33625942],"study_design_scores_gemma":[0.00019454618,0.00025399379,0.0110844765,0.0016094589,0.0002884482,0.0012946568,0.00031656027,0.55163705,0.0021600004,0.42037934,0.010593848,0.00018764751],"about_ca_topic_score_codex":0.0017823677,"about_ca_topic_score_gemma":0.00080830423,"teacher_disagreement_score":0.057452593,"about_ca_system_score_codex":0.0013941514,"about_ca_system_score_gemma":0.0009957701,"threshold_uncertainty_score":0.30384195},"labels":[],"label_agreement":null},{"id":"W2118844219","doi":"10.1002/sim.5805","title":"Reply to Taguri and Matsuyama","year":2013,"lang":"en","type":"letter","venue":"Statistics in Medicine","topic":"Influenza Virus Research Studies","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"National Institutes of Health","keywords":"Computer science; Computational biology; Biology","score_opus":0.07162140203662545,"score_gpt":0.40929713591478967,"score_spread":0.3376757338781642,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2118844219","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0001562979,0.0006053333,0.000050126517,0.98344606,0.0148891965,0.000013314584,0.000052721352,0.0000142934705,0.00077268446],"genre_scores_gemma":[0.0006844721,0.00024941904,0.000063172694,0.982646,0.013700079,0.000034626875,0.000013379112,0.000011880465,0.0025970316],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99124205,0.0025975294,0.0012497194,0.0010581408,0.0020118298,0.001840658],"domain_scores_gemma":[0.9697118,0.018134264,0.0026061255,0.0007459304,0.004188782,0.00461299],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010918114,0.0012215407,0.0027220894,0.0015055257,0.008039783,0.00825357,0.004175755,0.12903693,0.008134283],"category_scores_gemma":[0.0657783,0.0020087154,0.0021136838,0.001985284,0.0053617777,0.0053008036,0.004233736,0.07655242,0.008197604],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000044085242,0.0000127302665,0.00048041533,0.000039578405,0.00001733302,0.00048045957,0.0001449724,0.000026170355,0.000058357633,0.001024164,0.99577004,0.0019016657],"study_design_scores_gemma":[0.00018844305,0.000070956194,0.0036219407,0.000747905,0.00012186052,0.0010094228,0.0015081177,0.0005354801,0.0003194925,0.009562002,0.982152,0.0001624645],"about_ca_topic_score_codex":0.018517619,"about_ca_topic_score_gemma":0.03390352,"teacher_disagreement_score":0.12903693,"about_ca_system_score_codex":0.009246186,"about_ca_system_score_gemma":0.012416397,"threshold_uncertainty_score":0.0670861},"labels":[],"label_agreement":null},{"id":"W2119396122","doi":"10.1002/sim.1484","title":"A relative survival regression model using B‐spline functions to model non‐proportional hazards","year":2003,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":106,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Montreal General Hospital","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Ligue Contre le Cancer","keywords":"Proportional hazards model; Covariate; Relative survival; Statistics; Regression analysis; Survival analysis; Population; Inference; Regression; Hazard ratio; Relative risk; Econometrics; Mathematics; Medicine; Computer science; Cancer registry; Confidence interval; Artificial intelligence","score_opus":0.17991573856172782,"score_gpt":0.4606334294214673,"score_spread":0.28071769085973947,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2119396122","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0025802758,0.00019258891,0.99636555,0.00016901451,0.000058701575,0.000052602394,0.00013282444,0.00015929223,0.00028925054],"genre_scores_gemma":[0.15144983,0.0015424949,0.8358515,0.00035196275,0.00027001972,0.0016488773,0.001107987,0.00028161154,0.007495775],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99311405,0.005416442,0.00020345993,0.0004934742,0.0006026496,0.00016995515],"domain_scores_gemma":[0.98531336,0.012350308,0.00062998576,0.00069378316,0.000833099,0.00017956202],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020593692,0.0012516583,0.0017370931,0.002313128,0.00073559984,0.0012582846,0.0032321604,0.0019263129,0.0042660725],"category_scores_gemma":[0.032058068,0.0006806714,0.002909524,0.0034367188,0.0010298741,0.0017276158,0.001768606,0.00457129,0.0013896299],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00028475703,0.00016548084,0.004751734,0.00038687454,0.0005236462,0.00035300624,0.0004936596,0.60773873,0.0019204747,0.26131716,0.007108387,0.11495607],"study_design_scores_gemma":[0.00008022288,0.00017512048,0.0007471122,0.0000436091,0.00008808432,0.00017147689,0.0000363621,0.92189485,0.00032019257,0.06943823,0.0069580367,0.000046704066],"about_ca_topic_score_codex":0.006530201,"about_ca_topic_score_gemma":0.0041720667,"teacher_disagreement_score":0.020593692,"about_ca_system_score_codex":0.0007825036,"about_ca_system_score_gemma":0.0019480046,"threshold_uncertainty_score":0.10891116},"labels":[],"label_agreement":null},{"id":"W2119511462","doi":"10.1002/sim.6433","title":"Penalized regression procedures for variable selection in the potential outcomes framework","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"National Institute of Environmental Health Sciences; National Cancer Institute; National Institute on Drug Abuse; National Institutes of Health","keywords":"Causal inference; Imputation (statistics); Computer science; Inference; Missing data; Regression; Multivariate statistics; Feature selection; Regression analysis; Artificial intelligence; Model selection; Selection (genetic algorithm); Machine learning; Data mining; Econometrics; Statistics; Mathematics","score_opus":0.11775119733631027,"score_gpt":0.4738352817476763,"score_spread":0.356084084411366,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2119511462","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00033755656,0.0003000026,0.9983695,0.0003249143,0.00003428218,0.00004254171,0.000056773883,0.00009145577,0.0004429521],"genre_scores_gemma":[0.040985793,0.0018025762,0.9504919,0.0006004841,0.00048514776,0.0013012805,0.0005845635,0.0003865583,0.0033617276],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96886975,0.026371459,0.0006571953,0.0014638782,0.0022391237,0.00039855455],"domain_scores_gemma":[0.92590964,0.06366038,0.0029046147,0.0043589314,0.0027318643,0.00043463038],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.039579958,0.0019526955,0.002471559,0.0032904749,0.0010885913,0.0024047627,0.0041446257,0.0023290163,0.0087832175],"category_scores_gemma":[0.09556364,0.000922098,0.0025635445,0.004282696,0.0034959752,0.003630347,0.0036973206,0.0070728343,0.0024143907],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006356919,0.00006246069,0.00079894497,0.00028332745,0.00024305108,0.00011049477,0.00018510975,0.0711074,0.00035168586,0.83886224,0.0076330476,0.08029862],"study_design_scores_gemma":[0.00006811566,0.00005400755,0.00038049222,0.0001018099,0.000050298433,0.00006995098,0.000033357534,0.28067786,0.0003852117,0.7085497,0.009592368,0.000037014386],"about_ca_topic_score_codex":0.002940726,"about_ca_topic_score_gemma":0.0025345397,"teacher_disagreement_score":0.039579958,"about_ca_system_score_codex":0.0018008678,"about_ca_system_score_gemma":0.003619716,"threshold_uncertainty_score":0.20932132},"labels":[],"label_agreement":null},{"id":"W2119934595","doi":"10.1002/(sici)1097-0258(20000215)19:3<313::aid-sim370>3.0.co;2-k","title":"Bootstrap confidence intervals for the sensitivity of a quantitative diagnostic test","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":81,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Confidence interval; Sensitivity (control systems); Statistics; Parametric statistics; Nominal level; Inference; Computer science; Test (biology); Econometrics; Mathematics; Artificial intelligence","score_opus":0.16794440763566826,"score_gpt":0.46817699776453064,"score_spread":0.3002325901288624,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2119934595","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01720888,0.00462334,0.97147006,0.0007234876,0.00021161522,0.00015547956,0.00035260996,0.00061000505,0.004644422],"genre_scores_gemma":[0.6105453,0.0033334747,0.3806563,0.00080023194,0.00088577176,0.0010453039,0.0011677111,0.00045657562,0.0011093772],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9570984,0.026939007,0.0016962528,0.0030541262,0.010488342,0.0007238443],"domain_scores_gemma":[0.54352397,0.41330433,0.013769349,0.014726438,0.013504472,0.0011713487],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.062366687,0.0016601463,0.0024445157,0.007797282,0.00069999567,0.0030864256,0.0034393228,0.003121922,0.0041943355],"category_scores_gemma":[0.4227846,0.00069287873,0.0018869495,0.0036771877,0.003937952,0.0034921267,0.0033549918,0.0040756753,0.0011603461],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013564414,0.00026458356,0.025039665,0.0024435874,0.001995825,0.00094754994,0.001282145,0.2411949,0.003612814,0.427167,0.010612602,0.28408298],"study_design_scores_gemma":[0.00022117209,0.0005688206,0.012038116,0.001703963,0.00040845104,0.0012715382,0.0003315317,0.5183833,0.005948229,0.4465758,0.012282573,0.00026644883],"about_ca_topic_score_codex":0.00083208864,"about_ca_topic_score_gemma":0.00036772905,"teacher_disagreement_score":0.93763334,"about_ca_system_score_codex":0.0010854366,"about_ca_system_score_gemma":0.0010806734,"threshold_uncertainty_score":0.32983047},"labels":[],"label_agreement":null},{"id":"W2120330366","doi":"10.1002/sim.1096","title":"Meta‐analyses in systematic reviews of randomized controlled trials in perinatal medicine: comparison of fixed and random effects models","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":97,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Meta-analysis; Random effects model; Relative risk; Randomized controlled trial; Confidence interval; Fixed effects model; Medicine; Study heterogeneity; Statistics; Systematic review; Publication bias; MEDLINE; Internal medicine; Mathematics; Biology","score_opus":0.799571230144511,"score_gpt":0.6104975582558801,"score_spread":0.1890736718886309,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2120330366","genre_codex":"review","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0030774062,0.8102433,0.16910118,0.0046247966,0.0049573076,0.0053505152,0.0010528472,0.00054125243,0.0010513348],"genre_scores_gemma":[0.13025182,0.39533812,0.42211667,0.0055507957,0.0038232543,0.039632957,0.0016115262,0.00071656174,0.0009583171],"study_design_codex":"meta_analysis","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.2869693,0.64565134,0.041619096,0.0092020165,0.015644267,0.0009139728],"domain_scores_gemma":[0.34646586,0.60276824,0.02442819,0.018538726,0.0069464752,0.0008525296],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4016777,0.008311348,0.041194428,0.026463784,0.0015292612,0.012597998,0.0087876115,0.009347832,0.0046197176],"category_scores_gemma":[0.6850423,0.0069504012,0.07602271,0.025496181,0.0033077095,0.0125387665,0.0056787296,0.0131178135,0.00084795104],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0038236433,0.00009441301,0.0034053442,0.17327473,0.7280577,0.0002132802,0.0005291879,0.013532684,0.00015722653,0.009298833,0.004874473,0.06273856],"study_design_scores_gemma":[0.010252606,0.0021004523,0.004228236,0.09055498,0.77140427,0.00048870384,0.00036789753,0.022196278,0.0007327336,0.08250222,0.014550393,0.00062123465],"about_ca_topic_score_codex":0.0057694786,"about_ca_topic_score_gemma":0.009474587,"teacher_disagreement_score":0.5983223,"about_ca_system_score_codex":0.0099451095,"about_ca_system_score_gemma":0.009430412,"threshold_uncertainty_score":0.7378379},"labels":[],"label_agreement":null},{"id":"W2122527396","doi":"10.1002/sim.3801","title":"Missing data in the exposure of interest and marginal structural models: A simulation study based on the Framingham Heart Study","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"National Medical Research Council; National Health and Medical Research Council","keywords":"Missing data; Inverse probability; Framingham Heart Study; Marginal structural model; Statistics; Confounding; Propensity score matching; Computer science; Econometrics; Outcome (game theory); Mathematics; Medicine; Framingham Risk Score; Bayesian probability","score_opus":0.3765319600692848,"score_gpt":0.49942054001456876,"score_spread":0.12288857994528396,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2122527396","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92634994,0.0011893981,0.065184586,0.002060245,0.00010021054,0.00030841742,0.0008846046,0.00010859477,0.003814023],"genre_scores_gemma":[0.97081673,0.0005381205,0.026576292,0.00021738303,0.00006000052,0.00038636065,0.00062686123,0.000029905204,0.0007484317],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9837879,0.014773292,0.00023389491,0.0004824824,0.0003386949,0.00038372903],"domain_scores_gemma":[0.7724994,0.21669784,0.003070669,0.004139351,0.0025032808,0.0010894344],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.031875517,0.00081070163,0.0014627345,0.0013845447,0.0009030785,0.0010283251,0.0013831662,0.0017879154,0.0026969379],"category_scores_gemma":[0.08979665,0.0005439765,0.0021693716,0.00142932,0.0012759252,0.0015458644,0.0014420466,0.00249957,0.00016200611],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002508184,0.0012128444,0.049652185,0.0003322202,0.0010597595,0.00086833787,0.0011652537,0.86165416,0.00033406733,0.061282374,0.0033888638,0.016541794],"study_design_scores_gemma":[0.0005624424,0.00065536186,0.005766041,0.00011516354,0.00027268555,0.00014031511,0.0003688454,0.968814,0.00023161291,0.022083625,0.0009244482,0.00006535995],"about_ca_topic_score_codex":0.015275358,"about_ca_topic_score_gemma":0.008876367,"teacher_disagreement_score":0.9681245,"about_ca_system_score_codex":0.001351794,"about_ca_system_score_gemma":0.0012592005,"threshold_uncertainty_score":0.16857588},"labels":[],"label_agreement":null},{"id":"W2122697238","doi":"10.1002/sim.5761","title":"Relative risk regression models with inverse polynomials","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Distribution Estimation and Applications","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Inverse; Mathematics; Polynomial; Applied mathematics; Bounded function; Hazard; Polynomial regression; Function (biology); Log-linear model; Polynomial and rational function modeling; Statistics; Regression analysis; Linear model; Mathematical analysis","score_opus":0.06394317640654087,"score_gpt":0.3758390173818452,"score_spread":0.3118958409753043,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2122697238","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00940689,0.00055956113,0.9871334,0.0005006316,0.00004978878,0.000032707227,0.00012140973,0.00016183956,0.0020337026],"genre_scores_gemma":[0.72570616,0.003043515,0.25110832,0.00044009977,0.00039832987,0.00048478754,0.00060799345,0.00026242968,0.017948397],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99282867,0.0042150402,0.00021245013,0.0010012401,0.0013399057,0.00040267114],"domain_scores_gemma":[0.9878314,0.008995014,0.0012611602,0.00092920143,0.0008413839,0.00014184265],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009266416,0.0013199257,0.0015133569,0.0017321273,0.00043747496,0.0022832246,0.003471944,0.0017641422,0.0028266986],"category_scores_gemma":[0.0313473,0.0007805526,0.0018768144,0.0022608193,0.0016454081,0.003088768,0.0018054415,0.0036727688,0.0012139544],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00011587253,0.000061115,0.0021990137,0.00013884615,0.000115057446,0.0002444671,0.00029529352,0.557781,0.00072941504,0.39709672,0.0016521293,0.039571036],"study_design_scores_gemma":[0.000023617678,0.000061450526,0.0004067863,0.000022513075,0.000042394167,0.00013307914,0.000029427902,0.853559,0.00024164902,0.14263701,0.0028133562,0.0000297023],"about_ca_topic_score_codex":0.0047050216,"about_ca_topic_score_gemma":0.0024671203,"teacher_disagreement_score":0.009266416,"about_ca_system_score_codex":0.0014072391,"about_ca_system_score_gemma":0.0014276427,"threshold_uncertainty_score":0.049006104},"labels":[],"label_agreement":null},{"id":"W2124694383","doi":"10.1002/sim.2329","title":"Interval estimation of the mean response in a log‐regression model","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"National Cancer Institute; American Lebanese Syrian Associated Charities","keywords":"Statistics; Regression analysis; Mathematics; Sample size determination; Prediction interval; Confidence interval; Interval estimation; Interval (graph theory); Regression; Inference; Econometrics; Computer science","score_opus":0.15493030895431098,"score_gpt":0.5154423581120995,"score_spread":0.3605120491577885,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2124694383","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0071792887,0.00019457616,0.99187976,0.000085258966,0.000020598774,0.000052349893,0.00005137281,0.00013539968,0.0004014401],"genre_scores_gemma":[0.36990136,0.000650854,0.62640554,0.00022083455,0.00011716238,0.00093201024,0.00049641903,0.000104896164,0.0011710213],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9771752,0.01709473,0.0006081892,0.0021764457,0.0024953217,0.0004502307],"domain_scores_gemma":[0.8838178,0.106106326,0.0035652362,0.0035292355,0.0025417197,0.00043975151],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02637091,0.00091716833,0.0023867255,0.001763965,0.00037892064,0.0013205367,0.0028428545,0.0022207692,0.0028344183],"category_scores_gemma":[0.106121354,0.00056489353,0.0013723616,0.0015927675,0.0013804546,0.0021384554,0.0016182346,0.0029668715,0.0005028433],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020308513,0.000623097,0.0130162025,0.0010173772,0.0007276587,0.00047534276,0.0009506733,0.4207107,0.007750339,0.20728917,0.0030941807,0.34231436],"study_design_scores_gemma":[0.000112477595,0.00041491186,0.0026349206,0.000084180865,0.00011286053,0.0001650599,0.00006259646,0.91565734,0.0022151629,0.07656909,0.001894813,0.00007659946],"about_ca_topic_score_codex":0.0010185015,"about_ca_topic_score_gemma":0.0005286769,"teacher_disagreement_score":0.02637091,"about_ca_system_score_codex":0.00083002134,"about_ca_system_score_gemma":0.0011152145,"threshold_uncertainty_score":0.13946438},"labels":[],"label_agreement":null},{"id":"W2125238080","doi":"10.1002/sim.5684","title":"Estimating the intensity of ward admission and its effect on emergency department access block","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Emergency and Acute Care Studies","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"Australian Research Council; Queensland Health","keywords":"Poisson regression; Emergency department; Block (permutation group theory); Hospital admission; Medicine; Computer science; Emergency medicine; Internal medicine; Mathematics; Population","score_opus":0.03647757346316049,"score_gpt":0.387662823291902,"score_spread":0.3511852498287415,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2125238080","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9962388,0.00009513984,0.0028391809,0.00017699713,0.000008480724,0.00003738204,0.00022067052,0.000020119496,0.00036312448],"genre_scores_gemma":[0.9969555,0.00006482624,0.0020674008,0.000029991292,0.000017940789,0.00004166381,0.00040263296,0.000010665715,0.00040936997],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.97919583,0.016791996,0.00080258894,0.0013908512,0.00092731236,0.0008914636],"domain_scores_gemma":[0.78567547,0.18029031,0.021560553,0.0067826402,0.0024732596,0.0032177286],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.018397415,0.0005527963,0.0011152697,0.0015635949,0.00049398624,0.0013238423,0.0014140899,0.0011725381,0.0024413015],"category_scores_gemma":[0.08375945,0.0005706936,0.0017649712,0.0023383386,0.0012999338,0.0011648629,0.0023280547,0.002660443,0.00026937563],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00071567297,0.00031921917,0.980294,0.000029467405,0.0007069696,0.000087636545,0.00048061984,0.0078149205,0.00031036802,0.0005034456,0.00020967072,0.008528002],"study_design_scores_gemma":[0.000035019173,0.0006025905,0.9544636,0.000022126902,0.00030476923,0.00005357083,0.00051198533,0.0426547,0.00039691257,0.0005502143,0.0003730535,0.000031393083],"about_ca_topic_score_codex":0.025867198,"about_ca_topic_score_gemma":0.017751092,"teacher_disagreement_score":0.025867198,"about_ca_system_score_codex":0.0013592117,"about_ca_system_score_gemma":0.0014582513,"threshold_uncertainty_score":0.097296},"labels":[],"label_agreement":null},{"id":"W2125296802","doi":"10.1002/sim.5327","title":"An efficient alternative to the stratified Cox model analysis","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Gilead Sciences (Canada)","funders":"","keywords":"Stratum; Proportional hazards model; Statistics; Inference; Categorical variable; Sample size determination; Hazard ratio; Econometrics; Mathematics; Computer science; Confidence interval; Geology; Artificial intelligence","score_opus":0.5116409653596984,"score_gpt":0.6119579326220749,"score_spread":0.10031696726237649,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2125296802","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011783764,0.00026577455,0.9969608,0.00027028858,0.00012383318,0.00012993992,0.00012871667,0.000176738,0.00076543796],"genre_scores_gemma":[0.09272475,0.0007522554,0.9005823,0.0005432796,0.0003671364,0.0010876871,0.00049073214,0.00020883062,0.0032429942],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9843588,0.012155136,0.00042162914,0.00091098645,0.0017872169,0.00036635183],"domain_scores_gemma":[0.9849251,0.010279644,0.00066635315,0.0026629632,0.0012550333,0.00021098134],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.019466765,0.0011622444,0.002245722,0.0018862125,0.00050497235,0.0017960834,0.0022034266,0.001385718,0.010027364],"category_scores_gemma":[0.051187117,0.0006731838,0.00241895,0.002295627,0.0006639561,0.0020683657,0.0022054908,0.0027816074,0.001949444],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00091732264,0.0002269565,0.0041610915,0.00060488965,0.0012480491,0.0004682492,0.000346038,0.087604925,0.0019659656,0.4958272,0.017223837,0.3894054],"study_design_scores_gemma":[0.00041542755,0.00054521114,0.0013448029,0.00016689609,0.00042912218,0.0005830932,0.00008680216,0.5080723,0.0013559625,0.45611805,0.030780189,0.00010206961],"about_ca_topic_score_codex":0.0018018916,"about_ca_topic_score_gemma":0.0020777301,"teacher_disagreement_score":0.019466765,"about_ca_system_score_codex":0.000861847,"about_ca_system_score_gemma":0.0044999155,"threshold_uncertainty_score":0.10295135},"labels":[],"label_agreement":null},{"id":"W2126049444","doi":"10.1002/sim.3697","title":"Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity‐score matched samples","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":6487,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Covariate; Statistics; Matching (statistics); Quantile; Mathematics; Medicine","score_opus":0.2114497428934903,"score_gpt":0.42025483801624836,"score_spread":0.20880509512275805,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2126049444","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.019821076,0.0011584124,0.9687081,0.0006563682,0.00032063096,0.0013083165,0.002772326,0.0014111813,0.003843636],"genre_scores_gemma":[0.3184685,0.0007240424,0.6667436,0.00094305497,0.0003946361,0.005830506,0.0042151785,0.0007186478,0.0019619437],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.880947,0.09332233,0.0062487973,0.0068518925,0.011639399,0.0009905659],"domain_scores_gemma":[0.6266996,0.32235757,0.0240376,0.01971458,0.0062293005,0.0009612389],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.091588035,0.0015410535,0.0026246258,0.010805282,0.0011945798,0.0027790968,0.003842496,0.0034303325,0.013733185],"category_scores_gemma":[0.35920033,0.0010315916,0.002396284,0.0075444845,0.0039576227,0.003772966,0.0030415764,0.0034888461,0.0020550897],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0035098165,0.0005444972,0.16194321,0.0034802172,0.005085194,0.0011750255,0.0018956327,0.025033046,0.004060078,0.26699764,0.033703484,0.49257213],"study_design_scores_gemma":[0.0013958912,0.0026014582,0.12813275,0.0015717695,0.0016916456,0.0029879673,0.0013979635,0.30865055,0.012307788,0.4764229,0.062398642,0.00044061488],"about_ca_topic_score_codex":0.0019102904,"about_ca_topic_score_gemma":0.0010333966,"teacher_disagreement_score":0.091588035,"about_ca_system_score_codex":0.0014513901,"about_ca_system_score_gemma":0.0024643394,"threshold_uncertainty_score":0.48436964},"labels":[],"label_agreement":null},{"id":"W2126565679","doi":"10.1002/sim.6750","title":"A rare variant association test in family‐based designs and non‐normal quantitative traits","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic Associations and Epidemiology","field":"Biochemistry, Genetics and Molecular Biology","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Jewish General Hospital; Université du Québec à Montréal; Université Laval","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; National Institute for Health and Care Research; Department of Health and Social Care; Wellcome Trust; Wellcome","keywords":"Normality; Quantitative trait locus; Phenotype; Genetic association; Association test; Trait; Statistics; Biology; Computational biology; Genetics; Computer science; Mathematics; Single-nucleotide polymorphism; Genotype","score_opus":0.04158283716710916,"score_gpt":0.32908024858684204,"score_spread":0.2874974114197329,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2126565679","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02747171,0.00015437603,0.97068375,0.00017773893,0.00007756712,0.00021961333,0.00021922495,0.00029724196,0.0006987409],"genre_scores_gemma":[0.44349873,0.00018684327,0.5520892,0.00027052732,0.00012314058,0.0014357953,0.00066464616,0.00010014333,0.0016309812],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9507985,0.043010704,0.0008264946,0.0025568411,0.0023599842,0.00044751537],"domain_scores_gemma":[0.830395,0.15798533,0.003112909,0.0054446575,0.0022865029,0.0007756706],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0463071,0.0008650536,0.0018775736,0.0024484443,0.0008652746,0.0012655455,0.0028254758,0.0025902255,0.005325083],"category_scores_gemma":[0.13293791,0.0004993465,0.0017661655,0.0033955197,0.0028616386,0.001815065,0.002015979,0.0018592452,0.0009028191],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0031347058,0.0005665184,0.07581297,0.0008755082,0.0024763485,0.0017130426,0.0007120154,0.20195447,0.0049296375,0.26303786,0.007164233,0.43762273],"study_design_scores_gemma":[0.0005853724,0.0020047254,0.014405373,0.00009966721,0.00032053285,0.001412103,0.00017501334,0.8820794,0.0017078537,0.09330113,0.003789086,0.00011973039],"about_ca_topic_score_codex":0.0017669719,"about_ca_topic_score_gemma":0.0016273081,"teacher_disagreement_score":0.0463071,"about_ca_system_score_codex":0.0006790251,"about_ca_system_score_gemma":0.0026435116,"threshold_uncertainty_score":0.24489826},"labels":[],"label_agreement":null},{"id":"W2126948068","doi":"10.1002/sim.2747","title":"Assessing local influence in principal component analysis with application to haematology study data","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"","keywords":"Principal component analysis; Data set; Eigenvalues and eigenvectors; Data mining; Mathematics; Perturbation (astronomy); Statistics; Computer science; Sensitivity (control systems)","score_opus":0.11621309066097045,"score_gpt":0.5022851333630338,"score_spread":0.38607204270206336,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2126948068","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.045499347,0.00052599894,0.95259976,0.00015772204,0.000029309173,0.00015082875,0.00005833218,0.0002573285,0.00072130613],"genre_scores_gemma":[0.62122166,0.0005619866,0.37708285,0.0000649239,0.00012555388,0.00025147683,0.00021596695,0.000080486236,0.00039521416],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98889077,0.0067242216,0.0005949137,0.0010244689,0.002538739,0.00022677027],"domain_scores_gemma":[0.9593375,0.032829482,0.0025034545,0.002066033,0.0027869854,0.00047648873],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01539838,0.0011240606,0.0015024651,0.005277758,0.0009645053,0.0016556177,0.00077952433,0.001071352,0.0006788963],"category_scores_gemma":[0.06478317,0.0003904218,0.0012317108,0.004035806,0.0018333164,0.0015437374,0.0017299461,0.0013958844,0.00021666322],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00068319176,0.0001806173,0.047117356,0.00079602725,0.0010544104,0.0008814567,0.0019998585,0.32854173,0.0126993945,0.03737025,0.0019234553,0.5667523],"study_design_scores_gemma":[0.00002951985,0.0002701764,0.028297277,0.000093998635,0.0001774933,0.00037131616,0.00041657183,0.9104357,0.0057092113,0.051805764,0.002269017,0.00012397578],"about_ca_topic_score_codex":0.001807683,"about_ca_topic_score_gemma":0.0017713543,"teacher_disagreement_score":0.01539838,"about_ca_system_score_codex":0.00067112397,"about_ca_system_score_gemma":0.0009117159,"threshold_uncertainty_score":0.08143538},"labels":[],"label_agreement":null},{"id":"W2127105291","doi":"10.1002/sim.4463","title":"Bayesian approach to predicting cancer incidence for an area without cancer registration by using cancer incidence data from nearby areas","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Global Cancer Incidence and Screening","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Cancer registry; Incidence (geometry); Cancer; Cancer incidence; Mortality rate; Statistics; Population; Demography; Medicine; Mathematics; Surgery; Environmental health; Internal medicine","score_opus":0.2254399786217776,"score_gpt":0.4520156101574731,"score_spread":0.22657563153569552,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2127105291","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.23593567,0.00045146912,0.75874954,0.00044876517,0.000031066043,0.00013735339,0.00070823217,0.00052996253,0.0030079721],"genre_scores_gemma":[0.8255923,0.00039281574,0.17086467,0.000081680046,0.00005867012,0.0002246316,0.0015563866,0.00005116972,0.0011776659],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9983157,0.0009793546,0.000081211794,0.00026605933,0.0002849439,0.00007289282],"domain_scores_gemma":[0.9965334,0.002351907,0.00037464005,0.0002162862,0.00043040322,0.00009325679],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0045851823,0.00059842435,0.00076012383,0.0018442282,0.00036417705,0.00095126097,0.0014306123,0.0006735762,0.0010327494],"category_scores_gemma":[0.012079239,0.00051872386,0.0012410518,0.0010738295,0.00032049287,0.0012743742,0.00095590355,0.00076014927,0.00030761398],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00021438776,0.00016562355,0.07179325,0.0000750948,0.00026810588,0.00009743472,0.00019620883,0.8219113,0.0006431489,0.0060259746,0.0007242733,0.09788523],"study_design_scores_gemma":[0.000032030344,0.000071869996,0.0137588475,0.00002614405,0.00006033871,0.00006616978,0.00007149448,0.9782321,0.00029521793,0.006524059,0.0008225469,0.000039299153],"about_ca_topic_score_codex":0.021702444,"about_ca_topic_score_gemma":0.02687324,"teacher_disagreement_score":0.021702444,"about_ca_system_score_codex":0.0009666281,"about_ca_system_score_gemma":0.0014601259,"threshold_uncertainty_score":0.043152213},"labels":[],"label_agreement":null},{"id":"W2128212246","doi":"10.1002/sim.4098","title":"Mixture cure model with random effects for the analysis of a multi‐center tonsil cancer study","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":38,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"National Cancer Institute; Natural Sciences and Engineering Research Council of Canada; Queen's University","keywords":"Statistics; Maximum likelihood; Covariate; Random effects model; Mathematics; Gaussian; Cure rate; Mixture model; Correlation; Medicine; Surgery; Internal medicine","score_opus":0.01772856284049867,"score_gpt":0.3572900698339538,"score_spread":0.3395615069934551,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2128212246","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.024909595,0.001371365,0.97100484,0.0009125212,0.00016075988,0.00044399884,0.00041513593,0.00029350255,0.0004883522],"genre_scores_gemma":[0.4272011,0.0021635802,0.555949,0.0008199073,0.0004301841,0.0065641054,0.0023268878,0.00023646018,0.004308846],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.97332776,0.022097224,0.00061758136,0.0021079597,0.0013913119,0.00045814159],"domain_scores_gemma":[0.93448645,0.054976933,0.0034370816,0.00508498,0.001304605,0.0007098949],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.053928208,0.0018263462,0.0041376585,0.00513828,0.0014617925,0.0023507343,0.0051811947,0.0037880028,0.0069737383],"category_scores_gemma":[0.08147974,0.0017532728,0.005439742,0.0039273594,0.0027222214,0.0028480187,0.0037747642,0.00558223,0.00091767457],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0027940555,0.00044993896,0.040050413,0.0009469942,0.005095414,0.0012631238,0.0018917682,0.4013771,0.0020488072,0.43340194,0.0067559266,0.10392458],"study_design_scores_gemma":[0.00044560886,0.00041798808,0.005585429,0.0001840001,0.0007550993,0.0002841472,0.00014278432,0.81136024,0.0003918315,0.1747543,0.005539806,0.00013877671],"about_ca_topic_score_codex":0.007536687,"about_ca_topic_score_gemma":0.0053446144,"teacher_disagreement_score":0.053928208,"about_ca_system_score_codex":0.0017773508,"about_ca_system_score_gemma":0.00221154,"threshold_uncertainty_score":0.28520304},"labels":[],"label_agreement":null},{"id":"W2129017495","doi":"10.1002/sim.5723","title":"Correction: ‘Generating survival times to simulate Cox proportional hazards models with time‐varying covariates’","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences","funders":"","keywords":"Covariate; Weibull distribution; Proportional hazards model; Statistics; Mathematics; Cumulative distribution function; Event (particle physics); Hazard; Expression (computer science); Gompertz function; Function (biology); Econometrics; Probability density function; Computer science; Physics","score_opus":0.23952068522826894,"score_gpt":0.49038527983643865,"score_spread":0.2508645946081697,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2129017495","genre_codex":"editorial","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0022008216,0.0032613503,0.17146401,0.07982668,0.67110497,0.00048088588,0.031751875,0.030059507,0.009849876],"genre_scores_gemma":[0.10130817,0.006728839,0.37145373,0.15445396,0.10455327,0.0038636292,0.04639276,0.040744446,0.17050117],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9771827,0.008457882,0.003317157,0.0024267624,0.0077305855,0.0008849549],"domain_scores_gemma":[0.7148532,0.17269681,0.011370915,0.027156949,0.07158563,0.0023364667],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026329976,0.0035937887,0.0024711618,0.005484976,0.002515406,0.004143435,0.006331096,0.006872267,0.15772806],"category_scores_gemma":[0.35610902,0.0023738751,0.0035011536,0.00673868,0.0028342959,0.0036329166,0.0051753093,0.011338246,0.056974374],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016727425,0.00002310467,0.0005525183,0.00068556145,0.000114240094,0.00021001791,0.00024262548,0.001375174,0.00028057068,0.005028384,0.9667964,0.02452416],"study_design_scores_gemma":[0.00024251429,0.00006250387,0.0014451646,0.0009633815,0.00014563554,0.00054858485,0.00011129543,0.009277415,0.001856037,0.015466858,0.9697017,0.00017885669],"about_ca_topic_score_codex":0.014474553,"about_ca_topic_score_gemma":0.01364513,"teacher_disagreement_score":0.15772806,"about_ca_system_score_codex":0.004284052,"about_ca_system_score_gemma":0.0074036233,"threshold_uncertainty_score":0.5276529},"labels":[],"label_agreement":null},{"id":"W2129802796","doi":"10.1002/sim.2754","title":"Pseudoscore‐based estimation from biased observations","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"National Institute of Allergy and Infectious Diseases; Natural Sciences and Engineering Research Council of Canada","keywords":"Estimation; Computer science; Statistics; Econometrics; Mathematics","score_opus":0.15895267835464572,"score_gpt":0.41798995805961153,"score_spread":0.2590372797049658,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2129802796","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.023230115,0.00021578089,0.97534287,0.00021857528,0.000035933794,0.00005284117,0.00011864006,0.00016398769,0.00062121253],"genre_scores_gemma":[0.5808156,0.0007762948,0.41152897,0.00050076254,0.00019517425,0.00067871093,0.0016681592,0.0001927726,0.003643535],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9883736,0.008579548,0.00045843312,0.0009892216,0.0013323802,0.00026683943],"domain_scores_gemma":[0.9158097,0.06812863,0.0039491556,0.008177329,0.0034845276,0.0004507222],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026405286,0.00092042325,0.001974044,0.0018390765,0.00064831116,0.0025673527,0.0030519047,0.0020156116,0.003020556],"category_scores_gemma":[0.1195145,0.000880618,0.0013073855,0.0021295059,0.002574338,0.0041981335,0.002964817,0.0018005215,0.0008101614],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00066738884,0.00013720553,0.03588651,0.00049644517,0.00066696573,0.000570291,0.00071807776,0.32011762,0.0019415643,0.39277408,0.004045952,0.24197802],"study_design_scores_gemma":[0.00006309157,0.000107913635,0.004419921,0.00010195585,0.000058501326,0.0002647577,0.00007003455,0.82055366,0.001324953,0.17029218,0.0027006816,0.000042321695],"about_ca_topic_score_codex":0.0019800852,"about_ca_topic_score_gemma":0.0018400247,"teacher_disagreement_score":0.026405286,"about_ca_system_score_codex":0.0009167866,"about_ca_system_score_gemma":0.0015774799,"threshold_uncertainty_score":0.13964617},"labels":[],"label_agreement":null},{"id":"W2130247335","doi":"10.1002/sim.3869","title":"Controlling ecological bias in evidence synthesis of trials reporting on collapsed and overlapping covariate categories","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; University of Bristol; Heart and Stroke Foundation of Canada","keywords":"Covariate; Econometrics; Statistics; Computer science; Mathematics","score_opus":0.8103766064910152,"score_gpt":0.5788043668938704,"score_spread":0.2315722395971448,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2130247335","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009863491,0.081265315,0.8889797,0.0055728997,0.0018299671,0.007522679,0.0016511058,0.0008617492,0.0024530592],"genre_scores_gemma":[0.23261335,0.022286793,0.7090721,0.005187202,0.0011731526,0.027243981,0.0013995037,0.00036632377,0.0006575717],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.09742166,0.8147301,0.055210225,0.012754983,0.019036885,0.0008461197],"domain_scores_gemma":[0.09144821,0.8258041,0.040946614,0.03531573,0.006009204,0.0004762127],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6679286,0.0050514643,0.020958986,0.018041128,0.003024972,0.010962292,0.0064661633,0.009037632,0.0034940075],"category_scores_gemma":[0.87987125,0.004837518,0.02328575,0.025661664,0.008699381,0.011901881,0.013619266,0.00899821,0.0006098287],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.007894008,0.00024103107,0.02190195,0.12518707,0.21584347,0.0007805162,0.0064655957,0.044751875,0.0012588301,0.16361603,0.007574017,0.4044856],"study_design_scores_gemma":[0.006962589,0.0019105178,0.010602166,0.033400178,0.08868378,0.00071364315,0.0006507866,0.052673012,0.0029293764,0.77727324,0.023396565,0.00080414396],"about_ca_topic_score_codex":0.0038593577,"about_ca_topic_score_gemma":0.005025576,"teacher_disagreement_score":0.33207142,"about_ca_system_score_codex":0.007337427,"about_ca_system_score_gemma":0.013078212,"threshold_uncertainty_score":0.40950322},"labels":[],"label_agreement":null},{"id":"W2130606506","doi":"10.1002/1097-0258(20010115)20:1<63::aid-sim656>3.0.co;2-2","title":"Statistical analysis of repeated measurements with informative censoring times","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"National Institute of Allergy and Infectious Diseases; Natural Sciences and Engineering Research Council of Canada; National Institutes of Health","keywords":"Censoring (clinical trials); Parametric statistics; Computer science; Missing data; Statistical hypothesis testing; Statistics; Regression; Econometrics; Artificial intelligence; Machine learning; Mathematics","score_opus":0.10166540638061615,"score_gpt":0.4125996493320424,"score_spread":0.31093424295142624,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2130606506","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03343811,0.0021146273,0.9624317,0.0005710019,0.00022355765,0.00025094333,0.00028051084,0.00024790314,0.0004415863],"genre_scores_gemma":[0.5704014,0.0025916675,0.42062375,0.0006133196,0.00053315685,0.0026978175,0.0010200539,0.00013527452,0.0013835117],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9115317,0.07151772,0.0030936643,0.0052485866,0.007874322,0.0007339993],"domain_scores_gemma":[0.7056676,0.23833115,0.020734793,0.03114632,0.0034277143,0.00069240184],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08227947,0.0012095879,0.0036061998,0.0026969507,0.0007105644,0.0015495288,0.0032337338,0.0021293135,0.0022074392],"category_scores_gemma":[0.27700478,0.00066263176,0.0033005613,0.003662464,0.0037494865,0.0020406672,0.002235846,0.003335426,0.00032561523],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0027182866,0.00058898353,0.068726316,0.0037133587,0.012719196,0.0029299082,0.0024292665,0.11261427,0.0049968823,0.25802693,0.0056753005,0.5248613],"study_design_scores_gemma":[0.00042751725,0.0030464043,0.04558114,0.0006310043,0.0017058299,0.0010241976,0.00049051427,0.34816658,0.0048145666,0.5867741,0.007148858,0.00018923476],"about_ca_topic_score_codex":0.0007462359,"about_ca_topic_score_gemma":0.00051154575,"teacher_disagreement_score":0.08227947,"about_ca_system_score_codex":0.0008875512,"about_ca_system_score_gemma":0.0020142396,"threshold_uncertainty_score":0.43514067},"labels":[],"label_agreement":null},{"id":"W2130728435","doi":"10.1002/sim.2103","title":"The partial area under the summary ROC curve","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":246,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Receiver operating characteristic; Truncation (statistics); Measure (data warehouse); Sensitivity (control systems); Statistics; Area under the curve; Area under curve; Mathematics; Computer science; Medicine; Data mining; Internal medicine","score_opus":0.6102396939187572,"score_gpt":0.5384282700556492,"score_spread":0.07181142386310801,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2130728435","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.059938345,0.07337894,0.80560446,0.005104248,0.0018257351,0.0018326467,0.021954564,0.005140054,0.025220932],"genre_scores_gemma":[0.7351241,0.018143784,0.21976694,0.003309276,0.0021241505,0.0047290535,0.012150491,0.0010742919,0.003577921],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9160307,0.05673062,0.0072804643,0.010384002,0.008655848,0.0009184812],"domain_scores_gemma":[0.7268672,0.20960116,0.02621201,0.019356217,0.016604468,0.0013589297],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.062411077,0.0020965212,0.0046617226,0.010874622,0.0006143747,0.0065352796,0.002476352,0.0033232425,0.0068636998],"category_scores_gemma":[0.24851876,0.0006307205,0.006036657,0.008307933,0.002796534,0.005195881,0.0028248671,0.0026991551,0.0030692094],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0023925623,0.00015833195,0.113518,0.012459564,0.017496195,0.0009072157,0.0011292927,0.08147021,0.002296921,0.07091708,0.051283583,0.645971],"study_design_scores_gemma":[0.0007197437,0.0058247484,0.106877096,0.007984846,0.012656151,0.012343314,0.0018150439,0.32082486,0.006316218,0.2869674,0.23624723,0.0014233794],"about_ca_topic_score_codex":0.0017377556,"about_ca_topic_score_gemma":0.00077601447,"teacher_disagreement_score":0.93758893,"about_ca_system_score_codex":0.0019940312,"about_ca_system_score_gemma":0.002388953,"threshold_uncertainty_score":0.3300653},"labels":[],"label_agreement":null},{"id":"W2131336573","doi":"10.1002/sim.3398","title":"A geometric confidence ellipse approach to the estimation of the ratio of two variables","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Ellipse; Tangent; Confidence interval; Coverage probability; Mathematics; Confidence distribution; Confidence region; Point estimation; Statistics; Applied mathematics; Geometry","score_opus":0.41984205646303874,"score_gpt":0.5298368237881074,"score_spread":0.10999476732506863,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2131336573","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.001576259,0.0004499833,0.9961054,0.00022766265,0.000038463495,0.000037718786,0.00005317016,0.00011361599,0.0013977645],"genre_scores_gemma":[0.20920704,0.0020878972,0.78422934,0.00041356904,0.000556518,0.000628937,0.0005786027,0.00031312677,0.0019849278],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9629285,0.0269798,0.0013257606,0.003785857,0.0044251992,0.00055498746],"domain_scores_gemma":[0.8696444,0.11216042,0.006229772,0.0063864966,0.005069253,0.00050961063],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.039739367,0.0021226935,0.0023813215,0.0065766466,0.0007564923,0.004319749,0.0036342128,0.0028240813,0.005081816],"category_scores_gemma":[0.20641124,0.0010152338,0.002123325,0.0048199818,0.0051948433,0.005545603,0.0048755007,0.004518394,0.0013607292],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015899177,0.000053506086,0.005009685,0.00044706737,0.0003576063,0.00035890838,0.000568881,0.23851195,0.0010195907,0.60273033,0.003678744,0.14710473],"study_design_scores_gemma":[0.000051892024,0.00019850678,0.0021761064,0.00023653758,0.00010276042,0.00070751936,0.00015913874,0.57770514,0.0018230119,0.40260395,0.0141280135,0.00010740829],"about_ca_topic_score_codex":0.002211842,"about_ca_topic_score_gemma":0.00068328716,"teacher_disagreement_score":0.039739367,"about_ca_system_score_codex":0.0016711646,"about_ca_system_score_gemma":0.0014555029,"threshold_uncertainty_score":0.21016437},"labels":[],"label_agreement":null},{"id":"W2131644669","doi":"10.1002/sim.2078","title":"Cost-effectiveness analysis for multinational clinical trials","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":52,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; SickKids Foundation; University of Toronto; Population Health Research Institute; St. Joseph’s Healthcare Hamilton; Hospital for Sick Children","funders":"","keywords":"Univariate; Multinational corporation; Econometrics; Estimator; Clinical trial; Estimation; Multivariate statistics; Medicine; Statistics; Multivariate analysis; Computer science; Actuarial science; Economics; Mathematics; Finance; Internal medicine","score_opus":0.7655298801975778,"score_gpt":0.6423986283181136,"score_spread":0.12313125187946428,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2131644669","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.012312286,0.020715995,0.93396443,0.006971404,0.00096320495,0.00856863,0.0023063112,0.0008432544,0.013354506],"genre_scores_gemma":[0.26712716,0.011615759,0.670178,0.0027168011,0.0010147338,0.039659172,0.0027518084,0.00044757058,0.0044889874],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.74876195,0.23560195,0.005386434,0.0030860398,0.0063919025,0.00077177567],"domain_scores_gemma":[0.5827905,0.383451,0.0159797,0.011799103,0.0050060228,0.0009736465],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.18802911,0.0023731117,0.005600864,0.006707236,0.00082523003,0.0040644957,0.002607346,0.002975495,0.01680107],"category_scores_gemma":[0.44690844,0.0013579768,0.0050294707,0.007815319,0.0017323147,0.0048237205,0.003527987,0.0053340746,0.001122175],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004525649,0.00035466725,0.0124449385,0.010030354,0.0131388195,0.0007807756,0.00058462366,0.2852974,0.0003973409,0.38007417,0.029419681,0.2629516],"study_design_scores_gemma":[0.0019417586,0.0010193312,0.0051893676,0.0033629155,0.003332225,0.0005216895,0.00020949882,0.5904849,0.00031648466,0.36346215,0.030030208,0.00012956894],"about_ca_topic_score_codex":0.003343174,"about_ca_topic_score_gemma":0.0029818562,"teacher_disagreement_score":0.8119709,"about_ca_system_score_codex":0.0050891405,"about_ca_system_score_gemma":0.0059076655,"threshold_uncertainty_score":0.99440485},"labels":[],"label_agreement":null},{"id":"W2131715846","doi":"10.1002/sim.3645","title":"Two‐stage design of clinical trials involving recurrent events","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":28,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Interim; Interim analysis; Sample size determination; Clinical trial; Event (particle physics); Poisson regression; Poisson distribution; Population; Statistics; Stage (stratigraphy); Medicine; Computer science; Mathematics; Internal medicine","score_opus":0.8501905449859165,"score_gpt":0.7057092213771811,"score_spread":0.14448132360873545,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2131715846","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.027031166,0.0006096423,0.9518466,0.0008717079,0.00045168182,0.017432787,0.00027060608,0.0003286088,0.0011571664],"genre_scores_gemma":[0.26919302,0.0005801589,0.6616175,0.000817553,0.00019496877,0.06460874,0.0004046722,0.000054735046,0.0025286032],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8817289,0.10698707,0.002816283,0.003996459,0.003170351,0.0013009363],"domain_scores_gemma":[0.91812634,0.0631569,0.005933694,0.0073728217,0.0036176941,0.0017925734],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08288683,0.002294785,0.0046798494,0.0011294935,0.000584585,0.0017542404,0.0034566761,0.0034639833,0.005285288],"category_scores_gemma":[0.102499835,0.001962837,0.0033081262,0.00094297534,0.0020959505,0.0018439743,0.0023570433,0.0033497359,0.00082278537],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.07910421,0.004177748,0.008671756,0.0056682876,0.003235433,0.00083339924,0.0011303421,0.34284073,0.011501735,0.27165872,0.0054587657,0.26571882],"study_design_scores_gemma":[0.03225679,0.034442548,0.003138046,0.0003855571,0.0012577401,0.00031138095,0.00008361007,0.7715888,0.005750759,0.1419228,0.008498037,0.0003638762],"about_ca_topic_score_codex":0.000342392,"about_ca_topic_score_gemma":0.00044336615,"teacher_disagreement_score":0.9171132,"about_ca_system_score_codex":0.0013682968,"about_ca_system_score_gemma":0.0043451847,"threshold_uncertainty_score":0.4383527},"labels":[],"label_agreement":null},{"id":"W2131839274","doi":"10.1002/sim.6621","title":"Terminating observation within matched pairs of subjects in a matched cohort analysis: a Monte Carlo simulation study","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital; Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care","keywords":"Censoring (clinical trials); Monte Carlo method; Statistics; Observational study; Cohort; Event (particle physics); Computer science; Cohort study; Econometrics; Mathematics; Physics","score_opus":0.17266836408797095,"score_gpt":0.45042957816131257,"score_spread":0.27776121407334164,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2131839274","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5243766,0.0017215031,0.46761996,0.0013102945,0.00019113049,0.0017423836,0.00038755286,0.000120599856,0.0025300533],"genre_scores_gemma":[0.8796818,0.0005348828,0.117089465,0.00041033034,0.00006258757,0.0013555897,0.00027445325,0.000031391603,0.0005594341],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.944,0.04919599,0.0014257882,0.002977881,0.0015888427,0.0008113892],"domain_scores_gemma":[0.51194084,0.45912036,0.009009577,0.014540388,0.0038913114,0.0014975218],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1235493,0.00086992054,0.0021291613,0.0012838037,0.0013562746,0.0022855462,0.002533654,0.0025994107,0.0026827636],"category_scores_gemma":[0.2662152,0.0008396019,0.0034106793,0.0016681419,0.0019322195,0.002842933,0.0024212913,0.0038246398,0.0002962041],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.015926518,0.0053056814,0.19350022,0.0010659519,0.0070528067,0.00256752,0.0035389354,0.44155023,0.001521581,0.22495888,0.0029252393,0.10008651],"study_design_scores_gemma":[0.0011180755,0.0017024236,0.0073128454,0.00025092124,0.0016534275,0.00035662114,0.00041878916,0.9263402,0.00088934624,0.058060464,0.0017733655,0.00012350193],"about_ca_topic_score_codex":0.009209069,"about_ca_topic_score_gemma":0.006097357,"teacher_disagreement_score":0.8764507,"about_ca_system_score_codex":0.0022209235,"about_ca_system_score_gemma":0.0038093387,"threshold_uncertainty_score":0.653399},"labels":[],"label_agreement":null},{"id":"W2132469784","doi":"10.1002/sim.3719","title":"Quadratic inference functions in marginal models for longitudinal data","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":65,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Paul's Hospital; Alberta Cancer Foundation","funders":"National Science Foundation","keywords":"Inference; Computer science; Robustness (evolution); Statistical inference; Quadratic equation; Macro; Model selection; Applied mathematics; Econometrics; Machine learning; Statistics; Mathematics; Artificial intelligence","score_opus":0.331254591012354,"score_gpt":0.49301774444195917,"score_spread":0.16176315342960518,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2132469784","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00059004925,0.0006603176,0.99784195,0.00021963021,0.000033638626,0.000025218373,0.00008234774,0.000107111104,0.00043970486],"genre_scores_gemma":[0.066327766,0.0038781115,0.9210572,0.00072752486,0.00058883004,0.0010926203,0.0009146823,0.000607969,0.004805238],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97178644,0.022017607,0.0008908593,0.0022739782,0.0025112915,0.0005198577],"domain_scores_gemma":[0.893611,0.09656991,0.002495728,0.0035376363,0.0032862807,0.00049952057],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.052567847,0.0017471329,0.0023167622,0.0031477127,0.00093667186,0.0024876583,0.0033220742,0.0024739678,0.007638624],"category_scores_gemma":[0.14301483,0.0013553475,0.003414015,0.0033420061,0.004253682,0.0052976417,0.0036710873,0.005789198,0.0018338591],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000095147494,0.00005175912,0.0016976894,0.00051172334,0.00024472616,0.0001830612,0.00050621835,0.07304196,0.0004177548,0.8196739,0.005661885,0.097914174],"study_design_scores_gemma":[0.000029540917,0.00007056985,0.0008278948,0.00013211623,0.00007672172,0.0001311446,0.000048312446,0.35068834,0.00039264574,0.6369658,0.010583717,0.000053267544],"about_ca_topic_score_codex":0.0056736814,"about_ca_topic_score_gemma":0.0036644936,"teacher_disagreement_score":0.052567847,"about_ca_system_score_codex":0.0024938888,"about_ca_system_score_gemma":0.0033192288,"threshold_uncertainty_score":0.2780087},"labels":[],"label_agreement":null},{"id":"W2133169911","doi":"10.1002/sim.2913","title":"Impact of approximating or ignoring within‐study covariances in multivariate meta‐analyses","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":58,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Royal Victoria Hospital; Montreal General Hospital; Montreal Children's Hospital; McGill University; Canadian Association of Radiation Oncology","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Fonds Québécois de la Recherche sur la Nature et les Technologies","keywords":"Multivariate statistics; Covariance; Statistics; Econometrics; Meta-analysis; Multivariate analysis; Correlation; Scale (ratio); Mathematics; Medicine","score_opus":0.8667966181484852,"score_gpt":0.6644247021312913,"score_spread":0.20237191601719384,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2133169911","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04285001,0.03031246,0.9144061,0.006307187,0.0011053145,0.0007962105,0.00074971165,0.0011786467,0.0022943069],"genre_scores_gemma":[0.6030669,0.0067701563,0.38370425,0.0033843277,0.00037605883,0.0011123404,0.0005006038,0.0005823149,0.0005030887],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.33590978,0.5887495,0.041758195,0.012664845,0.019645287,0.0012723532],"domain_scores_gemma":[0.118245766,0.8215124,0.02381152,0.029930884,0.0060262852,0.00047303064],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5388361,0.004081587,0.008080182,0.0035068647,0.0015996319,0.0064433385,0.004327876,0.0046775206,0.00095895625],"category_scores_gemma":[0.83711284,0.0032494334,0.009839881,0.009834615,0.003854911,0.0073649534,0.00528299,0.0070291413,0.00035901184],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.006421841,0.00024472259,0.06765036,0.013815142,0.082340576,0.0020298325,0.00453971,0.51347405,0.0023966695,0.055846628,0.0076482263,0.24359222],"study_design_scores_gemma":[0.0025545107,0.0021403912,0.028976114,0.011477472,0.055031303,0.002199771,0.0010601864,0.5980702,0.009802574,0.26844734,0.019147888,0.0010921861],"about_ca_topic_score_codex":0.011087019,"about_ca_topic_score_gemma":0.008321687,"teacher_disagreement_score":0.46116388,"about_ca_system_score_codex":0.0047505265,"about_ca_system_score_gemma":0.005853184,"threshold_uncertainty_score":0.5686972},"labels":[],"label_agreement":null},{"id":"W2133300801","doi":"10.1002/sim.6085","title":"Value of information methods for assessing a new diagnostic test","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto; Public Health Ontario","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Value of information; Test (biology); Value (mathematics); Sample (material); Sample size determination; Statistics; Econometrics; Machine learning; Artificial intelligence; Mathematics","score_opus":0.3956805353606469,"score_gpt":0.635194064645554,"score_spread":0.23951352928490705,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2133300801","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.002345177,0.009923095,0.9790877,0.0032135558,0.00038174077,0.00041438756,0.00022489256,0.00009010852,0.004319309],"genre_scores_gemma":[0.16429053,0.017473882,0.80685496,0.0020684174,0.0025115781,0.0036638724,0.00047910112,0.0002012252,0.0024564257],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.84905326,0.12738961,0.0042776256,0.003916718,0.014599387,0.00076341577],"domain_scores_gemma":[0.4640341,0.50943696,0.008957458,0.008483521,0.008217316,0.0008707047],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.110515,0.0031960784,0.004518846,0.013018494,0.0011311894,0.0073078475,0.0037445612,0.0057282965,0.004878002],"category_scores_gemma":[0.40120646,0.0010990299,0.002812253,0.0072615123,0.006011594,0.0075042495,0.0044006016,0.00735771,0.0008560622],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004747336,0.00012240259,0.00391245,0.002201897,0.001399164,0.00030209665,0.00028333138,0.08006483,0.0003837429,0.7489447,0.0057210173,0.1561896],"study_design_scores_gemma":[0.000103987244,0.00022721502,0.0006828071,0.0010663058,0.00039129474,0.0002473422,0.00009324192,0.14676763,0.0004908416,0.84324926,0.00656252,0.00011751864],"about_ca_topic_score_codex":0.0016762854,"about_ca_topic_score_gemma":0.00088831433,"teacher_disagreement_score":0.889485,"about_ca_system_score_codex":0.0068166,"about_ca_system_score_gemma":0.005905339,"threshold_uncertainty_score":0.58446616},"labels":[],"label_agreement":null},{"id":"W2133313519","doi":"10.1002/sim.5891","title":"A spatial scan statistic for compound Poisson data","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Data-Driven Disease Surveillance","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Canadian Institutes of Health Research; Alberta Innovates; University of Alberta; Government of Alberta","keywords":"Scan statistic; Poisson distribution; Cluster analysis; Statistic; Cluster (spacecraft); Spatial epidemiology; Computer science; Statistics; Spatial analysis; Data mining; Poisson regression; Count data; Epidemiology; Econometrics; Medicine; Artificial intelligence; Mathematics; Environmental health; Pathology","score_opus":0.0479821941350101,"score_gpt":0.37055638338498254,"score_spread":0.32257418924997244,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2133313519","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014904385,0.00040665746,0.9789761,0.0009799806,0.00016654513,0.00016411828,0.0012207365,0.00064449816,0.0025370289],"genre_scores_gemma":[0.36550698,0.00097590423,0.62109274,0.0008342604,0.0007586634,0.00174553,0.0046899035,0.00055245945,0.0038435224],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9916231,0.0055207447,0.00037247068,0.0007629592,0.0015010926,0.00021963807],"domain_scores_gemma":[0.93109965,0.059113294,0.0026940485,0.0031246948,0.0033305213,0.0006377424],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012999299,0.0005467107,0.0011647413,0.0036782993,0.0005881932,0.0014327153,0.0020693117,0.0016374431,0.0067629414],"category_scores_gemma":[0.07536087,0.00046771285,0.0013596644,0.004176225,0.00206629,0.002416945,0.001859278,0.0024054,0.0011568443],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00053572794,0.00015392032,0.044944838,0.0005532395,0.0004519291,0.0012155267,0.0006522685,0.108390346,0.003586868,0.592215,0.030272076,0.21702826],"study_design_scores_gemma":[0.00015596848,0.00033774858,0.0064650034,0.00014480906,0.00010624761,0.0014002454,0.0002481599,0.6695946,0.0017793713,0.29562828,0.024039708,0.00009984372],"about_ca_topic_score_codex":0.00205525,"about_ca_topic_score_gemma":0.001699016,"teacher_disagreement_score":0.012999299,"about_ca_system_score_codex":0.00082697754,"about_ca_system_score_gemma":0.002298793,"threshold_uncertainty_score":0.0687477},"labels":[],"label_agreement":null},{"id":"W2133519275","doi":"10.1002/sim.5599","title":"An information criterion for marginal structural models","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":42,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre","funders":"Eunice Kennedy Shriver National Institute of Child Health and Human Development; National Center for Research Resources; National Institute of Child Health and Human Development; National Institute of Allergy and Infectious Diseases; National Cancer Institute; National Institute on Aging","keywords":"Marginal structural model; Marginal model; Econometrics; Specification; Inference; Computer science; Metric (unit); Observational study; Parametric statistics; Causal inference; Regression; Semiparametric regression; Marginal likelihood; Regression analysis; Statistics; Mathematics; Machine learning; Bayesian probability; Artificial intelligence; Economics","score_opus":0.1399503500513534,"score_gpt":0.47609368767591403,"score_spread":0.3361433376245606,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2133519275","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0047072936,0.00028398278,0.9928281,0.0004694949,0.000022387669,0.00007278231,0.00022472,0.0001336345,0.0012575962],"genre_scores_gemma":[0.34119007,0.0014091856,0.6502185,0.0009605248,0.0003925837,0.0013557014,0.0014839073,0.0003964824,0.002593098],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9799519,0.015276574,0.00083541,0.001239483,0.002233625,0.00046306313],"domain_scores_gemma":[0.83564305,0.14899895,0.0037331085,0.0059602587,0.0049171457,0.00074737536],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.033657312,0.0013363252,0.0025247568,0.004179405,0.0010503387,0.0021829482,0.0031654611,0.0026795287,0.0053373547],"category_scores_gemma":[0.1350731,0.0008615724,0.00271779,0.003282657,0.004392643,0.004976911,0.0034540717,0.0038412956,0.0008328338],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008433645,0.00005386924,0.0026400788,0.0003662811,0.00020833002,0.00016355597,0.00032825352,0.093467936,0.0006058139,0.8613619,0.0026688578,0.038050797],"study_design_scores_gemma":[0.000034678673,0.00010352468,0.0008250524,0.00013332877,0.00004646152,0.00010084551,0.000049654303,0.29792726,0.0004370673,0.6976032,0.0026934615,0.000045430796],"about_ca_topic_score_codex":0.0020234808,"about_ca_topic_score_gemma":0.0012757576,"teacher_disagreement_score":0.033657312,"about_ca_system_score_codex":0.0021561545,"about_ca_system_score_gemma":0.0025850763,"threshold_uncertainty_score":0.17799902},"labels":[],"label_agreement":null},{"id":"W2133594561","doi":"10.1002/sim.5938","title":"Estimating successive cancer risks in Lynch Syndrome families using a progressive three‐state model","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic factors in colorectal cancer","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Calgary; Alberta Health Services; Memorial University of Newfoundland; Public Health Ontario; University of Toronto; Mount Sinai Hospital; Western University","funders":"Canadian Institutes of Health Research","keywords":"Penetrance; Colorectal cancer; Estimator; Cancer; Statistics; Lynch syndrome; Medicine; Genotype; Estimation; Computer science; Genetics; DNA mismatch repair; Mathematics; Biology; Internal medicine; Gene; Phenotype","score_opus":0.060183370546366306,"score_gpt":0.3910755280911346,"score_spread":0.3308921575447683,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2133594561","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.65087277,0.00016442339,0.34695098,0.0004209933,0.000012147378,0.00012183001,0.00042297042,0.00017550348,0.00085832516],"genre_scores_gemma":[0.9606221,0.00016677997,0.036709942,0.000062895924,0.000009466261,0.00015325769,0.0004554543,0.000021424235,0.0017986471],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9985757,0.00084039505,0.000051590257,0.0003074876,0.000086275315,0.0001385782],"domain_scores_gemma":[0.9834283,0.014790239,0.00082299736,0.00045277487,0.00030475485,0.00020083909],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005967664,0.0005852696,0.0010779236,0.00080148847,0.0004273906,0.00090877706,0.001273021,0.001009849,0.0012954669],"category_scores_gemma":[0.015507404,0.00065071636,0.0013679106,0.0006810676,0.0010175132,0.00080627104,0.0010671685,0.0012065156,0.00016133413],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023667252,0.00007562417,0.03893437,0.00002655519,0.00019043271,0.0003037455,0.00031569038,0.93744874,0.00058111746,0.011672139,0.0002227875,0.009992217],"study_design_scores_gemma":[0.000024573335,0.00006359721,0.00468601,0.0000059960066,0.000049886996,0.000058048958,0.000038718284,0.98847896,0.0001552625,0.0062781037,0.00014214993,0.000018689416],"about_ca_topic_score_codex":0.037883952,"about_ca_topic_score_gemma":0.02329762,"teacher_disagreement_score":0.037883952,"about_ca_system_score_codex":0.0013853862,"about_ca_system_score_gemma":0.0012985744,"threshold_uncertainty_score":0.07532686},"labels":[],"label_agreement":null},{"id":"W2133707942","doi":"10.1002/sim.5567","title":"Impact of variability on the choice of biosimilarity limits in assessing follow‐on biologics","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Biosimilars and Bioanalytical Methods","field":"Immunology and Microbiology","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Biosimilar; Bioequivalence; Econometrics; Limit (mathematics); Statistics; Computer science; Economics; Mathematics; Pharmacology; Medicine; Biotechnology; Biology","score_opus":0.09628629694275934,"score_gpt":0.42570780713145306,"score_spread":0.3294215101886937,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2133707942","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.27913138,0.0020349156,0.7115012,0.0010002403,0.00012981214,0.00017790671,0.00016163579,0.00035221266,0.005510687],"genre_scores_gemma":[0.9001717,0.00024434825,0.098767065,0.000213299,0.000062349485,0.000106213505,0.000116886025,0.00009549543,0.00022265314],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.95759296,0.02732181,0.00266305,0.0027063855,0.009168166,0.00054766424],"domain_scores_gemma":[0.7457767,0.22000028,0.011850987,0.013948781,0.007808896,0.00061433145],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04763254,0.0007525269,0.0009441655,0.0012595226,0.000465684,0.0024746845,0.0012303524,0.0015222613,0.0005631241],"category_scores_gemma":[0.16624501,0.00030811797,0.0013226992,0.0011554394,0.0019178304,0.002360568,0.0019830868,0.0019276004,0.00016472347],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0034343381,0.00063906924,0.08903258,0.0007665143,0.0013309163,0.0012086697,0.0011419085,0.40546945,0.055820573,0.080055416,0.0020155518,0.35908496],"study_design_scores_gemma":[0.00024215228,0.0019845716,0.06746147,0.00047634367,0.00053583784,0.0017208128,0.00039197874,0.69127697,0.08965103,0.14028099,0.005545213,0.00043262693],"about_ca_topic_score_codex":0.0005218705,"about_ca_topic_score_gemma":0.0003147784,"teacher_disagreement_score":0.04763254,"about_ca_system_score_codex":0.00072715245,"about_ca_system_score_gemma":0.00080331735,"threshold_uncertainty_score":0.251908},"labels":[],"label_agreement":null},{"id":"W2134646536","doi":"10.1002/sim.2899","title":"Bayesian modelling of tuberculosis clustering from DNA fingerprint data","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Tuberculosis Research and Epidemiology","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Montreal General Hospital; McGill University","funders":"","keywords":"Genotyping; Bayesian probability; Cluster analysis; Categorical variable; Bayes' theorem; Computer science; Statistics; Data mining; Computational biology; Genetics; Biology; Artificial intelligence; Genotype; Mathematics; Machine learning","score_opus":0.10086367248325231,"score_gpt":0.3940453108420791,"score_spread":0.2931816383588268,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2134646536","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1612053,0.0011977793,0.83125645,0.0013431619,0.000070938695,0.00023069122,0.0011719586,0.0005087271,0.0030149897],"genre_scores_gemma":[0.868133,0.0012378268,0.1198919,0.00026112044,0.000120857265,0.0005089053,0.0016404423,0.00018176978,0.00802409],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9973869,0.0012760487,0.000105871266,0.00054692035,0.0003931643,0.00029109613],"domain_scores_gemma":[0.9838171,0.012302533,0.001724009,0.0005922115,0.0011149133,0.00044928322],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009560229,0.00093146163,0.0018633779,0.0020499805,0.0007009626,0.0023973421,0.0029017157,0.0026155568,0.0020047247],"category_scores_gemma":[0.030196868,0.0013993084,0.0015605823,0.0017261329,0.0019337189,0.002076637,0.0015854555,0.0020655221,0.0006379174],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019666525,0.000063096064,0.009695933,0.000113742986,0.00011903474,0.00018827504,0.00041391054,0.92823803,0.0009883015,0.04085912,0.00094425963,0.018179588],"study_design_scores_gemma":[0.000013660576,0.000015095433,0.0014354447,0.000018778546,0.000010848514,0.00002812514,0.000017640541,0.98557854,0.00008950784,0.012446504,0.00032306684,0.000022784992],"about_ca_topic_score_codex":0.036481585,"about_ca_topic_score_gemma":0.020747414,"teacher_disagreement_score":0.036481585,"about_ca_system_score_codex":0.0027593735,"about_ca_system_score_gemma":0.001277674,"threshold_uncertainty_score":0.072538495},"labels":[],"label_agreement":null},{"id":"W2135029321","doi":"10.1002/sim.1301","title":"Issues in the meta‐analysis of cluster randomized trials","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":731,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cancer Care Ontario; Western University","funders":"","keywords":"Meta-analysis; Randomization; Randomized controlled trial; Cluster (spacecraft); Computer science; Research design; Econometrics; Statistics; Medicine; Mathematics; Surgery","score_opus":0.8482658612698941,"score_gpt":0.6110797934992076,"score_spread":0.23718606777068652,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2135029321","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0025790022,0.3635793,0.48228472,0.120329686,0.02318634,0.002663543,0.00075769465,0.001217509,0.0034021912],"genre_scores_gemma":[0.13757119,0.0982587,0.6579384,0.06265447,0.024992933,0.015347057,0.00047426613,0.001386301,0.0013767398],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.11926002,0.7811422,0.059866913,0.009463499,0.029271765,0.0009955857],"domain_scores_gemma":[0.06565815,0.8848082,0.017847255,0.018216617,0.012581572,0.0008882469],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.73829263,0.0058272206,0.02295667,0.018633137,0.004805279,0.018199304,0.014700036,0.0155131845,0.0024535917],"category_scores_gemma":[0.88531834,0.0067354925,0.027047379,0.024850022,0.016047252,0.01511536,0.010352614,0.025785249,0.0010181245],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.005495173,0.00016578974,0.005945275,0.11185629,0.14880854,0.0014673169,0.006941953,0.017201291,0.00057706155,0.23749976,0.0843548,0.37968677],"study_design_scores_gemma":[0.0038464298,0.0012528695,0.004179916,0.049771328,0.057146378,0.0015218772,0.0013594518,0.027756803,0.0020717527,0.78146505,0.06872618,0.0009018634],"about_ca_topic_score_codex":0.006496266,"about_ca_topic_score_gemma":0.008145034,"teacher_disagreement_score":0.26170737,"about_ca_system_score_codex":0.012270263,"about_ca_system_score_gemma":0.01993515,"threshold_uncertainty_score":0.32273185},"labels":[],"label_agreement":null},{"id":"W2136825758","doi":"10.1002/sim.1601","title":"Estimating linear regression models in the presence of a censored independent variable","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":51,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada; Ontario Ministry of Health and Long-Term Care","keywords":"Statistics; Ordinary least squares; Censored regression model; Variables; Mathematics; Linear regression; Regression analysis; Econometrics; Monte Carlo method; Censoring (clinical trials); Local regression; Segmented regression; Polynomial regression","score_opus":0.07086286344301371,"score_gpt":0.41269204990285613,"score_spread":0.3418291864598424,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2136825758","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09841992,0.0006290716,0.8994127,0.00038847097,0.000041624022,0.000102273145,0.00007726154,0.00025917793,0.00066941656],"genre_scores_gemma":[0.5240985,0.0008087063,0.47304258,0.00029373978,0.00008732345,0.00030169848,0.00032689553,0.000071939794,0.0009685023],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.92618495,0.06687321,0.00125626,0.0024301868,0.002668935,0.000586453],"domain_scores_gemma":[0.6724223,0.30720246,0.008972724,0.008295174,0.0027378197,0.00036945855],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06096867,0.0012705748,0.002256812,0.0014703927,0.000592213,0.0021310498,0.0017753671,0.0017609458,0.00096394954],"category_scores_gemma":[0.25164092,0.0011750036,0.0014187531,0.0018020844,0.001607602,0.003234693,0.0022970636,0.0019517664,0.0003147175],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00089562684,0.00035828835,0.05338308,0.0007385308,0.0016064958,0.00071186456,0.0014073038,0.6497557,0.0015498905,0.116550244,0.0010609225,0.171982],"study_design_scores_gemma":[0.00014751308,0.00031481555,0.006781556,0.00014874409,0.00023577854,0.00020392676,0.00016055639,0.8560592,0.0015036653,0.13266827,0.0016723151,0.00010367328],"about_ca_topic_score_codex":0.006317836,"about_ca_topic_score_gemma":0.007409611,"teacher_disagreement_score":0.06096867,"about_ca_system_score_codex":0.0014240185,"about_ca_system_score_gemma":0.002465418,"threshold_uncertainty_score":0.322437},"labels":[],"label_agreement":null},{"id":"W2136972950","doi":"10.1002/sim.4453","title":"Hierarchical priors for bias parameters in Bayesian sensitivity analysis for unmeasured confounding","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University; University of British Columbia; Simon Fraser University","funders":"","keywords":"Prior probability; Covariate; Bayesian probability; Econometrics; Confounding; Computer science; Statistics; Mathematics","score_opus":0.15468768461269244,"score_gpt":0.44139871257239166,"score_spread":0.2867110279596992,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2136972950","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.002432843,0.0004823196,0.9956163,0.00045274358,0.000030416571,0.00012355833,0.00008140525,0.00011940536,0.0006608916],"genre_scores_gemma":[0.2201408,0.001362805,0.7745187,0.00075785763,0.00020835598,0.0014619487,0.00028444428,0.0002327376,0.001032268],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.91438794,0.0756698,0.0020256431,0.0033026726,0.0039402954,0.0006736633],"domain_scores_gemma":[0.67382586,0.30313691,0.0066042743,0.012908725,0.0030136816,0.00051048544],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.12276398,0.001968331,0.0030352785,0.0043070535,0.0014541335,0.0034314676,0.003828986,0.004168617,0.004457063],"category_scores_gemma":[0.3234968,0.0019722783,0.0039150617,0.0034419748,0.0050126216,0.0053542224,0.0047348225,0.007831172,0.00047454092],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002713239,0.00007632637,0.0037665968,0.0009528971,0.0011110522,0.0003393453,0.0007983941,0.25864905,0.0009130624,0.64700437,0.002477358,0.08364014],"study_design_scores_gemma":[0.0001325626,0.00006126328,0.0008642519,0.00031472242,0.00026083714,0.000112568385,0.000054635584,0.22887005,0.0006555423,0.76600444,0.0025947334,0.000074399686],"about_ca_topic_score_codex":0.004327987,"about_ca_topic_score_gemma":0.0036236336,"teacher_disagreement_score":0.12276398,"about_ca_system_score_codex":0.0035217062,"about_ca_system_score_gemma":0.0031311156,"threshold_uncertainty_score":0.64924574},"labels":[],"label_agreement":null},{"id":"W2137538399","doi":"10.1002/sim.5522","title":"Event‐weighted proportional hazards modelling for recurrent gap time data","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; University of Guelph","funders":"","keywords":"Resampling; Proportional hazards model; Statistics; Independence (probability theory); Computer science; Event (particle physics); Marginal model; Econometrics; Mathematics; Inverse; Applied mathematics; Regression analysis","score_opus":0.2797240781651692,"score_gpt":0.47789476376986334,"score_spread":0.19817068560469414,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2137538399","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014024705,0.00015428699,0.9845707,0.00023382579,0.000060834667,0.00020656022,0.00028669817,0.00020548019,0.0002569514],"genre_scores_gemma":[0.49704078,0.0007290827,0.49044654,0.00036525753,0.0002035037,0.003992075,0.0015137779,0.00024687647,0.0054621035],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9830573,0.012306452,0.0006404716,0.0019772628,0.0014726793,0.00054585515],"domain_scores_gemma":[0.94998854,0.039203107,0.0035501518,0.005610943,0.0012562883,0.0003908914],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.043652967,0.0011653585,0.002072401,0.0017961652,0.0006048881,0.0014678689,0.0057999166,0.0022131673,0.0047282143],"category_scores_gemma":[0.086243175,0.0007509732,0.0028968127,0.002247368,0.0018661789,0.0023942194,0.0030755175,0.0041369474,0.0007330239],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0021386736,0.00036679365,0.02735433,0.000906776,0.0016625233,0.0012214242,0.0020485919,0.36496642,0.0030966506,0.39896098,0.004368614,0.19290817],"study_design_scores_gemma":[0.00013326776,0.0003783491,0.003344208,0.00006402196,0.00018797745,0.0002751656,0.0001377948,0.75288045,0.001236845,0.23717351,0.004111275,0.000077156605],"about_ca_topic_score_codex":0.0026921853,"about_ca_topic_score_gemma":0.0018268487,"teacher_disagreement_score":0.043652967,"about_ca_system_score_codex":0.0010142013,"about_ca_system_score_gemma":0.0013961177,"threshold_uncertainty_score":0.23086172},"labels":[],"label_agreement":null},{"id":"W2137903189","doi":"10.1002/sim.6572","title":"The Africa Center for Biostatistical Excellence: a proposal for enhancing biostatistics capacity for sub‐Saharan Africa","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health and Medical Research Impacts","field":"Medicine","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"Fogarty International Center; National Institutes of Health; U.S. Public Health Service; Wellcome Trust; U.S. President’s Emergency Plan for AIDS Relief","keywords":"Biostatistics; Brainstorming; Center of excellence; Economic shortage; Excellence; Capacity building; Medical education; Capacity development; Statistician; Library science; Political science; Medicine; Computer science; Geography; Mathematics; Statistics; Environmental planning; Public health; Artificial intelligence; Nursing","score_opus":0.17510408876753647,"score_gpt":0.43606887980124204,"score_spread":0.26096479103370557,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2137903189","genre_codex":"commentary","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0038834247,0.0035353436,0.024093634,0.92880017,0.024925707,0.005527492,0.00025578943,0.00057088956,0.008407595],"genre_scores_gemma":[0.048387516,0.0065206736,0.3940856,0.48990557,0.009289817,0.012041834,0.00072993094,0.0005324412,0.038506582],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.93813163,0.03008673,0.004612247,0.0032476303,0.011010417,0.012911352],"domain_scores_gemma":[0.7963744,0.037992295,0.00803775,0.006179209,0.03379546,0.117620885],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11905884,0.0024718365,0.0016901884,0.0043616947,0.0072756987,0.016617822,0.0076520485,0.03488111,0.009442821],"category_scores_gemma":[0.1114494,0.002029964,0.004250486,0.0024812284,0.008671242,0.013023916,0.022512743,0.026982091,0.0047254846],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00067320303,0.0018572385,0.0045961114,0.0021843023,0.0001884656,0.0041870526,0.007848572,0.0026483978,0.0048960424,0.11857087,0.75061744,0.10173228],"study_design_scores_gemma":[0.0011375279,0.00091588177,0.005116155,0.0022694662,0.0001534629,0.0008370608,0.005989462,0.0015534469,0.0015508216,0.02243563,0.957696,0.0003450601],"about_ca_topic_score_codex":0.010912164,"about_ca_topic_score_gemma":0.016491516,"teacher_disagreement_score":0.11905884,"about_ca_system_score_codex":0.009114094,"about_ca_system_score_gemma":0.20569167,"threshold_uncertainty_score":0.62965083},"labels":[],"label_agreement":null},{"id":"W2138393952","doi":"10.1002/sim.3095","title":"Construction of confidence limits about effect measures: A general approach","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":252,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"","keywords":"Confidence interval; Computer science; Statistics; Log-normal distribution; Econometrics; Mathematics","score_opus":0.5277622902536679,"score_gpt":0.527346483686242,"score_spread":0.0004158065674259026,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2138393952","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00053523976,0.00054840045,0.9974643,0.00025190169,0.00005931224,0.00019837472,0.000059061396,0.00019602955,0.00068734336],"genre_scores_gemma":[0.046814013,0.0015619872,0.9465151,0.00048135745,0.000446223,0.0030538975,0.0003218,0.00027996345,0.0005257062],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.77886844,0.16739966,0.015682558,0.011091065,0.025364768,0.0015935361],"domain_scores_gemma":[0.37555867,0.56826013,0.011848562,0.025214475,0.018133394,0.0009847729],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.24463311,0.004117253,0.0065523162,0.016417455,0.0017672024,0.012070923,0.0100186765,0.008598731,0.008173862],"category_scores_gemma":[0.6110272,0.0030900813,0.007555156,0.011875269,0.008545684,0.011677547,0.009474022,0.01434381,0.0020633393],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00042515772,0.00013877821,0.0018669958,0.0027881262,0.0015054733,0.0007204249,0.0020460167,0.040134776,0.0011940879,0.71923226,0.0043907203,0.22555713],"study_design_scores_gemma":[0.00032450157,0.00038750476,0.00089528394,0.0016892406,0.00054699724,0.00072546914,0.00026241798,0.099916354,0.002273939,0.87534136,0.017450662,0.0001862237],"about_ca_topic_score_codex":0.0019745212,"about_ca_topic_score_gemma":0.0006066632,"teacher_disagreement_score":0.7553669,"about_ca_system_score_codex":0.0032638328,"about_ca_system_score_gemma":0.005303432,"threshold_uncertainty_score":0.93150187},"labels":[],"label_agreement":null},{"id":"W2138878140","doi":"10.1002/sim.2069","title":"The value of information and optimal clinical trial design","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":186,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Population Health Research Institute; SickKids Foundation; Hospital for Sick Children","funders":"Natural Sciences and Engineering Research Council of Canada; Sick Kids Foundation","keywords":"Type I and type II errors; Sample size determination; Null hypothesis; Value (mathematics); Sample (material); Set (abstract data type); Value of information; Computer science; Perspective (graphical); Statistics; Clinical trial; Medicine; Mathematics; Artificial intelligence; Internal medicine","score_opus":0.5587308049738138,"score_gpt":0.6187845387614808,"score_spread":0.060053733787666985,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2138878140","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0044415765,0.014743169,0.9270771,0.03322632,0.0011232211,0.0032280723,0.00037028868,0.0002288017,0.015561357],"genre_scores_gemma":[0.22159123,0.011927223,0.7372145,0.009704753,0.0022721894,0.014859778,0.00024377044,0.00014721112,0.0020394071],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.48807156,0.47848794,0.010112958,0.0067745186,0.0147017585,0.00185132],"domain_scores_gemma":[0.51010007,0.457734,0.012192645,0.011071906,0.00668196,0.0022194127],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2538993,0.00278958,0.0084427,0.005413027,0.0015888633,0.009079531,0.0042382632,0.010079818,0.0049278764],"category_scores_gemma":[0.47155812,0.0022690345,0.0027458156,0.0041907104,0.014117369,0.010096751,0.0059959656,0.009394662,0.0016056623],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0028150186,0.00021144055,0.0014798823,0.0026762858,0.0010112827,0.0002760562,0.0005277171,0.044802345,0.00024636753,0.795221,0.0073634335,0.14336926],"study_design_scores_gemma":[0.0012379307,0.0005935722,0.0003101188,0.0010609108,0.0002873836,0.00013440849,0.00006181325,0.026285773,0.00028053208,0.9629857,0.0066818437,0.000079891506],"about_ca_topic_score_codex":0.0008697406,"about_ca_topic_score_gemma":0.0006673049,"teacher_disagreement_score":0.74610066,"about_ca_system_score_codex":0.007548877,"about_ca_system_score_gemma":0.01306581,"threshold_uncertainty_score":0.920075},"labels":[],"label_agreement":null},{"id":"W2138986862","doi":"10.1002/sim.5452","title":"Generating survival times to simulate Cox proportional hazards models with time‐varying covariates","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Distribution Estimation and Applications","field":"Mathematics","cited_by":198,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Covariate; Proportional hazards model; Statistics; Weibull distribution; Gompertz function; Mathematics; Statistical model; Econometrics; Computer science","score_opus":0.06665911353813758,"score_gpt":0.3846642828112441,"score_spread":0.3180051692731065,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2138986862","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.030316757,0.0001402072,0.9659995,0.00036404055,0.00006855208,0.00045409027,0.0005706556,0.000602116,0.0014841392],"genre_scores_gemma":[0.37660256,0.000540782,0.61260784,0.00045564232,0.00006689229,0.00348267,0.0022942084,0.00028263833,0.003666778],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9957432,0.003089623,0.00020766411,0.0003533555,0.00043788465,0.0001682368],"domain_scores_gemma":[0.9463238,0.047455702,0.002166143,0.0020006655,0.0017102036,0.00034339415],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013742909,0.0008771619,0.000987488,0.0013547763,0.00047372744,0.0011900074,0.0020860615,0.0015558626,0.00673787],"category_scores_gemma":[0.06181479,0.0006107449,0.001374934,0.001330083,0.0010954646,0.0014024442,0.0011693815,0.0028537076,0.00094754045],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023016064,0.00011286947,0.0044671306,0.00010007667,0.0000728067,0.00015825439,0.00024691195,0.8820576,0.00057237403,0.08739276,0.0018599707,0.022729164],"study_design_scores_gemma":[0.0000780585,0.00006864387,0.0003253694,0.000034403158,0.000020295385,0.000049182872,0.000032841468,0.9577894,0.00081817596,0.03936449,0.0013965853,0.000022523336],"about_ca_topic_score_codex":0.003922246,"about_ca_topic_score_gemma":0.003447239,"teacher_disagreement_score":0.013742909,"about_ca_system_score_codex":0.0014071454,"about_ca_system_score_gemma":0.0018825461,"threshold_uncertainty_score":0.072680295},"labels":[],"label_agreement":null},{"id":"W2139017261","doi":"10.1002/sim.6265","title":"STRengthening Analytical Thinking for Observational Studies: the STRATOS initiative","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":135,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Economic and Social Research Council; Medical Research Council; Cancer Research UK; McGill University","keywords":"Observational study; Computer science; Econometrics; Data science; Management science; Statistics; Mathematics; Economics","score_opus":0.8755938676556354,"score_gpt":0.6135030834304239,"score_spread":0.2620907842252115,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2139017261","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0020828766,0.055461954,0.290276,0.6112762,0.01956657,0.003138135,0.0011640026,0.0033679667,0.013666369],"genre_scores_gemma":[0.012056978,0.020227615,0.89474523,0.054985244,0.007174719,0.0064681647,0.0011997101,0.0010349744,0.002107276],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.3701505,0.50715697,0.04951082,0.009240393,0.06025872,0.0036825405],"domain_scores_gemma":[0.13740051,0.7038296,0.03305851,0.037952654,0.0703753,0.017383337],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5997813,0.0027941177,0.0041728197,0.015236354,0.0027934667,0.018339345,0.008295348,0.010739701,0.00400746],"category_scores_gemma":[0.6437423,0.0025932738,0.00640218,0.010950809,0.013801003,0.014506142,0.03375689,0.024778383,0.0022513578],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00047329054,0.00019889075,0.0019212402,0.013214503,0.001172517,0.00040465008,0.007833522,0.0029890174,0.0007014501,0.2701397,0.2698875,0.43106368],"study_design_scores_gemma":[0.00060496037,0.00025533343,0.0014881222,0.03272739,0.0006076242,0.00033499152,0.0014255949,0.003613395,0.000770407,0.35701078,0.6009552,0.00020621948],"about_ca_topic_score_codex":0.0043304595,"about_ca_topic_score_gemma":0.0064660045,"teacher_disagreement_score":0.40021873,"about_ca_system_score_codex":0.012302287,"about_ca_system_score_gemma":0.10833733,"threshold_uncertainty_score":0.493541},"labels":[],"label_agreement":null},{"id":"W2139115743","doi":"10.1002/sim.6633","title":"Knowledge translation in biostatistics: a survey of current practices, preferences, and barriers to the dissemination and uptake of new statistical methods","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McMaster University; Institute for Work & Health; Institute of Health Services and Policy Research; McGill University; Institute for Clinical Evaluative Sciences; University of Toronto; SickKids Foundation; Hospital for Sick Children; Public Health Ontario","funders":"Canadian Institutes of Health Research","keywords":"Biostatistics; Statistical software; Computer science; Knowledge translation; Medical education; Perception; Principal (computer security); Data science; Psychology; Public health; Knowledge management; Medicine; Nursing","score_opus":0.8373272864966954,"score_gpt":0.65655898842761,"score_spread":0.18076829806908545,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2139115743","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9918743,0.001208898,0.00046827443,0.0035977524,0.000021174079,0.0001700964,0.00030031777,0.00001766224,0.0023416323],"genre_scores_gemma":[0.9943921,0.0020320031,0.0010301364,0.00156854,0.00002508528,0.00010809329,0.00020362504,0.000012098204,0.0006283222],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9856113,0.0048087128,0.001579273,0.0005707403,0.006003541,0.0014263978],"domain_scores_gemma":[0.90556556,0.043416634,0.02120856,0.0019560296,0.019390596,0.008462539],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02283813,0.00017368914,0.00048083384,0.0028212871,0.0013593852,0.0026063493,0.0009181332,0.0012472211,0.0024537598],"category_scores_gemma":[0.07316956,0.00048354547,0.0004939945,0.0034241735,0.0020619682,0.0021014542,0.0018151436,0.0013907837,0.0004213453],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012792149,0.00023692171,0.8620387,0.0007836287,0.000053388605,0.00043891385,0.07717307,0.000113483475,0.0007171959,0.0002200885,0.0036301538,0.05446659],"study_design_scores_gemma":[0.0000415446,0.0005175245,0.83143955,0.0009046715,0.000044744233,0.0009087911,0.15131389,0.0009118657,0.00029297188,0.00028278926,0.013268168,0.00007358084],"about_ca_topic_score_codex":0.08153422,"about_ca_topic_score_gemma":0.060752176,"teacher_disagreement_score":0.9771619,"about_ca_system_score_codex":0.0043336228,"about_ca_system_score_gemma":0.013256056,"threshold_uncertainty_score":0.16211927},"labels":[],"label_agreement":null},{"id":"W2140085139","doi":"10.1002/sim.3552","title":"Bayesian adjustment for covariate measurement errors: A flexible parametric approach","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Covariate; Bayesian probability; Statistics; Parametric statistics; Computer science; Econometrics; Semiparametric model; Observational error; Mathematics","score_opus":0.17226786948241857,"score_gpt":0.41952464059558425,"score_spread":0.24725677111316569,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2140085139","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0014331925,0.00017263474,0.99786216,0.00012250569,0.000022534858,0.000048398302,0.00003645458,0.00012986777,0.00017225501],"genre_scores_gemma":[0.1937344,0.001015952,0.80018353,0.0005204269,0.00032450847,0.0008575795,0.0005960272,0.00031055015,0.002457056],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.978308,0.015333068,0.0007529999,0.0025131328,0.002453308,0.0006394629],"domain_scores_gemma":[0.93676114,0.050195593,0.0032928123,0.0060697887,0.0032148047,0.0004658164],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03519347,0.0017815374,0.0031560638,0.0024290415,0.0013834039,0.0022128737,0.006269305,0.0028679303,0.0031684344],"category_scores_gemma":[0.11739619,0.0015766008,0.0035882418,0.0037404026,0.0025979003,0.003884283,0.00466786,0.0053171967,0.00079391815],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00048942247,0.00020024827,0.008535261,0.0006004284,0.00131316,0.00065908505,0.0009989714,0.4717046,0.0018590906,0.18474224,0.004917161,0.32398036],"study_design_scores_gemma":[0.00009586139,0.000107341955,0.0016650102,0.00011058987,0.00021555816,0.00021699784,0.00007143379,0.8465583,0.0007231845,0.14583407,0.0043132934,0.00008841176],"about_ca_topic_score_codex":0.008193914,"about_ca_topic_score_gemma":0.0062805354,"teacher_disagreement_score":0.03519347,"about_ca_system_score_codex":0.0013890319,"about_ca_system_score_gemma":0.0041013625,"threshold_uncertainty_score":0.18612307},"labels":[],"label_agreement":null},{"id":"W2141260650","doi":"10.1002/sim.2572","title":"Longitudinal variable selection by cross‐validation in the case of many covariates","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"","keywords":"Covariate; Computer science; Markov chain Monte Carlo; Variable (mathematics); Feature selection; Selection (genetic algorithm); Model selection; Markov chain; Scale (ratio); Variety (cybernetics); Econometrics; Marginal model; Statistics; Data mining; Machine learning; Regression analysis; Artificial intelligence; Mathematics; Bayesian probability","score_opus":0.04109148622515056,"score_gpt":0.40199404916491205,"score_spread":0.3609025629397615,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2141260650","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00555255,0.00037435218,0.992955,0.00024473027,0.00006846029,0.00007395513,0.000064218046,0.00025461422,0.00041209813],"genre_scores_gemma":[0.16368327,0.00053318706,0.8317836,0.00042367404,0.00019051592,0.00078354817,0.0005794117,0.00022267344,0.0018001683],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9507992,0.044145886,0.0009315206,0.002111357,0.0015856357,0.0004262584],"domain_scores_gemma":[0.90756375,0.07755186,0.002615385,0.009177942,0.0026023888,0.0004887296],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.070951335,0.0013439064,0.0020978726,0.0023598627,0.0015263392,0.0016067657,0.002647209,0.0029671665,0.0033371544],"category_scores_gemma":[0.13582636,0.00090889534,0.0019189818,0.0028002106,0.002258853,0.0023314862,0.0028010001,0.004087268,0.0009988962],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00060742506,0.00035948283,0.02613168,0.0005163834,0.0021589869,0.0012739215,0.0009064885,0.4964702,0.0017412539,0.16269481,0.010839459,0.29629993],"study_design_scores_gemma":[0.0000910461,0.00019103156,0.0027604073,0.0001555345,0.00013136145,0.00029449188,0.00008790833,0.85672045,0.0013174637,0.1313119,0.0068754614,0.000063019135],"about_ca_topic_score_codex":0.004980221,"about_ca_topic_score_gemma":0.0055903993,"teacher_disagreement_score":0.070951335,"about_ca_system_score_codex":0.00069393637,"about_ca_system_score_gemma":0.0023267693,"threshold_uncertainty_score":0.37523103},"labels":[],"label_agreement":null},{"id":"W2141508928","doi":"10.1002/sim.1794","title":"Regression methods for cost-effectiveness analysis with censored data","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":92,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; Population Health Research Institute; University of Toronto; Hospital for Sick Children","funders":"National Cancer Institute; Natural Sciences and Engineering Research Council of Canada","keywords":"Censoring (clinical trials); Statistics; Weighting; Regression analysis; Inverse probability weighting; Survival analysis; Regression; Inverse probability; Econometrics; Proportional hazards model; Computer science; Censored regression model; Mathematics; Medicine; Propensity score matching","score_opus":0.49929086056699135,"score_gpt":0.5848901660240776,"score_spread":0.08559930545708627,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2141508928","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00028664435,0.00040980786,0.99865746,0.00011031517,0.00004145709,0.00007888496,0.00007979428,0.0001690863,0.00016658826],"genre_scores_gemma":[0.030541198,0.0022629777,0.9613883,0.00025185855,0.0003288259,0.0021629985,0.0008132984,0.00033378732,0.0019168351],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9547293,0.038140863,0.0015811543,0.0023061212,0.0029089202,0.00033354067],"domain_scores_gemma":[0.93885076,0.052441083,0.0028800413,0.0033156849,0.0023067393,0.00020569195],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04229507,0.0026102555,0.003685329,0.004887638,0.0006679196,0.0021142361,0.0037624135,0.0019698003,0.0057306136],"category_scores_gemma":[0.10206312,0.0014089972,0.0041760486,0.0054141744,0.0011063811,0.0021399518,0.0023499506,0.005198514,0.0027038546],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019568499,0.00022925857,0.0038347042,0.0012703125,0.0024354311,0.00030603263,0.00035086946,0.16916488,0.0016217945,0.41499418,0.011983668,0.3936131],"study_design_scores_gemma":[0.0001977039,0.00021591838,0.0019780076,0.00029021513,0.00043372603,0.00028194496,0.00004595195,0.6704718,0.00090072834,0.30705905,0.017961372,0.00016354943],"about_ca_topic_score_codex":0.0025670265,"about_ca_topic_score_gemma":0.0019024241,"teacher_disagreement_score":0.04229507,"about_ca_system_score_codex":0.0012667497,"about_ca_system_score_gemma":0.0020688213,"threshold_uncertainty_score":0.22368038},"labels":[],"label_agreement":null},{"id":"W2141588553","doi":"10.1002/sim.2662","title":"The merits of breaking the matches: a cautionary tale","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Matching (statistics); Outcome (game theory); Computer science; Econometrics; Cluster (spacecraft); Research design; Test (biology); Randomization; Statistics; Intervention (counseling); Restricted randomization; Scale (ratio); Mathematics; Clinical trial; Psychology; Medicine","score_opus":0.037239099333314964,"score_gpt":0.3812329515603517,"score_spread":0.3439938522270367,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2141588553","genre_codex":"commentary","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0023431175,0.023717966,0.068168715,0.8571701,0.041634277,0.00055937667,0.00030627864,0.00042027613,0.005679865],"genre_scores_gemma":[0.039018795,0.012710153,0.14731094,0.7347204,0.057214536,0.002249961,0.00009674308,0.00040124063,0.00627722],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8576683,0.09465544,0.012265663,0.009536369,0.024868187,0.0010060187],"domain_scores_gemma":[0.52129185,0.41689757,0.007182166,0.027735664,0.023477776,0.0034150034],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.22706033,0.0025119246,0.0054848874,0.0033289902,0.004554937,0.007749958,0.011595438,0.0170384,0.00447186],"category_scores_gemma":[0.48748514,0.0011298772,0.0036875359,0.0028322,0.037434608,0.017572988,0.006592188,0.059356842,0.0028964407],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020204606,0.00021160668,0.0029389467,0.0029047043,0.0012221583,0.0014568336,0.004455269,0.0021022202,0.00084966864,0.3504765,0.48893848,0.1424231],"study_design_scores_gemma":[0.00069680373,0.00041652084,0.001858395,0.0033693542,0.00035295764,0.0010896669,0.0018659275,0.0042555723,0.0010544134,0.723199,0.26148984,0.00035162637],"about_ca_topic_score_codex":0.0053323545,"about_ca_topic_score_gemma":0.006323943,"teacher_disagreement_score":0.7729397,"about_ca_system_score_codex":0.0034743042,"about_ca_system_score_gemma":0.0055215796,"threshold_uncertainty_score":0.9531722},"labels":[],"label_agreement":null},{"id":"W2141848375","doi":"10.1002/sim.5951","title":"Multiple‐objective response‐adaptive repeated measurement designs in clinical trials for binary responses","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Binary number; Computer science; Basis (linear algebra); Binary data; Function (biology); Mathematical optimization; Algorithm; Mathematics","score_opus":0.716823768290405,"score_gpt":0.608469569222505,"score_spread":0.1083541990679,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2141848375","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0068280385,0.00040015788,0.9892681,0.00027026812,0.00008773075,0.0024685431,0.00004620053,0.00015628066,0.00047470187],"genre_scores_gemma":[0.12731239,0.0003021806,0.85776395,0.00041560334,0.00008081155,0.013520487,0.00006234606,0.00003746235,0.00050473254],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7095317,0.27616587,0.0037939432,0.004210713,0.0054950584,0.00080256176],"domain_scores_gemma":[0.8964531,0.08541151,0.007839415,0.0072952243,0.0024930495,0.00050774],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.13113521,0.0019716085,0.0029991074,0.0022773643,0.00051394495,0.0013185247,0.0024785697,0.003010269,0.003785393],"category_scores_gemma":[0.1527485,0.0013244534,0.002669043,0.0016181113,0.0028144056,0.001540545,0.0015936238,0.0035091315,0.0007481463],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009793357,0.0021088733,0.003713897,0.004472901,0.0031206398,0.00034747378,0.0013313808,0.19889307,0.008104833,0.32619283,0.0031848014,0.43873587],"study_design_scores_gemma":[0.010178064,0.015325303,0.003740871,0.0008076056,0.0009355273,0.00026479343,0.00011022001,0.639492,0.0128042,0.3058377,0.010184706,0.00031900257],"about_ca_topic_score_codex":0.00021309043,"about_ca_topic_score_gemma":0.0002640077,"teacher_disagreement_score":0.13113521,"about_ca_system_score_codex":0.0013317524,"about_ca_system_score_gemma":0.002212423,"threshold_uncertainty_score":0.69351757},"labels":[],"label_agreement":null},{"id":"W2142200036","doi":"10.1002/sim.3110","title":"Reliability analysis for continuous measurements: Equivalence test for agreement","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Western Hospital; University Health Network; Memorial University of Newfoundland; Canadian Blood Services; University of Toronto","funders":"","keywords":"Repeatability; Equivalence (formal languages); Reliability (semiconductor); Inter-rater reliability; Computer science; Reliability engineering; Intra-rater reliability; Statistics; Test (biology); Consistency (knowledge bases); Mathematics; Rating scale; Artificial intelligence; Engineering","score_opus":0.20769671851629035,"score_gpt":0.4449493606354994,"score_spread":0.23725264211920905,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2142200036","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.022663984,0.0016050318,0.9627304,0.00045544494,0.00047481697,0.0034517588,0.00073841243,0.00050280354,0.0073773637],"genre_scores_gemma":[0.40047607,0.0010605893,0.5779858,0.000394271,0.00036695955,0.016708303,0.0011974993,0.0005557671,0.0012546061],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6621769,0.24161531,0.021819903,0.020460978,0.052121587,0.0018052773],"domain_scores_gemma":[0.386167,0.5346315,0.016359791,0.02859927,0.03315164,0.0010907608],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.21806082,0.0017636183,0.0047329627,0.00894103,0.0017093739,0.0032660798,0.003075668,0.0023308308,0.0052670618],"category_scores_gemma":[0.5577832,0.00095688493,0.0049238116,0.008624847,0.0053862967,0.0059233936,0.0055586873,0.00462281,0.0015519756],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004907642,0.0008974779,0.08637467,0.0101601165,0.00964405,0.0007209774,0.015450145,0.02454078,0.0047514043,0.18813539,0.016967436,0.6374499],"study_design_scores_gemma":[0.0018002593,0.010333047,0.14354296,0.004874429,0.00464783,0.0031456558,0.007238552,0.24099806,0.008948206,0.49896193,0.07434412,0.0011648976],"about_ca_topic_score_codex":0.001116363,"about_ca_topic_score_gemma":0.0007126562,"teacher_disagreement_score":0.21806082,"about_ca_system_score_codex":0.002057025,"about_ca_system_score_gemma":0.0044190898,"threshold_uncertainty_score":0.96427023},"labels":[],"label_agreement":null},{"id":"W2143015220","doi":"10.1002/sim.4328","title":"Information in the sample covariate distribution in prevalent cohorts: Addendum","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Nutritional Studies and Diet","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Waterloo","funders":"Medical Research Council; Health Canada; Pfizer Canada; University of Ottawa; Pfizer","keywords":"Addendum; Library science; Acknowledgement; Sample (material); Gerontology; Medicine; Political science; Family medicine; Law","score_opus":0.04777589730579668,"score_gpt":0.3156718643223418,"score_spread":0.2678959670165451,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2143015220","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0134387985,0.19709206,0.32639924,0.21230786,0.21212906,0.00042153598,0.022732368,0.0011072343,0.014371867],"genre_scores_gemma":[0.09889043,0.15004665,0.14652348,0.08616,0.47579458,0.0012811021,0.019198438,0.0014464116,0.020658867],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9865159,0.008650471,0.00092629425,0.001357134,0.0021847675,0.0003653875],"domain_scores_gemma":[0.8725126,0.10832875,0.0033446962,0.009020833,0.005931531,0.00086151296],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.016068868,0.0014319809,0.0020412384,0.0054363506,0.00089658215,0.0023202368,0.0032262972,0.00248169,0.0194039],"category_scores_gemma":[0.13483545,0.0009218938,0.003316457,0.0058674463,0.0013227648,0.0049818656,0.0031726223,0.005383738,0.004103131],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006222498,0.00021678113,0.018261375,0.00401362,0.0016078403,0.0026273043,0.0004474097,0.0041930424,0.0007050177,0.028457867,0.5935857,0.34526184],"study_design_scores_gemma":[0.00016495651,0.0009512501,0.035878576,0.0037314426,0.0017235767,0.00968283,0.00031153712,0.016867096,0.0026109396,0.1870902,0.7402564,0.0007312609],"about_ca_topic_score_codex":0.0033669726,"about_ca_topic_score_gemma":0.0041993735,"teacher_disagreement_score":0.9839311,"about_ca_system_score_codex":0.0012682551,"about_ca_system_score_gemma":0.0017706234,"threshold_uncertainty_score":0.08498126},"labels":[],"label_agreement":null},{"id":"W2143134812","doi":"10.1002/sim.4180","title":"Information in the sample covariate distribution in prevalent cohorts","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Ottawa; University of Waterloo","funders":"Canadian Institutes of Health Research","keywords":"Covariate; Statistics; Sample (material); Econometrics; Computer science; Mathematics","score_opus":0.13063176353298864,"score_gpt":0.39900020257419255,"score_spread":0.2683684390412039,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2143134812","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09093102,0.0018380372,0.9034881,0.0015876901,0.000046115558,0.00016479782,0.0005247313,0.00017128339,0.0012481933],"genre_scores_gemma":[0.7765704,0.0031709275,0.21495728,0.00085913955,0.00022843223,0.0006811647,0.0010814795,0.00009524125,0.0023559006],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97130966,0.02146732,0.0012193365,0.002259045,0.0031486338,0.00059598824],"domain_scores_gemma":[0.6950453,0.2658269,0.013016202,0.021187682,0.004257152,0.00066678313],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08932901,0.00057904737,0.001659098,0.0024260161,0.00088079716,0.0026578698,0.002238018,0.0020261507,0.0022847957],"category_scores_gemma":[0.3310021,0.0010348675,0.0014051648,0.0026235937,0.0036700848,0.0044144974,0.003835851,0.0027178945,0.00035964613],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009238453,0.00012438522,0.168507,0.000841836,0.00081104477,0.00123335,0.0028751867,0.12438023,0.0018713265,0.50480986,0.0026838782,0.19093806],"study_design_scores_gemma":[0.00016175052,0.00025718202,0.04088551,0.00042333582,0.0003153125,0.0007431922,0.0003705878,0.25569788,0.0016171483,0.693472,0.0059547382,0.000101417805],"about_ca_topic_score_codex":0.0051521026,"about_ca_topic_score_gemma":0.0043471325,"teacher_disagreement_score":0.910671,"about_ca_system_score_codex":0.0018140839,"about_ca_system_score_gemma":0.0022924608,"threshold_uncertainty_score":0.4724226},"labels":[],"label_agreement":null},{"id":"W2143710665","doi":"10.1002/sim.5529","title":"Mixture distributions in multi‐state modelling: Some considerations in a study of psoriatic arthritis","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"Medical Research Council","keywords":"Mixture model; Econometrics; Poisson distribution; Psoriatic arthritis; Computer science; Population; Inverse Gaussian distribution; Statistics; Mathematics; Arthritis; Medicine","score_opus":0.05109171537922794,"score_gpt":0.3417578543330217,"score_spread":0.2906661389537938,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2143710665","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.016058775,0.0040857675,0.96806693,0.008198844,0.00024134887,0.00016858395,0.00015614234,0.0001653157,0.002858316],"genre_scores_gemma":[0.36874098,0.007021179,0.6111444,0.002487958,0.0011432377,0.0016036998,0.00039984778,0.00020995567,0.0072488063],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98555195,0.011991288,0.0004020259,0.00085754116,0.00089244737,0.00030472054],"domain_scores_gemma":[0.9184251,0.07753323,0.0012127759,0.0012546672,0.0009973701,0.00057682546],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03413722,0.0011975414,0.0034096015,0.002218469,0.0016034555,0.0041263993,0.004302786,0.0064919805,0.0027001745],"category_scores_gemma":[0.06913113,0.0016516958,0.0036484143,0.0034049724,0.003991533,0.004375225,0.0035531356,0.0061493916,0.0005386817],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00014031897,0.00011290924,0.0067383493,0.00024060594,0.00028438217,0.0007293623,0.0013092051,0.38566515,0.00033958326,0.57845885,0.001805356,0.024175921],"study_design_scores_gemma":[0.000047304886,0.0000801336,0.00094013877,0.00012369845,0.000056054734,0.0001947069,0.00022441681,0.67685705,0.000072967276,0.3175622,0.0037531385,0.00008820835],"about_ca_topic_score_codex":0.02431355,"about_ca_topic_score_gemma":0.029424716,"teacher_disagreement_score":0.03413722,"about_ca_system_score_codex":0.0023782165,"about_ca_system_score_gemma":0.0023156286,"threshold_uncertainty_score":0.18053705},"labels":[],"label_agreement":null},{"id":"W2143887617","doi":"10.1002/sim.2748","title":"A new estimation method for the semiparametric accelerated failure time mixture cure model","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":97,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Memorial University of Newfoundland","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Mixture model; Identifiability; Accelerated failure time model; Computer science; Estimator; Parametric statistics; Semiparametric model; Semiparametric regression; Rank (graph theory); Nonparametric statistics; Estimation; Covariate; Mathematics; Statistics; Artificial intelligence; Machine learning","score_opus":0.07505516488088816,"score_gpt":0.42820516346295245,"score_spread":0.3531499985820643,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2143887617","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.001007324,0.000072463685,0.998541,0.000052130457,0.000014932266,0.000020376068,0.000032934786,0.00010035843,0.00015845134],"genre_scores_gemma":[0.11076334,0.00067416055,0.88278633,0.0002145466,0.00023450349,0.00064871425,0.00091571535,0.00021582842,0.0035468538],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9960162,0.0023217674,0.00018407979,0.0005699995,0.00075432484,0.00015361913],"domain_scores_gemma":[0.99244153,0.0050512687,0.00063361094,0.0006237698,0.0011076642,0.00014204207],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007855038,0.0010328356,0.0017751701,0.0021736645,0.0004416727,0.0011057878,0.002401018,0.0015638251,0.004190784],"category_scores_gemma":[0.022577066,0.00082108576,0.0020864,0.0016975042,0.000867725,0.0023302254,0.0021071145,0.0029522607,0.0012133272],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002689713,0.00016943783,0.0046889656,0.00048265123,0.00044957138,0.0002609032,0.00036004916,0.35466075,0.005055308,0.15111157,0.0066043497,0.4758875],"study_design_scores_gemma":[0.00004125401,0.0000593444,0.00079261773,0.000035593246,0.00006058791,0.00021641981,0.000025484282,0.96888316,0.00092980656,0.024165075,0.0047402503,0.000050424485],"about_ca_topic_score_codex":0.0021657094,"about_ca_topic_score_gemma":0.0015606381,"teacher_disagreement_score":0.007855038,"about_ca_system_score_codex":0.00067740143,"about_ca_system_score_gemma":0.0016418463,"threshold_uncertainty_score":0.041541934},"labels":[],"label_agreement":null},{"id":"W2144168223","doi":"10.1002/sim.5984","title":"The use of propensity score methods with survival or time‐to‐event outcomes: reporting measures of effect similar to those used in randomized experiments","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1441,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Observational study; Marginal structural model; Inverse probability weighting; Medicine; Covariate; Randomized controlled trial; Confounding; Average treatment effect; Statistics; Hazard ratio; Selection bias; Population; Survival analysis; Matching (statistics); Proportional hazards model; Confidence interval; Internal medicine; Mathematics","score_opus":0.45447001760427186,"score_gpt":0.5290019207742545,"score_spread":0.0745319031699826,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2144168223","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0007856977,0.0074109593,0.9824929,0.0023354762,0.0010918033,0.0013159605,0.001651181,0.000509948,0.0024059967],"genre_scores_gemma":[0.03131308,0.015987866,0.9277407,0.004357662,0.0027333263,0.012668828,0.0022718462,0.0008063878,0.00212038],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.63943064,0.2924809,0.02367798,0.010842903,0.03267649,0.0008911327],"domain_scores_gemma":[0.429904,0.4315079,0.056376863,0.06826769,0.01300956,0.0009340929],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.21747023,0.0041576964,0.004999245,0.010260479,0.0015755405,0.0082487855,0.006004136,0.008134787,0.010013883],"category_scores_gemma":[0.5252035,0.0018896988,0.008818037,0.018248314,0.0061053378,0.012372686,0.008749089,0.012322884,0.0034431724],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004632436,0.00022067092,0.007671661,0.011343051,0.0060817325,0.00021697623,0.0017149311,0.019727986,0.0010024324,0.55355334,0.055481337,0.34252262],"study_design_scores_gemma":[0.0006025329,0.0006282846,0.004716466,0.0050013294,0.0018165468,0.0007356432,0.00039693873,0.034774665,0.0034194917,0.78427774,0.16314241,0.00048801024],"about_ca_topic_score_codex":0.0029941082,"about_ca_topic_score_gemma":0.0019002688,"teacher_disagreement_score":0.7825298,"about_ca_system_score_codex":0.003038803,"about_ca_system_score_gemma":0.008339749,"threshold_uncertainty_score":0.96499854},"labels":[],"label_agreement":null},{"id":"W2144420447","doi":"10.1002/sim.1099","title":"Properties of the summary receiver operating characteristic (SROC) curve for diagnostic test data","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Ovarian cancer diagnosis and treatment","field":"Medicine","cited_by":694,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"Canadian Institutes of Health Research","keywords":"Receiver operating characteristic; Diagnostic odds ratio; Area under the curve; Statistics; Homogeneous; Meta-analysis; Odds ratio; Area under curve; Coverage probability; Standard error; Mathematics; Medicine; Confidence interval; Internal medicine; Combinatorics","score_opus":0.08867909890496116,"score_gpt":0.31968617351874695,"score_spread":0.23100707461378578,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2144420447","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.049537703,0.081837505,0.8478735,0.0050774375,0.0007023217,0.0013634305,0.004823257,0.002173863,0.006610891],"genre_scores_gemma":[0.7465599,0.015489857,0.22142316,0.0040599974,0.0012748475,0.0031662832,0.0058548967,0.0011000348,0.0010709438],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.82942617,0.13361956,0.0115103675,0.011578042,0.012682069,0.0011837821],"domain_scores_gemma":[0.27895913,0.64087474,0.042117264,0.02456966,0.012537418,0.0009417826],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.21378648,0.0022956282,0.0059440862,0.011714365,0.0007488153,0.0046078246,0.0027813867,0.004419207,0.0021288411],"category_scores_gemma":[0.56279063,0.0009767555,0.008485937,0.009762902,0.0038171096,0.0055694,0.0032058645,0.0038773592,0.0010205329],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004695915,0.00017853736,0.23860902,0.01615573,0.0406846,0.001629629,0.0019805208,0.21857841,0.0021732089,0.06749182,0.01889123,0.38893142],"study_design_scores_gemma":[0.0008034006,0.0041627795,0.12526396,0.0067589083,0.018306281,0.008458802,0.0011497064,0.51054794,0.0036426191,0.27407122,0.045846578,0.0009878373],"about_ca_topic_score_codex":0.0019233641,"about_ca_topic_score_gemma":0.0006943571,"teacher_disagreement_score":0.9940559,"about_ca_system_score_codex":0.00236818,"about_ca_system_score_gemma":0.0018483661,"threshold_uncertainty_score":0.96954125},"labels":[],"label_agreement":null},{"id":"W2144443379","doi":"10.1002/sim.1613","title":"A Bayesian analysis of the 4‐year follow‐up data of the Wisconsin epidemiologic study of diabetic retinopathy","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"National Eye Institute","keywords":"Covariate; Bayesian probability; Baseline (sea); Diabetic retinopathy; Medicine; Population; Data set; Statistics; Computer science; Demography; Mathematics; Diabetes mellitus; Environmental health","score_opus":0.1301802120599601,"score_gpt":0.4246668987784154,"score_spread":0.2944866867184553,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2144443379","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.45351487,0.0027674478,0.5369327,0.001818972,0.00006316984,0.00023596584,0.0027217597,0.00017645817,0.0017686212],"genre_scores_gemma":[0.8272296,0.0022163275,0.16281311,0.00025578993,0.0001316189,0.00036914807,0.0048479126,0.000061890714,0.002074545],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99300283,0.005432715,0.0001861809,0.0005582487,0.00058556465,0.00023440763],"domain_scores_gemma":[0.96226186,0.03177569,0.0024843304,0.0016989205,0.001446091,0.00033322826],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026186204,0.000609116,0.0011627271,0.0026330673,0.00075509545,0.0012352742,0.0011762975,0.0009736759,0.0010537133],"category_scores_gemma":[0.07750899,0.0006015317,0.0014871471,0.0020914318,0.0007847411,0.0011708395,0.0011679166,0.0015896887,0.00018408167],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019031406,0.00041099233,0.1329163,0.00062087976,0.0017436368,0.0009678887,0.0015302377,0.4427278,0.004109716,0.16065343,0.0070429174,0.24537307],"study_design_scores_gemma":[0.00018093872,0.0003695531,0.073044725,0.00024907608,0.00073521124,0.00033180404,0.0003276875,0.79417294,0.0009759744,0.12276889,0.0066648056,0.00017837237],"about_ca_topic_score_codex":0.020730782,"about_ca_topic_score_gemma":0.020076806,"teacher_disagreement_score":0.026186204,"about_ca_system_score_codex":0.0010653767,"about_ca_system_score_gemma":0.0018390389,"threshold_uncertainty_score":0.13848752},"labels":[],"label_agreement":null},{"id":"W2145394717","doi":"10.1002/sim.4373","title":"Short‐term cancer mortality projections: A comparative study of prediction methods","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Colorectal Cancer Screening and Detection","field":"Medicine","cited_by":39,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Public Health Agency of Canada; HIV Legal Network; Simon Fraser University","funders":"","keywords":"Term (time); Cancer; Statistics; Computer science; Econometrics; Mathematics; Medicine; Internal medicine","score_opus":0.22206470472513787,"score_gpt":0.4958846590628696,"score_spread":0.27381995433773176,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2145394717","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.42960623,0.034692306,0.5020197,0.0040514795,0.0005372877,0.0010644607,0.008074559,0.0019437394,0.01801019],"genre_scores_gemma":[0.82798696,0.0113059655,0.15099052,0.0002544405,0.00025667466,0.0007725252,0.006629342,0.00047237435,0.0013312113],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9654868,0.025040615,0.0017111669,0.0016094836,0.005846938,0.00030506548],"domain_scores_gemma":[0.8209467,0.15314029,0.004917604,0.007484212,0.012768469,0.000742746],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.070966154,0.0009244173,0.0010824786,0.0040054168,0.00047190097,0.00183875,0.0019451391,0.0010148953,0.0021718226],"category_scores_gemma":[0.11495943,0.00064899825,0.0018444923,0.003606981,0.00051488716,0.0029136823,0.0026527378,0.0014256346,0.00044172764],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0025780003,0.00037023958,0.17303962,0.00207631,0.0027892818,0.00009754806,0.0009903996,0.21689013,0.00053679355,0.014314717,0.005999821,0.58031714],"study_design_scores_gemma":[0.00040766987,0.0024019885,0.22147034,0.002194105,0.0013802202,0.00055115565,0.0013539678,0.73091096,0.0028184121,0.017329276,0.018794702,0.00038725056],"about_ca_topic_score_codex":0.006906739,"about_ca_topic_score_gemma":0.005899114,"teacher_disagreement_score":0.070966154,"about_ca_system_score_codex":0.0014628043,"about_ca_system_score_gemma":0.0020574331,"threshold_uncertainty_score":0.3753094},"labels":[],"label_agreement":null},{"id":"W2146976162","doi":"10.1002/(sici)1097-0258(20000215)19:3<373::aid-sim337>3.0.co;2-y","title":"Testing the equality of two dependent kappa statistics","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":89,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Guelph; Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Statistics; Kappa; Correlation; Monte Carlo method; Mathematics; Cohen's kappa; Goodness of fit; Dependency (UML); Sample size determination; Variable (mathematics); Computer science; Artificial intelligence","score_opus":0.2670104097609483,"score_gpt":0.4556427207423286,"score_spread":0.18863231098138028,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2146976162","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.058271896,0.000765665,0.927771,0.00034207373,0.00032273767,0.0017461998,0.0008243784,0.0009541623,0.009001925],"genre_scores_gemma":[0.48032713,0.00044449553,0.5114686,0.00025269846,0.00014132175,0.0047022756,0.00094646093,0.0004831482,0.0012338781],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6833567,0.20184842,0.024418235,0.026637187,0.06068893,0.0030505066],"domain_scores_gemma":[0.29304883,0.58917,0.031476486,0.04178861,0.041994408,0.0025216897],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.20248723,0.0017242522,0.0034663195,0.010813658,0.0026796486,0.0046273475,0.0045209727,0.0030467499,0.006440107],"category_scores_gemma":[0.64147663,0.001136942,0.0034698679,0.0055543347,0.008107116,0.0061326753,0.008412695,0.0036412317,0.0020405254],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0048524304,0.0004249489,0.08294743,0.0039980374,0.0050793365,0.0012062038,0.011640886,0.023954839,0.009584379,0.21625814,0.012147631,0.6279057],"study_design_scores_gemma":[0.0012101297,0.0053887633,0.15781859,0.0028396081,0.002276931,0.0061951396,0.008677211,0.1980949,0.031480923,0.54214966,0.042150002,0.0017180804],"about_ca_topic_score_codex":0.0014448762,"about_ca_topic_score_gemma":0.0010685647,"teacher_disagreement_score":0.20248723,"about_ca_system_score_codex":0.0020563537,"about_ca_system_score_gemma":0.0042512994,"threshold_uncertainty_score":0.9834752},"labels":[],"label_agreement":null},{"id":"W2147497323","doi":"10.1002/sim.4176","title":"Bayesian inference of gene–environment interaction from incomplete data: What happens when information on environment is disjoint from data on gene and disease?","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic Associations and Epidemiology","field":"Biochemistry, Genetics and Molecular Biology","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta; University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Environmental data; Inference; Mendelian randomization; Population; Disjoint sets; Computer science; Disease; Bayesian probability; Genotype; Statistics; Econometrics; Biology; Genetics; Mathematics; Medicine; Environmental health; Gene; Artificial intelligence; Ecology","score_opus":0.056889990450539225,"score_gpt":0.2997278995458502,"score_spread":0.242837909095311,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2147497323","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07019207,0.005897139,0.9038604,0.01580029,0.00017996573,0.00010114011,0.00067828834,0.00023575802,0.0030549318],"genre_scores_gemma":[0.74723387,0.009041357,0.2310943,0.0061538294,0.0018064647,0.0005413723,0.0015273951,0.00038807275,0.0022132944],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9436587,0.04391938,0.0018953851,0.0057166046,0.0034959717,0.0013139196],"domain_scores_gemma":[0.38302535,0.5876481,0.012085562,0.012906385,0.0027159965,0.0016187328],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.12793338,0.0012981119,0.007442432,0.0031847125,0.0023331705,0.0061765243,0.0056370716,0.0054919203,0.00236306],"category_scores_gemma":[0.4344216,0.0032401786,0.0033643465,0.0046132496,0.014089615,0.0204818,0.005659825,0.008070591,0.00034081822],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000696056,0.00016418318,0.05142697,0.0016628407,0.0034245772,0.0016958201,0.0032478909,0.15485027,0.0010144714,0.683426,0.0044475296,0.09394341],"study_design_scores_gemma":[0.00010621417,0.000051978157,0.0045501837,0.00019078863,0.00021392781,0.00028621216,0.00017494097,0.083828144,0.00028755903,0.9085891,0.0016513474,0.00006962488],"about_ca_topic_score_codex":0.00914379,"about_ca_topic_score_gemma":0.0061675305,"teacher_disagreement_score":0.12793338,"about_ca_system_score_codex":0.0027498154,"about_ca_system_score_gemma":0.0024120044,"threshold_uncertainty_score":0.6765845},"labels":[],"label_agreement":null},{"id":"W2147506537","doi":"10.1002/sim.5571","title":"Scientific considerations for assessing biosimilar products","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Biosimilars and Bioanalytical Methods","field":"Immunology and Microbiology","cited_by":31,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Biosimilar; Interchangeability; Risk analysis (engineering); Bioequivalence; Drug approval; Business; Computer science; Drug; Medicine; Management science; Pharmacology; Engineering; Pharmacokinetics","score_opus":0.09295416366638011,"score_gpt":0.3965204339961872,"score_spread":0.3035662703298071,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2147506537","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014891076,0.030018931,0.4448512,0.44300956,0.009906362,0.0017700236,0.0011447025,0.00059840427,0.053809732],"genre_scores_gemma":[0.24074332,0.014185886,0.6437595,0.07172061,0.021083675,0.0034319737,0.0008002799,0.00045707793,0.00381762],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.41032955,0.3800625,0.06523246,0.010696951,0.13198476,0.0016938614],"domain_scores_gemma":[0.18570961,0.61788523,0.045286022,0.034450643,0.11299653,0.0036718764],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.48267892,0.0024917272,0.0047699744,0.015932044,0.005207162,0.017110696,0.0069804345,0.018051447,0.0038778987],"category_scores_gemma":[0.6729284,0.0015698441,0.0030738835,0.01042568,0.028340744,0.020089917,0.0073735174,0.015651155,0.0018741497],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00046264075,0.0002931686,0.0075689047,0.0018370516,0.00028513523,0.00038578425,0.0013413164,0.004897249,0.0013501022,0.80672413,0.04629325,0.12856129],"study_design_scores_gemma":[0.00014792783,0.00041960934,0.0030814023,0.0019257268,0.0001456115,0.0005190391,0.0009756172,0.0093531795,0.0022712105,0.9012163,0.07978543,0.00015897422],"about_ca_topic_score_codex":0.0021383832,"about_ca_topic_score_gemma":0.0016300939,"teacher_disagreement_score":0.5173211,"about_ca_system_score_codex":0.009429317,"about_ca_system_score_gemma":0.018618664,"threshold_uncertainty_score":0.637949},"labels":[],"label_agreement":null},{"id":"W2147898530","doi":"10.1002/sim.3078","title":"CoPlot: A tool for visualizing multivariate data in medicine","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Hydrology and Drought Analysis","field":"Environmental Science","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"","keywords":"Multivariate statistics; Multidimensional scaling; Computer science; Multivariate analysis; Data mining; Set (abstract data type); Data science; Data visualization; Data set; Visualization; Interpretation (philosophy); Artificial intelligence; Machine learning","score_opus":0.04385822664028715,"score_gpt":0.39284796707206726,"score_spread":0.3489897404317801,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2147898530","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.004934779,0.0016254826,0.9087839,0.0019735154,0.00075732375,0.00055021676,0.018267523,0.057853058,0.005254258],"genre_scores_gemma":[0.045926217,0.0016774473,0.9298602,0.00061611226,0.00045967445,0.0020655505,0.009858515,0.0075551933,0.001981146],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9916849,0.0047557848,0.0010595325,0.00078067125,0.0014603446,0.0002586911],"domain_scores_gemma":[0.9446387,0.043414015,0.003283358,0.00352475,0.0041562733,0.0009828764],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013125612,0.0030799475,0.002813861,0.013614149,0.001558102,0.0064613405,0.0023679459,0.0018813789,0.044413537],"category_scores_gemma":[0.06383004,0.0014016972,0.0028912919,0.012983573,0.0011476551,0.0071948976,0.00519932,0.0042771217,0.007227092],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014801968,0.00031970223,0.008275038,0.0056109927,0.0011870085,0.0012452443,0.004893789,0.020200169,0.005526204,0.067491636,0.39029187,0.49347812],"study_design_scores_gemma":[0.00066676724,0.00057850406,0.0143125765,0.0019750216,0.0005021325,0.0023421897,0.002092543,0.18907134,0.009787436,0.20923087,0.5685023,0.00093829934],"about_ca_topic_score_codex":0.0034521718,"about_ca_topic_score_gemma":0.0032376416,"teacher_disagreement_score":0.044413537,"about_ca_system_score_codex":0.001006894,"about_ca_system_score_gemma":0.0031341028,"threshold_uncertainty_score":0.1485781},"labels":[],"label_agreement":null},{"id":"W2148216078","doi":"10.1002/sim.3619","title":"Modelling heterogeneity in clustered count data with extra zeros using compound Poisson random effect","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Memorial University of Newfoundland; University of New Brunswick","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Count data; Poisson distribution; Random effects model; Overdispersion; Poisson regression; Statistics; Zero-inflated model; Quasi-likelihood; Variance (accounting); Multilevel model; Computer science; Hierarchical database model; Compound Poisson distribution; Mathematics; Econometrics; Data mining; Medicine; Population; Meta-analysis","score_opus":0.15086653236908357,"score_gpt":0.43161931460224834,"score_spread":0.2807527822331648,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2148216078","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06757997,0.00025400892,0.9308188,0.0003326416,0.000034980876,0.00011583848,0.00021413514,0.00011894297,0.0005306268],"genre_scores_gemma":[0.6897313,0.0005644977,0.3057801,0.00025178946,0.0001013552,0.0005806767,0.0006015274,0.00009662808,0.0022920775],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9855921,0.010676825,0.0005553001,0.0016690328,0.00108193,0.00042482174],"domain_scores_gemma":[0.9152073,0.07378791,0.0056885527,0.0036356028,0.0012624975,0.0004180925],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03485744,0.00083725475,0.0022203224,0.0020479544,0.0006104387,0.0018044172,0.0035726829,0.0017778202,0.002129288],"category_scores_gemma":[0.065637454,0.0008360541,0.0021735635,0.0028352311,0.001993457,0.0025606041,0.0021480424,0.0017810109,0.0003140012],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00058372127,0.00018007355,0.032642197,0.00048256177,0.0008643468,0.0018021846,0.001668493,0.47808126,0.001685037,0.4262696,0.0015810417,0.054159414],"study_design_scores_gemma":[0.00010962147,0.00016187149,0.004157715,0.00006423072,0.0001856535,0.00026174998,0.00017124762,0.7809782,0.0006313236,0.21197127,0.001226911,0.000080177306],"about_ca_topic_score_codex":0.004216321,"about_ca_topic_score_gemma":0.0036229333,"teacher_disagreement_score":0.03485744,"about_ca_system_score_codex":0.0013360432,"about_ca_system_score_gemma":0.0015305637,"threshold_uncertainty_score":0.18434596},"labels":[],"label_agreement":null},{"id":"W2148357625","doi":"10.1002/sim.4434","title":"Weighted multiple testing correction for correlated tests","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hamilton Health Sciences; McMaster University; Population Health Research Institute","funders":"","keywords":"Bonferroni correction; Type I and type II errors; Multiple comparisons problem; Sequence (biology); Mathematics; Correlation; Statistics; False discovery rate; Statistical hypothesis testing; Null hypothesis; Statistical power; Algorithm; Computer science","score_opus":0.6250874573824682,"score_gpt":0.5510253799376074,"score_spread":0.07406207744486082,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2148357625","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007366666,0.0014922678,0.97951376,0.0008163415,0.0031063396,0.002149413,0.0011394877,0.0019465543,0.0024691462],"genre_scores_gemma":[0.1301167,0.00070853083,0.84344196,0.0016683777,0.0009210748,0.013801763,0.001080917,0.002246881,0.0060138684],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.77284205,0.15953523,0.014920413,0.026449429,0.022954632,0.0032981914],"domain_scores_gemma":[0.59544104,0.31958374,0.017938163,0.044565737,0.020745164,0.0017260911],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.12897101,0.0038084306,0.006204396,0.0045466963,0.002236364,0.00293503,0.007994893,0.0039091697,0.030330561],"category_scores_gemma":[0.44686925,0.0016773391,0.008376761,0.007562024,0.0049817995,0.004006644,0.0057449983,0.011343946,0.004017001],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0057355063,0.00088267744,0.026788743,0.00831818,0.01532031,0.0035822701,0.0033202856,0.03361041,0.0075440966,0.11732143,0.073692136,0.703884],"study_design_scores_gemma":[0.0031026537,0.0065794704,0.051776133,0.00510653,0.007333037,0.0038693321,0.0013885085,0.3341548,0.023512587,0.3583999,0.20398201,0.00079504965],"about_ca_topic_score_codex":0.003103874,"about_ca_topic_score_gemma":0.002425963,"teacher_disagreement_score":0.12897101,"about_ca_system_score_codex":0.003341989,"about_ca_system_score_gemma":0.006559804,"threshold_uncertainty_score":0.68207204},"labels":[],"label_agreement":null},{"id":"W2148441587","doi":"10.1002/sim.2757","title":"A residuals‐based transition model for longitudinal analysis with estimation in the presence of missing data","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"National Institute on Drug Abuse; National Institute on Alcohol Abuse and Alcoholism; Rural Development Administration","keywords":"Imputation (statistics); Missing data; Covariate; Econometrics; Computer science; Longitudinal data; Statistics; Regression; Autoregressive model; Data mining; Mathematics","score_opus":0.14013846414275608,"score_gpt":0.4426742623897916,"score_spread":0.30253579824703547,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2148441587","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0038431017,0.00015103439,0.99504304,0.00027362307,0.000039156646,0.0000323277,0.0001558632,0.00016003824,0.00030182043],"genre_scores_gemma":[0.32558334,0.0011168489,0.6620296,0.0005085667,0.0003170779,0.0013330528,0.0016398864,0.00035950248,0.007112131],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9914226,0.005823921,0.0003973439,0.0011273542,0.00079245627,0.0004363819],"domain_scores_gemma":[0.9656384,0.027172774,0.0023148127,0.0027127904,0.0016734683,0.00048770654],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020706033,0.0008213784,0.0017195825,0.001680073,0.00052346004,0.0017085474,0.005441597,0.002272735,0.0044694026],"category_scores_gemma":[0.053435497,0.00093038747,0.0022707724,0.0024177318,0.0022305134,0.0037812789,0.0025764094,0.0037140923,0.0012836551],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00029736586,0.00012239105,0.007834814,0.00027129415,0.00033420516,0.0004236926,0.0008343906,0.3039754,0.0009620102,0.60934263,0.004136541,0.071465276],"study_design_scores_gemma":[0.000051672916,0.00011357807,0.0012306493,0.000053051026,0.00008808463,0.00011371429,0.000048011236,0.7929044,0.000279871,0.20082234,0.004236847,0.000057789683],"about_ca_topic_score_codex":0.00549545,"about_ca_topic_score_gemma":0.003565171,"teacher_disagreement_score":0.020706033,"about_ca_system_score_codex":0.0011105689,"about_ca_system_score_gemma":0.0019464829,"threshold_uncertainty_score":0.109505236},"labels":[],"label_agreement":null},{"id":"W2150563978","doi":"10.1002/sim.1018","title":"Cost‐effectiveness analysis when the WTA is greater than the WTP","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Joseph’s Healthcare Hamilton; McMaster University; St. Joseph's Hospital","funders":"","keywords":"Willingness to pay; Interpretability; Willingness to accept; Cost effectiveness; Actuarial science; Cost–benefit analysis; Econometrics; Medicine; Computer science; Economics; Risk analysis (engineering); Microeconomics","score_opus":0.4059551580570371,"score_gpt":0.48038368130171705,"score_spread":0.07442852324467997,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2150563978","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.10367439,0.12934312,0.64058846,0.03575239,0.003262089,0.004584475,0.0048800083,0.00047214588,0.077443],"genre_scores_gemma":[0.84553844,0.02716739,0.10487362,0.0066242525,0.002112576,0.0058445698,0.0015771979,0.00020903816,0.0060529467],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.94080514,0.050483562,0.0022989924,0.0013519952,0.004083213,0.0009770585],"domain_scores_gemma":[0.84135604,0.1504575,0.0035351554,0.0023284645,0.0017935587,0.000529331],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06217551,0.0015790203,0.005888327,0.0035064449,0.00036706726,0.0035512317,0.0014686099,0.003486824,0.00962643],"category_scores_gemma":[0.14236283,0.00069281465,0.004131207,0.003336948,0.0011653306,0.0050107073,0.0015255074,0.004191658,0.0005200412],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009158541,0.00053967466,0.012294998,0.010281816,0.015283547,0.0020154135,0.00034635921,0.21695916,0.0016802505,0.3506688,0.020907827,0.35986364],"study_design_scores_gemma":[0.0013755306,0.0016647079,0.013335434,0.0034220372,0.0073763123,0.0022397721,0.00040099074,0.26977444,0.0010085907,0.6544385,0.04471808,0.000245603],"about_ca_topic_score_codex":0.0012208143,"about_ca_topic_score_gemma":0.0008240568,"teacher_disagreement_score":0.06217551,"about_ca_system_score_codex":0.0022222605,"about_ca_system_score_gemma":0.0024165853,"threshold_uncertainty_score":0.32881945},"labels":[],"label_agreement":null},{"id":"W2151248340","doi":"10.1002/sim.5889","title":"Using confidence intervals to compare several correlated areas under the receiver operating characteristic curves","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"Ontario Ministry of Research and Innovation; Natural Sciences and Engineering Research Council of Canada","keywords":"Receiver operating characteristic; Confidence interval; Statistics; Computer science; Mathematics","score_opus":0.5440103651787387,"score_gpt":0.569237814890378,"score_spread":0.02522744971163926,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2151248340","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009455297,0.0015079293,0.9872569,0.00016449203,0.00010378776,0.00015096221,0.00013972296,0.00044980148,0.0007712156],"genre_scores_gemma":[0.26275793,0.0012261698,0.73278064,0.00020654175,0.0003056805,0.001162616,0.00063930376,0.00035710484,0.00056402007],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.90036666,0.07422528,0.0052798362,0.008050125,0.0113165965,0.0007614835],"domain_scores_gemma":[0.4983309,0.44993454,0.0201045,0.019120535,0.011404974,0.001104614],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.106792346,0.0023275067,0.0027516491,0.011309339,0.00069816323,0.0045324317,0.0033680527,0.0026905714,0.0028448368],"category_scores_gemma":[0.40720505,0.0007798887,0.003002922,0.0061637904,0.002942821,0.004187023,0.0045805513,0.003918855,0.0006531775],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001835842,0.0003762616,0.027498642,0.0022300398,0.004785649,0.00089269545,0.0015256933,0.20460899,0.0049377624,0.15770783,0.004184332,0.58941627],"study_design_scores_gemma":[0.00043540855,0.00196383,0.016510502,0.0010653766,0.0014194702,0.0016064361,0.00043695333,0.75370395,0.009132903,0.2014686,0.011729854,0.00052669545],"about_ca_topic_score_codex":0.0011890722,"about_ca_topic_score_gemma":0.000578067,"teacher_disagreement_score":0.106792346,"about_ca_system_score_codex":0.0012412901,"about_ca_system_score_gemma":0.0018254903,"threshold_uncertainty_score":0.5647787},"labels":[],"label_agreement":null},{"id":"W2151617875","doi":"10.1002/sim.2116","title":"Bayesian modelling of imperfect ascertainment methods in cancer studies","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic Associations and Epidemiology","field":"Biochemistry, Genetics and Molecular Biology","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Jewish General Hospital; Montreal General Hospital","funders":"Canadian Arthritis Network; Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Cancer registry; Cancer; Population; Statistics; Bayesian probability; Incidence (geometry); Gold standard (test); Medicine; Mathematics; Internal medicine; Environmental health","score_opus":0.05725572175604364,"score_gpt":0.44193500957810483,"score_spread":0.3846792878220612,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2151617875","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.018733433,0.0032662498,0.96947014,0.0023453815,0.00016154205,0.0007049527,0.0011871507,0.0002566549,0.003874468],"genre_scores_gemma":[0.4273071,0.007491774,0.5384567,0.0013301921,0.0007036997,0.00673571,0.0032562155,0.0003822229,0.014336382],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9276131,0.056014847,0.0032081578,0.007218637,0.004357464,0.0015878002],"domain_scores_gemma":[0.6527553,0.3131576,0.016016804,0.010588814,0.0060996334,0.0013818373],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.10669846,0.002023521,0.0051516774,0.0048526577,0.0020906813,0.0067250878,0.008624505,0.005989019,0.0072562173],"category_scores_gemma":[0.24684443,0.0039714323,0.004894448,0.007102758,0.005408485,0.0067853066,0.005375099,0.006553058,0.0012001598],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00032543257,0.00010342003,0.011993412,0.00082577235,0.0007396114,0.0006723291,0.0021279894,0.4455711,0.00025849586,0.49500346,0.003180548,0.03919836],"study_design_scores_gemma":[0.00023145825,0.000096291464,0.0027834324,0.00043192742,0.00027624352,0.0002945104,0.00016062977,0.58171827,0.0001713565,0.4064316,0.007284203,0.00012011355],"about_ca_topic_score_codex":0.031232173,"about_ca_topic_score_gemma":0.021808797,"teacher_disagreement_score":0.10669846,"about_ca_system_score_codex":0.006326793,"about_ca_system_score_gemma":0.004018619,"threshold_uncertainty_score":0.5642822},"labels":[],"label_agreement":null},{"id":"W2151722761","doi":"10.1002/sim.3094","title":"Hierarchical Bayesian spatiotemporal analysis of revascularization odds using smoothing splines","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université de Sherbrooke; Simon Fraser University","funders":"Fundação para a Ciência e a Tecnologia","keywords":"Markov chain Monte Carlo; Odds; Bayesian probability; Random effects model; Smoothing; Bayesian inference; Statistics; Computer science; Inference; Econometrics; Markov chain; Mathematics; Artificial intelligence; Logistic regression; Medicine; Internal medicine","score_opus":0.25380343357457186,"score_gpt":0.47357443580603237,"score_spread":0.2197710022314605,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2151722761","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.054166894,0.00037913487,0.9430668,0.0004262832,0.00002111245,0.00006960199,0.0004736785,0.00033895124,0.0010576698],"genre_scores_gemma":[0.775979,0.0006838314,0.21790221,0.000092984796,0.000054442822,0.00026192458,0.00091461826,0.00012449708,0.003986451],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9957023,0.002982306,0.00013651061,0.00045265467,0.00042652042,0.00029977039],"domain_scores_gemma":[0.982312,0.01346366,0.0016254019,0.0012390383,0.001095088,0.00026477335],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012700583,0.00053092133,0.0011045197,0.0029330966,0.0007151507,0.001543722,0.0021528606,0.0010296303,0.003922885],"category_scores_gemma":[0.03787704,0.0006228176,0.0017776976,0.0043252897,0.0013233791,0.0016958001,0.0016068583,0.0014351911,0.00032586613],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015211562,0.00005948141,0.031451564,0.00011015022,0.0004227393,0.0002479244,0.00063958776,0.57731915,0.00059518433,0.30346715,0.0019285537,0.0836064],"study_design_scores_gemma":[0.000017168744,0.000022831558,0.0074147764,0.00002407215,0.000055042347,0.000049440205,0.000078018755,0.9103209,0.00010599243,0.08073614,0.0011485462,0.000027038213],"about_ca_topic_score_codex":0.11983287,"about_ca_topic_score_gemma":0.07566234,"teacher_disagreement_score":0.11983287,"about_ca_system_score_codex":0.0023976455,"about_ca_system_score_gemma":0.0026346794,"threshold_uncertainty_score":0.23827076},"labels":[],"label_agreement":null},{"id":"W2151923367","doi":"10.1002/sim.2868","title":"Regression B‐spline smoothing in Bayesian disease mapping: with an application to patient safety surveillance","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":50,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Child and Family Research Institute; University of British Columbia; Providence Health Care","funders":"","keywords":"Smoothing; Computer science; Prior probability; Bayesian probability; Deviance information criterion; Bayes' theorem; Marginal likelihood; Smoothing spline; Random effects model; Context (archaeology); Statistics; Bayesian inference; Econometrics; Mathematics; Artificial intelligence; Geography","score_opus":0.030658900740926413,"score_gpt":0.37988102876256025,"score_spread":0.3492221280216338,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2151923367","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008611183,0.0015649649,0.9877423,0.0007743814,0.00006206023,0.000045343262,0.00006935676,0.00020698208,0.0009234274],"genre_scores_gemma":[0.24232133,0.0048972997,0.7487208,0.00031611187,0.00030591575,0.0003602187,0.0003538749,0.00020286316,0.0025215955],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99416286,0.0045810076,0.00016965035,0.00042973002,0.000533815,0.00012306466],"domain_scores_gemma":[0.98101693,0.01598328,0.00093939586,0.00076991087,0.001034574,0.00025583484],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.015514388,0.00084805896,0.0014721727,0.0023142959,0.0008391576,0.0017400151,0.0016174581,0.0020985673,0.0016568777],"category_scores_gemma":[0.05140092,0.00070296484,0.0019855506,0.0036504588,0.0014573281,0.0013441036,0.0025471698,0.00232434,0.0003820733],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013560665,0.000100683144,0.007108402,0.00030520637,0.00019688359,0.00030262856,0.00068165234,0.6248103,0.0010868285,0.18101566,0.0030742227,0.18118194],"study_design_scores_gemma":[0.000020466912,0.00003672342,0.0010584934,0.00005459077,0.000025805452,0.000093213035,0.000058195757,0.9242647,0.00024913816,0.0704307,0.0036689905,0.00003899801],"about_ca_topic_score_codex":0.025123307,"about_ca_topic_score_gemma":0.013674807,"teacher_disagreement_score":0.025123307,"about_ca_system_score_codex":0.0011906504,"about_ca_system_score_gemma":0.0022162874,"threshold_uncertainty_score":0.08204889},"labels":[],"label_agreement":null},{"id":"W2153045803","doi":"10.1002/sim.3358","title":"Estimation method of the semiparametric mixture cure gamma frailty model","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Unobservable; Mixture model; Covariate; Expectation–maximization algorithm; Computer science; Imputation (statistics); Statistics; Accelerated failure time model; Econometrics; Maximum likelihood; Mathematics; Artificial intelligence; Missing data; Machine learning","score_opus":0.1371487357945214,"score_gpt":0.44413282114809916,"score_spread":0.3069840853535778,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2153045803","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008064206,0.00014382733,0.9912628,0.00008412271,0.0000115113935,0.000022722741,0.000055648397,0.00012054252,0.00023463866],"genre_scores_gemma":[0.52371967,0.0009820089,0.4700443,0.0001664052,0.00011796295,0.00050147716,0.0010789146,0.0001560112,0.0032332544],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9982287,0.0010829021,0.00007517016,0.00023756643,0.00026742223,0.00010816474],"domain_scores_gemma":[0.9952938,0.0032000619,0.00039677336,0.0005383722,0.0004589327,0.00011203969],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005191609,0.00080903975,0.0017229986,0.0018731456,0.00045162995,0.0009823032,0.0019970206,0.0012218102,0.0027798621],"category_scores_gemma":[0.016294474,0.0006128153,0.0020606597,0.0014792939,0.0007266319,0.001638509,0.0020704453,0.0021830301,0.0005601651],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00024483452,0.00011322599,0.010222349,0.0002735375,0.0004807089,0.00032670316,0.0005313696,0.6505422,0.0036711365,0.15264718,0.0033685416,0.17757817],"study_design_scores_gemma":[0.000019322448,0.000031057687,0.0010948123,0.000020767991,0.00004488047,0.000115601484,0.00002832785,0.96966344,0.00050109555,0.027373383,0.0010756403,0.000031706448],"about_ca_topic_score_codex":0.0029627103,"about_ca_topic_score_gemma":0.0017746735,"teacher_disagreement_score":0.005191609,"about_ca_system_score_codex":0.0006717652,"about_ca_system_score_gemma":0.0012456083,"threshold_uncertainty_score":0.027456164},"labels":[],"label_agreement":null},{"id":"W2153200786","doi":"10.1002/sim.6044","title":"An empirical comparison of univariate and multivariate meta‐analyses for categorical outcomes","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":60,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Sudbury","funders":"Agency for Healthcare Research and Quality; U.S. Department of Health and Human Services","keywords":"Univariate; Multivariate statistics; Categorical variable; Statistics; Multivariate analysis; Meta-analysis; Context (archaeology); Econometrics; Bayesian probability; Prior probability; Correlation; Mathematics; Computer science; Medicine; Internal medicine","score_opus":0.8756302530372296,"score_gpt":0.6790142308790336,"score_spread":0.19661602215819607,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2153200786","genre_codex":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.033494215,0.59875196,0.34337857,0.0045983614,0.0025609971,0.004728872,0.0067746416,0.0011285988,0.004583825],"genre_scores_gemma":[0.68916786,0.08318091,0.20170593,0.0038290257,0.0017284803,0.01287042,0.0051382403,0.00081777305,0.0015612451],"study_design_codex":"meta_analysis","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.65345836,0.27087957,0.036469404,0.018350309,0.01968421,0.001158109],"domain_scores_gemma":[0.32738072,0.6051559,0.032444052,0.027322358,0.00710364,0.00059338677],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3205521,0.0054213866,0.013947571,0.015501041,0.00071938045,0.007899261,0.005413967,0.0038299304,0.005186634],"category_scores_gemma":[0.5702699,0.002575983,0.058768973,0.014381532,0.0030925204,0.0083776405,0.004273393,0.005462421,0.00084624114],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0036168487,0.00006213024,0.01439446,0.057290226,0.8746851,0.00032879654,0.00024032546,0.007124341,0.0005192748,0.004320503,0.001708405,0.035709646],"study_design_scores_gemma":[0.0022663139,0.0011821383,0.014642901,0.015813291,0.91216046,0.0006371471,0.00032228284,0.011242053,0.0014435597,0.0300307,0.010016053,0.00024303728],"about_ca_topic_score_codex":0.0013384186,"about_ca_topic_score_gemma":0.0014994072,"teacher_disagreement_score":0.98605245,"about_ca_system_score_codex":0.002612957,"about_ca_system_score_gemma":0.0017279809,"threshold_uncertainty_score":0.83788025},"labels":[],"label_agreement":null},{"id":"W2153784930","doi":"10.1002/sim.1101","title":"Estimating cancer prevalence using mixture models for cancer survival","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"BC Cancer Agency","funders":"","keywords":"Hazard; Hazard ratio; Cancer; Disease; Medicine; Population; Cancer survival; Proportional hazards model; Demography; Survival analysis; Environmental health; Internal medicine; Confidence interval; Biology","score_opus":0.10269196972644017,"score_gpt":0.39283956274252446,"score_spread":0.2901475930160843,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2153784930","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.022356158,0.00094526925,0.9741726,0.00043482406,0.000056850397,0.00014671695,0.0006754968,0.00055832404,0.0006537151],"genre_scores_gemma":[0.49606165,0.0037308459,0.48541772,0.0003315611,0.00042552425,0.0018557587,0.0068576597,0.00030573434,0.0050135097],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9957333,0.002472986,0.00020307956,0.00073077,0.00059927045,0.00026049884],"domain_scores_gemma":[0.9824302,0.014318688,0.0011214521,0.0011919969,0.00073723303,0.00020034691],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009605247,0.0015075089,0.00257972,0.0059490725,0.0009393772,0.0025002358,0.0026902528,0.0028553242,0.0028777174],"category_scores_gemma":[0.03326819,0.0016518916,0.0038833804,0.0042918823,0.0011946372,0.0029578328,0.0022624463,0.002838613,0.0013234144],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004905501,0.00019411971,0.029513018,0.0002786745,0.00075341994,0.00018466396,0.0005100155,0.76133233,0.00092362723,0.06691938,0.0036431842,0.135257],"study_design_scores_gemma":[0.00004162168,0.000044634256,0.0030514954,0.000052855532,0.000092535236,0.00013372736,0.000034072567,0.9345431,0.00021717152,0.059948407,0.0017861307,0.000054200224],"about_ca_topic_score_codex":0.016686589,"about_ca_topic_score_gemma":0.013941859,"teacher_disagreement_score":0.016686589,"about_ca_system_score_codex":0.001830853,"about_ca_system_score_gemma":0.0014278032,"threshold_uncertainty_score":0.050798},"labels":[],"label_agreement":null},{"id":"W2154400783","doi":"10.1002/sim.821","title":"The use of mixture models for identifying high risks in disease mapping","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":36,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Simon Fraser University","funders":"","keywords":"Overdispersion; Poisson distribution; Frequentist inference; Econometrics; Statistics; Autocorrelation; Parametric model; Computer science; Parametric statistics; Bayes' theorem; Random effects model; Mixture model; Bayesian probability; Population; Inference; Bayesian inference; Mathematics; Negative binomial distribution; Artificial intelligence; Medicine","score_opus":0.16375861163825192,"score_gpt":0.38095295121453415,"score_spread":0.21719433957628223,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2154400783","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005622524,0.000970832,0.99252594,0.00019781245,0.000022919732,0.000039643048,0.000046125842,0.00014887872,0.00042530985],"genre_scores_gemma":[0.23022087,0.0025346458,0.7649816,0.00019243191,0.0001618368,0.00042376964,0.00037091938,0.00010952316,0.0010044167],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9852297,0.011663743,0.00036270518,0.0012039804,0.0013140453,0.00022592275],"domain_scores_gemma":[0.9401598,0.053608853,0.0022873958,0.002556634,0.001042328,0.00034497367],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02041375,0.0015818267,0.002214264,0.007174667,0.001036608,0.0027537686,0.0026648906,0.0031234154,0.0016794477],"category_scores_gemma":[0.09215462,0.001325273,0.0022061781,0.004820098,0.0026017632,0.0052712136,0.0033945714,0.002850975,0.0006291404],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00034636597,0.00014102375,0.01985694,0.00054613146,0.00095972995,0.0003277584,0.0010858583,0.49717,0.0012170295,0.21381383,0.0020554531,0.26248],"study_design_scores_gemma":[0.00006166176,0.000096860764,0.003020699,0.000115179995,0.00016546724,0.0003211737,0.00011135387,0.7064233,0.00059862516,0.28634393,0.0026240258,0.000117749485],"about_ca_topic_score_codex":0.0060680993,"about_ca_topic_score_gemma":0.0049379664,"teacher_disagreement_score":0.02041375,"about_ca_system_score_codex":0.001306964,"about_ca_system_score_gemma":0.0011911114,"threshold_uncertainty_score":0.10795957},"labels":[],"label_agreement":null},{"id":"W2154910915","doi":"10.1002/sim.2525","title":"Models for assisted conception data with embryo‐specific covariates","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Assisted Reproductive Technology and Twin Pregnancy","field":"Medicine","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"York University","keywords":"Covariate; Outcome (game theory); Random effects model; Embryo; Statistics; Computer science; Data set; Econometrics; Medicine; Mathematics; Biology; Internal medicine; Genetics","score_opus":0.05828472769053113,"score_gpt":0.33604692622526994,"score_spread":0.2777621985347388,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2154910915","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.11796262,0.0029169477,0.8571197,0.0050764303,0.0003344257,0.0008674066,0.011014904,0.0013369517,0.0033706233],"genre_scores_gemma":[0.72030425,0.0044715446,0.2254775,0.0011978114,0.0006733443,0.004257701,0.016057972,0.00041030793,0.027149512],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9881612,0.0076674805,0.0005330302,0.002136556,0.00076579954,0.00073596684],"domain_scores_gemma":[0.93391496,0.054014515,0.0058243866,0.0031893803,0.0021567359,0.0009000642],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.023833135,0.0019550696,0.004581328,0.002969403,0.0014420829,0.00443808,0.0074558705,0.0048093395,0.0095475055],"category_scores_gemma":[0.045985494,0.0019066092,0.0041490565,0.005223898,0.0025387954,0.0044856365,0.0033257122,0.005706454,0.0022759875],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00080432126,0.00019713484,0.023160463,0.0005165147,0.001278996,0.0011128323,0.001111426,0.6202296,0.0005665451,0.3202316,0.006779258,0.0240114],"study_design_scores_gemma":[0.00025103363,0.0001756446,0.005337569,0.00008642137,0.00019777918,0.00025982887,0.00013168353,0.8493093,0.00013729297,0.13787894,0.0061410996,0.00009337279],"about_ca_topic_score_codex":0.024032095,"about_ca_topic_score_gemma":0.02026211,"teacher_disagreement_score":0.024032095,"about_ca_system_score_codex":0.0035045375,"about_ca_system_score_gemma":0.0026843357,"threshold_uncertainty_score":0.1260432},"labels":[],"label_agreement":null},{"id":"W2156667592","doi":"10.1002/sim.5754","title":"Armitage Lecture 2011: the design and analysis of life history studies","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Data Analysis and Archiving","field":"Social Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Computer science; Statistics; Mathematics","score_opus":0.06506782028851961,"score_gpt":0.3518201170214288,"score_spread":0.2867522967329092,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2156667592","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0060265795,0.002787861,0.9112679,0.029312735,0.005590812,0.013582427,0.0046788063,0.0029760287,0.023777008],"genre_scores_gemma":[0.022900514,0.0030142518,0.8989829,0.002158326,0.0013360173,0.05810904,0.0022896333,0.0015195754,0.009689759],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8300191,0.14840406,0.0071494537,0.0041682557,0.008710865,0.0015482273],"domain_scores_gemma":[0.6677508,0.23340037,0.010168154,0.055114157,0.028541297,0.0050252243],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.19320178,0.002009137,0.0028576574,0.0045858924,0.002652749,0.010004711,0.0027681724,0.0032834026,0.03346403],"category_scores_gemma":[0.42575014,0.003668643,0.0021677765,0.0064349123,0.0047983583,0.004875392,0.008088353,0.00994105,0.010949926],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018439031,0.0002788416,0.00511854,0.0020295784,0.00064780324,0.00036309127,0.0053561428,0.008628451,0.0019827788,0.1263334,0.28074235,0.5666751],"study_design_scores_gemma":[0.001108753,0.0008434823,0.011052812,0.0040776436,0.00041213658,0.00033141382,0.0015196567,0.031041497,0.0028876765,0.26094347,0.68545544,0.0003260118],"about_ca_topic_score_codex":0.0026260903,"about_ca_topic_score_gemma":0.003194078,"teacher_disagreement_score":0.19320178,"about_ca_system_score_codex":0.0043860707,"about_ca_system_score_gemma":0.016234003,"threshold_uncertainty_score":0.99492586},"labels":[],"label_agreement":null},{"id":"W2157347940","doi":"10.1002/sim.698","title":"A comparison of methods to detect publication bias in meta‐analysis","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":1297,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Statistics; Publication bias; Funnel plot; Sample size determination; Meta-analysis; Type I and type II errors; Nominal level; Linear regression; Mathematics; Econometrics; Censoring (clinical trials); Logistic regression; Meta-regression; Statistical power; Logit; Confidence interval; Medicine","score_opus":0.8953613622891847,"score_gpt":0.6824856965509237,"score_spread":0.212875665738261,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2157347940","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.022130987,0.3918507,0.5524138,0.0042067403,0.0049168672,0.014744504,0.003970131,0.0019720243,0.0037941756],"genre_scores_gemma":[0.16968617,0.1060431,0.663268,0.002111973,0.001494926,0.0521397,0.0026046329,0.0011343665,0.0015171121],"study_design_codex":"meta_analysis","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.31056976,0.6002141,0.044009075,0.008195844,0.03623142,0.0007797424],"domain_scores_gemma":[0.19892383,0.75021607,0.01937294,0.0164789,0.014118878,0.00088941614],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4238874,0.0055337106,0.019934326,0.025121368,0.0014320129,0.007888622,0.006231738,0.0072948453,0.0059706066],"category_scores_gemma":[0.7209188,0.0036473954,0.03434489,0.018227646,0.0033306188,0.008440174,0.0053061415,0.006502141,0.0010496785],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.022192085,0.00032188595,0.012069029,0.10628083,0.48838094,0.0003563853,0.0014985091,0.011954365,0.000909183,0.015068805,0.008627956,0.33234],"study_design_scores_gemma":[0.0713966,0.018015632,0.05866822,0.09118789,0.4542598,0.0028650349,0.0016735524,0.085214086,0.0058610504,0.15785609,0.049460705,0.0035413094],"about_ca_topic_score_codex":0.0015383916,"about_ca_topic_score_gemma":0.002095102,"teacher_disagreement_score":0.5761126,"about_ca_system_score_codex":0.004764911,"about_ca_system_score_gemma":0.005311054,"threshold_uncertainty_score":0.71044946},"labels":[],"label_agreement":null},{"id":"W2157860153","doi":"10.1002/sim.3523","title":"Confidence interval construction for a difference between two dependent intraclass correlation coefficients","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":36,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"Ontario Ministry of Research and Innovation; Natural Sciences and Engineering Research Council of Canada","keywords":"Confidence interval; Intraclass correlation; Statistics; Point estimation; Interval estimation; Inference; Mathematics; Sample size determination; Confidence distribution; Nominal level; Variance (accounting); Correlation; Coverage probability; CDF-based nonparametric confidence interval; Standard error; Computer science; Reproducibility; Artificial intelligence","score_opus":0.06250443287661349,"score_gpt":0.41815188558518207,"score_spread":0.35564745270856857,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2157860153","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0017815572,0.00021094512,0.99685836,0.00012437637,0.000052535994,0.000071586655,0.00009180895,0.00020522648,0.0006035748],"genre_scores_gemma":[0.07435401,0.00044879332,0.9222498,0.00017390665,0.00016944754,0.0012404907,0.0006264285,0.0002302011,0.0005069672],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9565833,0.023786318,0.0033382445,0.005943328,0.0094810175,0.0008678584],"domain_scores_gemma":[0.66959554,0.27996078,0.013062696,0.01678945,0.01935102,0.0012404706],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07093043,0.0016792174,0.0021772915,0.009154476,0.0014450282,0.0037846898,0.005302833,0.0039354633,0.0069103865],"category_scores_gemma":[0.35109308,0.0009980729,0.0026069356,0.005185453,0.003920084,0.0048035686,0.0049988087,0.006986889,0.0015741879],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00079917663,0.00022501692,0.008873642,0.0011804102,0.000612643,0.00080013025,0.0016563928,0.060002625,0.004139573,0.55874836,0.008797389,0.35416454],"study_design_scores_gemma":[0.00033880657,0.0007007974,0.006879782,0.00114794,0.00040007228,0.0018638325,0.00043853858,0.36700562,0.011137004,0.5873063,0.02242679,0.00035448108],"about_ca_topic_score_codex":0.0014899189,"about_ca_topic_score_gemma":0.0006358287,"teacher_disagreement_score":0.07093043,"about_ca_system_score_codex":0.001338727,"about_ca_system_score_gemma":0.0021513342,"threshold_uncertainty_score":0.37512046},"labels":[],"label_agreement":null},{"id":"W2159452805","doi":"10.1002/sim.6791","title":"A tutorial on Bayesian bivariate meta‐analysis of mixed binary‐continuous outcomes with missing treatment effects","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital; Public Health Ontario; University Health Network; University of Toronto; Toronto Public Health; Osteoporosis Canada","funders":"Canadian Institutes of Health Research","keywords":"Bivariate analysis; Random effects model; Computer science; Meta-analysis; Categorical variable; Bayesian probability; Missing data; Univariate; Statistics; Econometrics; Data mining; Multivariate statistics; Medicine; Mathematics; Machine learning; Artificial intelligence","score_opus":0.5665462388726342,"score_gpt":0.5224215384526055,"score_spread":0.044124700420028695,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2159452805","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00062568043,0.064567775,0.8800992,0.0076168547,0.0038934418,0.0012241283,0.014294842,0.009978446,0.017699663],"genre_scores_gemma":[0.006559625,0.058015108,0.8763923,0.009585903,0.0035612688,0.00490561,0.009969094,0.004699302,0.026311781],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9973552,0.0014609938,0.0002602452,0.00026667962,0.0005971759,0.00005965267],"domain_scores_gemma":[0.983073,0.014751058,0.0006196876,0.0004721441,0.0008303653,0.00025371232],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.005935026,0.0029874854,0.0018138965,0.0029065346,0.00041525858,0.002099814,0.0022011916,0.002323188,0.094939396],"category_scores_gemma":[0.032120034,0.0014946872,0.003991772,0.0028462883,0.0005784213,0.003021445,0.0016092369,0.0039327745,0.027947908],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017619198,0.00009259182,0.00043192098,0.013845337,0.00072540814,0.0005208436,0.00028289072,0.014986605,0.0019732793,0.08127942,0.5279287,0.35775673],"study_design_scores_gemma":[0.00016054924,0.00013287588,0.00049103476,0.0037578112,0.00024849665,0.0007425464,0.00003981133,0.015438718,0.0010596357,0.15972896,0.8180574,0.00014224106],"about_ca_topic_score_codex":0.0022071567,"about_ca_topic_score_gemma":0.0037111084,"teacher_disagreement_score":0.994065,"about_ca_system_score_codex":0.0011659925,"about_ca_system_score_gemma":0.0023803092,"threshold_uncertainty_score":0.31760395},"labels":[],"label_agreement":null},{"id":"W2159741108","doi":"10.1002/sim.1335","title":"Power and sample size for DNA microarray studies","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Gene expression and cancer classification","field":"Biochemistry, Genetics and Molecular Biology","cited_by":171,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"National Heart, Lung, and Blood Institute","keywords":"Sample size determination; Statistics; Computer science; Sample (material); Power (physics); Computational biology; Mathematics; Biology; Chromatography; Chemistry; Physics","score_opus":0.03013445253274159,"score_gpt":0.33393307553954465,"score_spread":0.30379862300680305,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2159741108","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008290155,0.003994542,0.9712809,0.0056966455,0.0012982783,0.0021777472,0.0008089059,0.00042614175,0.0060267597],"genre_scores_gemma":[0.2887715,0.0029812579,0.66810113,0.005785738,0.0024970518,0.027305288,0.0010980868,0.0008980046,0.0025619236],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7735267,0.1806717,0.009004603,0.013473204,0.022198334,0.0011254202],"domain_scores_gemma":[0.35174036,0.60710794,0.010207049,0.02326674,0.0066239573,0.0010539927],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.20505919,0.0016824988,0.0049318434,0.0038972634,0.0019144799,0.004523454,0.004492934,0.0058241766,0.007486513],"category_scores_gemma":[0.65972406,0.0011101977,0.0033497629,0.0049035787,0.0076024146,0.007041269,0.004763548,0.0064338334,0.0016977562],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00271926,0.00024247226,0.017447222,0.005484432,0.0016214107,0.00095456734,0.0019115914,0.022618173,0.0038000315,0.63018644,0.023450064,0.2895643],"study_design_scores_gemma":[0.0015718284,0.001979613,0.011107355,0.0013208532,0.0007294107,0.0015877055,0.00047077294,0.06323206,0.00608895,0.8570251,0.05467336,0.00021287522],"about_ca_topic_score_codex":0.00052580633,"about_ca_topic_score_gemma":0.00030468823,"teacher_disagreement_score":0.20505919,"about_ca_system_score_codex":0.0022091023,"about_ca_system_score_gemma":0.0028265924,"threshold_uncertainty_score":0.9803035},"labels":[],"label_agreement":null},{"id":"W2159848039","doi":"10.1002/sim.5808","title":"State‐space size considerations for disease‐progression models","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Insurance, Mortality, Demography, Risk Management","field":"Social Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Markov chain; State space; Stochastic matrix; Computer science; Set (abstract data type); Disease; Markov model; Basis (linear algebra); State (computer science); Space (punctuation); Econometrics; Markov process; Statistics; Mathematics; Algorithm; Medicine","score_opus":0.03652017716464257,"score_gpt":0.3702021972862839,"score_spread":0.33368202012164133,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2159848039","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.018264793,0.0013500169,0.9642807,0.007641244,0.00017558147,0.00021825623,0.00043351864,0.0002762108,0.0073596565],"genre_scores_gemma":[0.69590825,0.0027042269,0.2859457,0.0020665142,0.0009372209,0.0018864634,0.0013972131,0.0003228091,0.008831669],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9903606,0.006718574,0.00048264308,0.0009119348,0.0012000167,0.0003262937],"domain_scores_gemma":[0.7732154,0.21181624,0.0031396658,0.0072734803,0.0034812212,0.0010739444],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.030587537,0.0010794994,0.0025748904,0.0015168114,0.0016457598,0.0032643527,0.0044964477,0.0026739365,0.0061375885],"category_scores_gemma":[0.13820758,0.001242008,0.0017575409,0.0015002397,0.0032535659,0.00797672,0.004188562,0.0062914127,0.0005620176],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00014928257,0.00008343797,0.0036476247,0.00013153552,0.000100774385,0.0001841064,0.00042561215,0.3633917,0.0004174216,0.6048682,0.0034711703,0.023129199],"study_design_scores_gemma":[0.00003285481,0.00003547834,0.00038858567,0.00004310828,0.000026215905,0.000047917987,0.000056867455,0.7158742,0.0001686168,0.28097415,0.0023330457,0.000019004372],"about_ca_topic_score_codex":0.01275813,"about_ca_topic_score_gemma":0.01042818,"teacher_disagreement_score":0.030587537,"about_ca_system_score_codex":0.0024634474,"about_ca_system_score_gemma":0.0025889603,"threshold_uncertainty_score":0.16176432},"labels":[],"label_agreement":null},{"id":"W2159915215","doi":"10.1002/sim.3012","title":"Effect of exposure misclassification on the mean squared error of population attributable risk and prevented fraction estimates","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Statistics; Mean squared error; Attributable risk; Fraction (chemistry); Observational error; Population; Variance (accounting); Econometrics; Accuracy and precision; Mathematics; Medicine; Chemistry; Environmental health; Economics","score_opus":0.07347620403333992,"score_gpt":0.4348047438118476,"score_spread":0.36132853977850765,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2159915215","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2244652,0.01577966,0.7406015,0.007575578,0.00068617245,0.00027916438,0.0010422779,0.000621434,0.008948923],"genre_scores_gemma":[0.9057229,0.0020238357,0.08954417,0.00067776715,0.00023778473,0.00014571378,0.00033590136,0.00015516774,0.0011568146],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8846031,0.091013335,0.0051760683,0.0073211496,0.010688609,0.0011977892],"domain_scores_gemma":[0.25944754,0.6899979,0.015871145,0.027635342,0.006584912,0.000463192],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.12997034,0.001026947,0.001969118,0.0026786025,0.0006352963,0.002226407,0.0020532412,0.0031337317,0.0017135114],"category_scores_gemma":[0.51344556,0.0008303417,0.0021645771,0.0026715943,0.0034605693,0.003923211,0.003403392,0.002776438,0.00031840062],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002287875,0.00012894622,0.21891937,0.0011891092,0.005486035,0.0009881969,0.0019303807,0.43983984,0.0015921321,0.13395397,0.002929836,0.19075434],"study_design_scores_gemma":[0.00039995398,0.0011896064,0.1373666,0.0014142541,0.0024073476,0.0033225785,0.0007102539,0.50068283,0.00832382,0.33650368,0.007228241,0.00045090463],"about_ca_topic_score_codex":0.0034899602,"about_ca_topic_score_gemma":0.002539689,"teacher_disagreement_score":0.8700297,"about_ca_system_score_codex":0.0019476829,"about_ca_system_score_gemma":0.0014414501,"threshold_uncertainty_score":0.68735707},"labels":[],"label_agreement":null},{"id":"W2159968952","doi":"10.1002/sim.3243","title":"Discussion of ‘A critical appraisal of propensity‐score matching in the medical literature between 1996 and 2003’","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences","funders":"Institute for Clinical Evaluative Sciences","keywords":"Critical appraisal; Matching (statistics); Library science; Citation; Propensity score matching; Psychology; Operations research; Computer science; Medicine; Alternative medicine; Mathematics; Pathology; Surgery","score_opus":0.12894851090037607,"score_gpt":0.4533669741319672,"score_spread":0.3244184632315911,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2159968952","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0020378572,0.08564556,0.0099525405,0.8730117,0.025910143,0.00017358348,0.0007389204,0.000060047623,0.0024696845],"genre_scores_gemma":[0.05879097,0.045340616,0.026232721,0.80876315,0.05728844,0.00078086747,0.00063544494,0.00015203057,0.0020156582],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.73682934,0.15257514,0.04086598,0.013494276,0.05142085,0.00481444],"domain_scores_gemma":[0.20151609,0.68991446,0.04865548,0.011385188,0.045970522,0.0025582789],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.369296,0.0010606949,0.0017864548,0.018984709,0.0039290274,0.01251598,0.010692033,0.023463609,0.005054329],"category_scores_gemma":[0.7181724,0.0014703843,0.0058056223,0.017484326,0.016536534,0.013353652,0.0054349536,0.015876824,0.0007538879],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00046086506,0.000052720254,0.007799437,0.007364628,0.0016272445,0.00089766126,0.0064607086,0.0013928731,0.00037386522,0.1657014,0.6944119,0.113456786],"study_design_scores_gemma":[0.00034325707,0.00019036312,0.021624336,0.025122399,0.0016027928,0.0016075673,0.0037751344,0.001547304,0.001506243,0.09346278,0.84889853,0.00031935144],"about_ca_topic_score_codex":0.030321775,"about_ca_topic_score_gemma":0.047267072,"teacher_disagreement_score":0.630704,"about_ca_system_score_codex":0.013252633,"about_ca_system_score_gemma":0.03097724,"threshold_uncertainty_score":0.7777703},"labels":[],"label_agreement":null},{"id":"W2160269073","doi":"10.1002/1097-0258(20000730)19:14<1901::aid-sim514>3.0.co;2-v","title":"The generalization of the odds ratio, risk ratio and risk difference tor �k tables","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":28,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Royal Victoria Hospital","funders":"","keywords":"Odds ratio; Confidence interval; Statistics; Pooling; Generalization; Statistic; Mathematics; Relative risk; Standard error; Diagnostic odds ratio; Odds; Absolute risk reduction; Type I and type II errors; Econometrics; Computer science; Logistic regression; Artificial intelligence","score_opus":0.21131445457799014,"score_gpt":0.4951114708605551,"score_spread":0.283797016282565,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2160269073","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0034811988,0.020440517,0.95794725,0.0049278736,0.0034664616,0.0006231724,0.002591535,0.0005630473,0.005958924],"genre_scores_gemma":[0.1411513,0.01993032,0.8160657,0.0047875643,0.0060166335,0.0045231907,0.0019995554,0.00047530528,0.0050504752],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9057269,0.0672646,0.008620823,0.0075742947,0.010218557,0.0005947661],"domain_scores_gemma":[0.77232474,0.19470006,0.012002977,0.015037825,0.0051367213,0.00079766044],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05550654,0.0016771826,0.0031293004,0.0064471923,0.0008926428,0.005091616,0.0035742857,0.0028427842,0.012015673],"category_scores_gemma":[0.27267313,0.0008928363,0.0032982202,0.008295364,0.005102948,0.007945155,0.0033485626,0.005876348,0.002650686],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005709894,0.00006913319,0.0073002595,0.0056547984,0.002141098,0.0006755069,0.0009716646,0.011363866,0.000662758,0.61336243,0.03816536,0.31906208],"study_design_scores_gemma":[0.00012719065,0.00034953438,0.002186833,0.0010397367,0.00046309052,0.0016340624,0.00024416213,0.017021094,0.00057624903,0.8920794,0.0841287,0.00014987634],"about_ca_topic_score_codex":0.0011350147,"about_ca_topic_score_gemma":0.0007114937,"teacher_disagreement_score":0.05550654,"about_ca_system_score_codex":0.002613122,"about_ca_system_score_gemma":0.0029586398,"threshold_uncertainty_score":0.29355013},"labels":[],"label_agreement":null},{"id":"W2160329317","doi":"10.1002/sim.1617","title":"Two‐sample scale tests for comparison of metabolic rates for styrene in previously exposed and unexposed groups","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"National Science Council","keywords":"Statistics; Styrene; Sample size determination; Estimator; Inference; Statistical inference; Scale (ratio); Sample (material); Mathematics; Econometrics; Chemistry; Chromatography; Computer science; Organic chemistry; Physics","score_opus":0.13106755298798983,"score_gpt":0.4701442731845786,"score_spread":0.33907672019658874,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2160329317","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.15104869,0.00031739188,0.8425857,0.00043568472,0.0005886434,0.0012830716,0.00059397577,0.0007093401,0.0024375406],"genre_scores_gemma":[0.6670263,0.00012538029,0.32617137,0.00028363662,0.0002221835,0.0044275173,0.00053781155,0.00019068575,0.0010151172],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.93120164,0.049714606,0.0033221748,0.0065003536,0.008272462,0.0009888624],"domain_scores_gemma":[0.7946714,0.17604282,0.008163594,0.016681802,0.0033469975,0.0010933948],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04456157,0.0010370881,0.0024164382,0.002371533,0.001349185,0.0013256806,0.003009077,0.0022465417,0.009124271],"category_scores_gemma":[0.17754129,0.0004732984,0.0026242589,0.0019255219,0.004747175,0.002969311,0.0020864916,0.0040084654,0.00067865726],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.015346087,0.0037479487,0.08929732,0.0036846397,0.008416989,0.0034503872,0.004987341,0.025237389,0.041887823,0.21511759,0.011713871,0.57711256],"study_design_scores_gemma":[0.0026839306,0.040462866,0.20763029,0.00056538515,0.0023612382,0.0037194446,0.0045039654,0.35568297,0.054478392,0.30318916,0.024073133,0.0006492263],"about_ca_topic_score_codex":0.00036649854,"about_ca_topic_score_gemma":0.00027062878,"teacher_disagreement_score":0.04456157,"about_ca_system_score_codex":0.0008533778,"about_ca_system_score_gemma":0.001349,"threshold_uncertainty_score":0.23566687},"labels":[],"label_agreement":null},{"id":"W2160393348","doi":"10.1002/sim.2206","title":"SARS incubation and quarantine times: when is an exposed individual known to be disease free?","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"COVID-19 epidemiological studies","field":"Mathematics","cited_by":36,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Medical Research Council","keywords":"Biostatistics; Queen (butterfly); Library science; Public health; History; Medicine; Computer science; Pathology","score_opus":0.2196670982091016,"score_gpt":0.4390063802517846,"score_spread":0.21933928204268302,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2160393348","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8752151,0.015874749,0.08451482,0.010846658,0.0002218241,0.00010275346,0.0015010163,0.000092505674,0.011630585],"genre_scores_gemma":[0.9931613,0.0016703603,0.0034344604,0.00012335854,0.00017161021,0.000034527588,0.00029362983,0.000011847226,0.0010989482],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.996185,0.0021266316,0.00027719658,0.00064314355,0.0003308759,0.00043715723],"domain_scores_gemma":[0.87131655,0.10639488,0.014650585,0.0036700882,0.0017037068,0.00226421],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011893747,0.0004943105,0.0013633079,0.0016485372,0.0006538006,0.0021119346,0.0014464946,0.0022344443,0.0046462147],"category_scores_gemma":[0.12217261,0.0004950341,0.00081819674,0.0016275297,0.0023359414,0.0062273503,0.0015236274,0.0022300251,0.00040212108],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0032394868,0.00030945617,0.64672613,0.000911685,0.0003580897,0.0031887616,0.0044853427,0.034013204,0.0019586293,0.15250807,0.0039654174,0.14833583],"study_design_scores_gemma":[0.00011283248,0.0017629783,0.53097194,0.00082892115,0.0005571164,0.0056784074,0.008223312,0.1720465,0.0021815079,0.26897302,0.008449236,0.00021420767],"about_ca_topic_score_codex":0.0051748697,"about_ca_topic_score_gemma":0.004288104,"teacher_disagreement_score":0.011893747,"about_ca_system_score_codex":0.0013199011,"about_ca_system_score_gemma":0.0010027325,"threshold_uncertainty_score":0.0629009},"labels":[],"label_agreement":null},{"id":"W2160577585","doi":"10.1002/sim.6466","title":"Frailty models for pneumonia to death with a left‐censored covariate","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University","funders":"National Institute of General Medical Sciences; Natural Sciences and Engineering Research Council of Canada","keywords":"Covariate; Estimator; Statistics; Inference; Mathematics; Nonparametric statistics; Multiplicative function; Hazard ratio; Proportional hazards model; Confidence interval; Computer science; Econometrics; Artificial intelligence","score_opus":0.5341691921754336,"score_gpt":0.4810747424250354,"score_spread":0.053094449750398176,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2160577585","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.17373343,0.0018459891,0.8181731,0.0016533926,0.000209478,0.00025942037,0.0014628366,0.00039157228,0.002270769],"genre_scores_gemma":[0.8959586,0.0019561935,0.081596784,0.0004903862,0.00023102961,0.0008945851,0.002207131,0.00010606459,0.016559275],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99759656,0.0012405566,0.00011249507,0.00041462865,0.00027035366,0.00036546143],"domain_scores_gemma":[0.99126244,0.0061493777,0.0010971504,0.0005761052,0.00058744376,0.00032743785],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009909364,0.0012138064,0.002034027,0.0017962484,0.0009608947,0.0012257607,0.003760205,0.0019575162,0.004416489],"category_scores_gemma":[0.022037243,0.0005344893,0.003075193,0.0016672894,0.0013345636,0.0018769599,0.0024465676,0.0035212894,0.00054507784],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006961921,0.00030878192,0.054102067,0.00048458448,0.00076185574,0.0020643093,0.0012685226,0.5899551,0.0016126243,0.2290499,0.00659313,0.11310297],"study_design_scores_gemma":[0.000108763,0.00021376662,0.00903451,0.00009625525,0.00019663056,0.0002850138,0.00017352767,0.8559693,0.0003451173,0.13050151,0.00299608,0.000079436075],"about_ca_topic_score_codex":0.016561788,"about_ca_topic_score_gemma":0.016464382,"teacher_disagreement_score":0.016561788,"about_ca_system_score_codex":0.001577398,"about_ca_system_score_gemma":0.0023692609,"threshold_uncertainty_score":0.05240631},"labels":[],"label_agreement":null},{"id":"W2160628682","doi":"10.1002/sim.5374","title":"A Bayesian method for estimating prevalence in the presence of a hidden sub‐population","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Statistics; Posterior probability; Bayesian probability; Population; Sampling (signal processing); Markov chain Monte Carlo; Sample (material); Gibbs sampling; Computer science; Econometrics; Mathematics; Demography","score_opus":0.03508214765957837,"score_gpt":0.3825456648440279,"score_spread":0.34746351718444957,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2160628682","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0046218317,0.00013503828,0.994649,0.00012734704,0.0000132197065,0.000045326076,0.000056338842,0.00007812413,0.00027380523],"genre_scores_gemma":[0.15830104,0.00054858957,0.8385036,0.00021530123,0.00010138985,0.00056814734,0.00039916293,0.000085646556,0.0012771069],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99389946,0.0038870822,0.00025421823,0.00086824957,0.00092757633,0.00016332272],"domain_scores_gemma":[0.9811323,0.01527483,0.001108135,0.0013094755,0.0009338171,0.00024147141],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01458608,0.00091124006,0.0019218076,0.0037366608,0.0010006579,0.0020311978,0.0031699964,0.00241027,0.002711784],"category_scores_gemma":[0.053439982,0.0011409182,0.0018382863,0.002591009,0.0021179405,0.003655561,0.0020404418,0.0030552216,0.0006246896],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00039740335,0.000252896,0.020017162,0.00054811075,0.0008432484,0.00039351775,0.0013338635,0.26311037,0.0066684084,0.3990511,0.0036100491,0.3037739],"study_design_scores_gemma":[0.00007034845,0.000084852385,0.003034758,0.00014352343,0.00013050831,0.0004971848,0.00008860108,0.7580731,0.0012107122,0.2329164,0.0036496297,0.00010040297],"about_ca_topic_score_codex":0.005063997,"about_ca_topic_score_gemma":0.004714043,"teacher_disagreement_score":0.01458608,"about_ca_system_score_codex":0.0012699389,"about_ca_system_score_gemma":0.0016261882,"threshold_uncertainty_score":0.07713944},"labels":[],"label_agreement":null},{"id":"W2161040745","doi":"10.1002/sim.2073","title":"Using generalized additive models to reduce residual confounding","year":2004,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":103,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Montreal General Hospital","funders":"","keywords":"Covariate; Confounding; Statistics; Logistic regression; Mathematics; Parametric statistics; Context (archaeology); Generalized linear model; Econometrics; Linear regression","score_opus":0.3116101435216381,"score_gpt":0.5048658643751648,"score_spread":0.19325572085352666,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2161040745","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0031662893,0.00077574456,0.9934495,0.00035071932,0.00015449717,0.00012923652,0.0003065706,0.0005886152,0.0010788841],"genre_scores_gemma":[0.12969592,0.0024438205,0.8584077,0.0007146479,0.00031080924,0.0014266436,0.0011008956,0.0007191746,0.0051804157],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98415923,0.012078535,0.00059117266,0.0015066495,0.0012829276,0.0003814747],"domain_scores_gemma":[0.9756437,0.019855196,0.001293358,0.0015716474,0.0014771161,0.00015903545],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.019493867,0.0028054162,0.0032419008,0.0027075536,0.00089577225,0.0033452385,0.004014543,0.0024534913,0.0076556634],"category_scores_gemma":[0.052948758,0.0012958118,0.005238638,0.0035455027,0.0011005584,0.0020763038,0.0036452336,0.0041571045,0.0016296501],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00030060372,0.00022520738,0.008845039,0.0015626871,0.0035418996,0.0011851116,0.0010570273,0.5401752,0.002183527,0.23730049,0.011308967,0.19231424],"study_design_scores_gemma":[0.00012558843,0.00023114451,0.0017677001,0.0002983254,0.0008455682,0.00032788972,0.00016027718,0.73966295,0.00096782844,0.22840813,0.027066484,0.00013817033],"about_ca_topic_score_codex":0.013381085,"about_ca_topic_score_gemma":0.013697355,"teacher_disagreement_score":0.019493867,"about_ca_system_score_codex":0.0014895145,"about_ca_system_score_gemma":0.0036471908,"threshold_uncertainty_score":0.10309464},"labels":[],"label_agreement":null},{"id":"W2162274668","doi":"10.1002/sim.2624","title":"Using mixed treatment comparisons and meta‐regression to perform indirect comparisons to estimate the efficacy of biologic treatments in rheumatoid arthritis","year":2006,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Rheumatoid Arthritis Research and Therapies","field":"Medicine","cited_by":152,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; St. Paul's Hospital","funders":"","keywords":"Meta-analysis; Medicine; Odds ratio; Rheumatoid arthritis; Internal medicine; Random effects model; Meta-regression; Placebo; Randomized controlled trial; Sample size determination; Statistics; Pathology; Mathematics","score_opus":0.14527511481052424,"score_gpt":0.45913485076603355,"score_spread":0.3138597359555093,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2162274668","genre_codex":"methods","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010374703,0.14272355,0.82674456,0.002987561,0.0033915695,0.0051901625,0.0038630585,0.002486928,0.0022378413],"genre_scores_gemma":[0.1483975,0.024904752,0.8024248,0.0017721285,0.0010157427,0.017218702,0.0019937141,0.0005846937,0.0016880155],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.69387347,0.2852209,0.008253265,0.0067651886,0.0053152787,0.0005719929],"domain_scores_gemma":[0.71724194,0.26088423,0.008715756,0.010558035,0.002168039,0.00043203702],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.17657143,0.0053471504,0.014093762,0.012907891,0.0007980403,0.005786788,0.005661392,0.0040387814,0.0059743263],"category_scores_gemma":[0.2701223,0.0024351405,0.041108694,0.010610695,0.0013157574,0.004351894,0.0038303288,0.005665165,0.0007165191],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.005457101,0.00018462806,0.00607146,0.05274395,0.70043516,0.00052658375,0.00048130855,0.07927027,0.00078889396,0.035515234,0.0082266005,0.11029885],"study_design_scores_gemma":[0.0073294356,0.0032129565,0.006324171,0.010141139,0.46180156,0.0006451927,0.00033532546,0.23933622,0.0032531605,0.23249567,0.034348335,0.00077673246],"about_ca_topic_score_codex":0.003484391,"about_ca_topic_score_gemma":0.0034890452,"teacher_disagreement_score":0.17657143,"about_ca_system_score_codex":0.0036066228,"about_ca_system_score_gemma":0.0033102022,"threshold_uncertainty_score":0.9338101},"labels":[],"label_agreement":null},{"id":"W2162875017","doi":"10.1002/sim.707","title":"Estimating the number needed to treat (NNT) index when the data are subject to error","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Number needed to treat; Statistics; Medicine; Event (particle physics); Randomized controlled trial; Reliability (semiconductor); Adverse effect; Confidence interval; Mathematics; Relative risk; Internal medicine","score_opus":0.4318392061353256,"score_gpt":0.49505066292947963,"score_spread":0.06321145679415402,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2162875017","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.021027256,0.017292526,0.9441683,0.005187206,0.0019563143,0.0048085437,0.0015632487,0.0004524726,0.0035441173],"genre_scores_gemma":[0.35868347,0.007926077,0.6022204,0.005902424,0.0015689194,0.019611083,0.0019912315,0.00026524745,0.0018311171],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6299058,0.314941,0.020561485,0.011512484,0.022024564,0.0010546576],"domain_scores_gemma":[0.31233567,0.6453544,0.021285027,0.013316305,0.0070789745,0.00062961545],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.28274786,0.0022491373,0.0100423945,0.004994032,0.00083011546,0.0043226597,0.00417856,0.0076288986,0.002409438],"category_scores_gemma":[0.6336968,0.0009896632,0.0050940053,0.0054257596,0.0039863875,0.0056052026,0.003727075,0.006949503,0.00069288135],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008273752,0.00066057086,0.087979056,0.013570574,0.039435398,0.00094369263,0.0015635645,0.22133525,0.001306488,0.11526048,0.020068036,0.4896031],"study_design_scores_gemma":[0.0022416685,0.002806828,0.022545459,0.003911488,0.011217979,0.0012368743,0.00046443686,0.45089933,0.002965905,0.47089568,0.03026258,0.00055175135],"about_ca_topic_score_codex":0.0015918517,"about_ca_topic_score_gemma":0.0009409605,"teacher_disagreement_score":0.71725214,"about_ca_system_score_codex":0.0041636177,"about_ca_system_score_gemma":0.0033947178,"threshold_uncertainty_score":0.88449955},"labels":[],"label_agreement":null},{"id":"W2163071463","doi":"10.1002/sim.1121","title":"Sample size re‐estimation in cluster randomization trials","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":91,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cancer Care Ontario; University of Toronto","funders":"","keywords":"Sample size determination; Randomization; Statistics; Cluster (spacecraft); Cluster randomised controlled trial; Computer science; Statistical power; Sample (material); Restricted randomization; Econometrics; Clinical trial; Mathematics; Randomized controlled trial; Medicine","score_opus":0.5611285954998531,"score_gpt":0.5843085060944774,"score_spread":0.02317991059462432,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2163071463","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0053843167,0.0018830624,0.98459387,0.0012323054,0.00054526184,0.004442074,0.0000938904,0.000384754,0.001440548],"genre_scores_gemma":[0.18489686,0.0010387505,0.7907106,0.0016269778,0.00036974368,0.019972254,0.00020608875,0.0002546432,0.0009240345],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.50568545,0.4637762,0.010105132,0.0065122666,0.013183618,0.0007372341],"domain_scores_gemma":[0.34777993,0.5699346,0.028921638,0.040934443,0.011486103,0.0009433175],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.30958125,0.0021467246,0.0049605644,0.0039582714,0.0008801922,0.0024399802,0.0038800344,0.0036942516,0.0035341291],"category_scores_gemma":[0.66172415,0.0012774664,0.003063992,0.0032083248,0.0038116868,0.003230787,0.0035863991,0.0051840832,0.00082015165],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009332203,0.00043424504,0.0070502213,0.008179449,0.0052037477,0.00042671355,0.002621354,0.0672251,0.0019307338,0.19750822,0.017142339,0.6829456],"study_design_scores_gemma":[0.008685493,0.0052060555,0.005631521,0.0055847066,0.0033469182,0.00081663695,0.00036952723,0.3423182,0.008364727,0.58395463,0.03528649,0.0004351079],"about_ca_topic_score_codex":0.000614202,"about_ca_topic_score_gemma":0.0005412968,"teacher_disagreement_score":0.6904187,"about_ca_system_score_codex":0.0021090584,"about_ca_system_score_gemma":0.0031435832,"threshold_uncertainty_score":0.85140926},"labels":[],"label_agreement":null},{"id":"W2163077119","doi":"10.1002/sim.3788","title":"Risk‐adjusted survival time monitoring with an updating exponentially weighted moving average (EWMA) control chart","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":99,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"EWMA chart; Control chart; Chart; Computer science; Statistics; Moving average; Set (abstract data type); Data mining; Medicine; Mathematics; Process (computing)","score_opus":0.047221581457835254,"score_gpt":0.3800120459529197,"score_spread":0.3327904644950844,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2163077119","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.020306675,0.00014540614,0.9771712,0.0001477268,0.000052877505,0.00018386354,0.00016177414,0.001482541,0.00034793708],"genre_scores_gemma":[0.44120735,0.00026214286,0.556173,0.00012071849,0.00014940788,0.00057982793,0.00059173425,0.0001787687,0.0007369686],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9932388,0.0031749764,0.00076314266,0.0010873885,0.0015772087,0.00015851547],"domain_scores_gemma":[0.96153146,0.023717063,0.0054678563,0.0034981566,0.0052483063,0.0005371646],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013174616,0.000975938,0.0013761596,0.001962399,0.0004859408,0.0018082365,0.0018042515,0.00073089363,0.0010666619],"category_scores_gemma":[0.055740263,0.00044552135,0.00062652817,0.0014678227,0.0006598677,0.0020454505,0.0008374291,0.001652159,0.00021029192],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015176736,0.00055481197,0.032904707,0.00022460298,0.00036340128,0.0003189501,0.00039745457,0.28330675,0.010840178,0.023148078,0.0054526706,0.6409707],"study_design_scores_gemma":[0.000076482764,0.00025426561,0.003490329,0.00001667595,0.000061539846,0.000095950454,0.000013621225,0.986535,0.0047517973,0.00321102,0.0014380804,0.00005517817],"about_ca_topic_score_codex":0.006051195,"about_ca_topic_score_gemma":0.003289266,"teacher_disagreement_score":0.013174616,"about_ca_system_score_codex":0.0006934819,"about_ca_system_score_gemma":0.0015522299,"threshold_uncertainty_score":0.06967485},"labels":[],"label_agreement":null},{"id":"W2163654519","doi":"10.1002/sim.4321","title":"Bayesian inference on joint models of HIV dynamics for time‐to‐event and longitudinal data with skewness and covariate measurement errors","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":56,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Institute of Allergy and Infectious Diseases; Natural Sciences and Engineering Research Council of Canada; National Institutes of Health","keywords":"Covariate; Skewness; Bayesian probability; Computer science; Event (particle physics); Econometrics; Random effects model; Bayesian inference; Statistics; Inference; Mathematics; Artificial intelligence; Medicine","score_opus":0.1320334529644685,"score_gpt":0.3338156581999011,"score_spread":0.2017822052354326,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2163654519","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017516581,0.00026850216,0.9812391,0.00030142016,0.000017176639,0.000039456474,0.00013436859,0.00012511159,0.0003582496],"genre_scores_gemma":[0.5886517,0.0021786687,0.40160057,0.0004057987,0.00022457314,0.00079695805,0.0015954621,0.00019495332,0.0043512373],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9919001,0.005882759,0.00026811965,0.001077804,0.00058925967,0.00028198882],"domain_scores_gemma":[0.9502545,0.044177324,0.002640016,0.0016905909,0.00082859857,0.00040894552],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020878132,0.0010953769,0.0022580675,0.0023690567,0.0008081071,0.0022179424,0.0023655724,0.0020957952,0.0015492811],"category_scores_gemma":[0.07240882,0.0017640035,0.0023685913,0.0023722944,0.0024993264,0.003943445,0.0022023271,0.0031635389,0.00038610207],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001508728,0.0000978934,0.007628796,0.00013987813,0.0003787386,0.00013587627,0.00042591442,0.6945099,0.00078476686,0.24760136,0.0012472424,0.04689879],"study_design_scores_gemma":[0.000025938065,0.000024933692,0.0014712907,0.000027903352,0.00003938972,0.00004042015,0.000027694972,0.8687706,0.00016465846,0.12877598,0.0006024148,0.00002882902],"about_ca_topic_score_codex":0.012923976,"about_ca_topic_score_gemma":0.012162761,"teacher_disagreement_score":0.020878132,"about_ca_system_score_codex":0.0022191352,"about_ca_system_score_gemma":0.0024473893,"threshold_uncertainty_score":0.11041546},"labels":[],"label_agreement":null},{"id":"W2164498941","doi":"10.1002/sim.2580","title":"A comparison of the ability of different propensity score models to balance measured variables between treated and untreated subjects: a Monte Carlo study","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1134,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Propensity score matching; Confounding; Observational study; Outcome (game theory); Statistics; Matching (statistics); Medicine; Average treatment effect; Variables; Econometrics; Mathematics","score_opus":0.20433493146265783,"score_gpt":0.40639983108407973,"score_spread":0.2020648996214219,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2164498941","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.84632206,0.0015261852,0.14794803,0.00065516296,0.00006792765,0.0006625701,0.0003037346,0.00015572982,0.0023586429],"genre_scores_gemma":[0.9491639,0.0005614044,0.04845314,0.00026119972,0.000036917125,0.00046521844,0.000405913,0.00005329924,0.0005989801],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9815192,0.015978612,0.0005199485,0.0008710177,0.00074026966,0.0003708512],"domain_scores_gemma":[0.740361,0.24029021,0.005693611,0.009473369,0.0030889215,0.0010929554],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04845132,0.0008524176,0.0016802121,0.0014264224,0.0007175,0.0013477242,0.001399815,0.0018137979,0.0011688459],"category_scores_gemma":[0.122049816,0.0007098514,0.0024247332,0.0011862394,0.0016824844,0.0019069549,0.0011314055,0.0016368141,0.00015688929],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0060498635,0.0012473614,0.048527118,0.00029503863,0.0025140315,0.0002295064,0.00064618123,0.86531645,0.0011689281,0.038750008,0.001361311,0.03389415],"study_design_scores_gemma":[0.0004035115,0.0011159499,0.007384622,0.00008313108,0.0002840767,0.00010269539,0.00010450498,0.9747498,0.0009780137,0.014049841,0.00067083293,0.000073083254],"about_ca_topic_score_codex":0.007454513,"about_ca_topic_score_gemma":0.003993871,"teacher_disagreement_score":0.9515487,"about_ca_system_score_codex":0.0019863462,"about_ca_system_score_gemma":0.0016740398,"threshold_uncertainty_score":0.2562381},"labels":[],"label_agreement":null},{"id":"W2164627526","doi":"10.1002/sim.6395","title":"Sample size and robust marginal methods for cluster‐randomized trials with censored event times","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Canadian Institutes of Health Research; Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs","keywords":"Censoring (clinical trials); Statistics; Sample size determination; Estimator; Copula (linguistics); Marginal model; Mathematics; Econometrics; Proportional hazards model; Censored regression model; Regression analysis; Computer science","score_opus":0.09310038489766441,"score_gpt":0.46693814710278747,"score_spread":0.37383776220512305,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2164627526","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0005795897,0.00069732044,0.9971705,0.0003961043,0.00009723362,0.00034860222,0.000063709434,0.00015370111,0.000493258],"genre_scores_gemma":[0.058867924,0.0012408167,0.92955536,0.0007553278,0.00043601313,0.007247296,0.00030312783,0.00030837688,0.0012856562],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9012337,0.088400565,0.0020587873,0.0033063937,0.004491565,0.0005090695],"domain_scores_gemma":[0.70727915,0.2632852,0.007347321,0.015062398,0.006037995,0.0009879564],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.119006455,0.001966274,0.0043015177,0.0039214375,0.00082665135,0.0025976163,0.0060066357,0.0029188497,0.006255871],"category_scores_gemma":[0.31938922,0.0013865194,0.0030646406,0.0031474526,0.0038704665,0.0035052486,0.0042563532,0.0062433057,0.0010388833],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00086033397,0.00015055959,0.001568548,0.001394114,0.0010956153,0.00019439163,0.0005446932,0.06869783,0.0006910928,0.7559228,0.005921375,0.16295873],"study_design_scores_gemma":[0.0006367421,0.00040569998,0.0006786133,0.000436999,0.00033919635,0.00013199286,0.000070106486,0.28552154,0.0007383274,0.7003754,0.0106035825,0.00006188274],"about_ca_topic_score_codex":0.0015746609,"about_ca_topic_score_gemma":0.0013135918,"teacher_disagreement_score":0.88099355,"about_ca_system_score_codex":0.002530425,"about_ca_system_score_gemma":0.004919265,"threshold_uncertainty_score":0.6293738},"labels":[],"label_agreement":null},{"id":"W2164652223","doi":"10.1002/sim.6387","title":"Effects of categorization method, regression type, and variable distribution on the inflation of Type‐I error rate when categorizing a confounding variable","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":42,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Q & T Research; Université de Sherbrooke","funders":"Fonds de Recherche du Québec - Santé; Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Statistics; Confounding; Econometrics; Proxy (statistics); Type I and type II errors; Logistic regression; Categorization; Latent variable; Variable (mathematics); Regression analysis; Regression; Mathematics; Inflation (cosmology); Linear regression; Computer science; Artificial intelligence","score_opus":0.059187419105748394,"score_gpt":0.4150197663453947,"score_spread":0.35583234723964635,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2164652223","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1650859,0.004557969,0.8150555,0.0040980116,0.00085904734,0.0017975597,0.0008753224,0.0016926312,0.005978076],"genre_scores_gemma":[0.67569643,0.00112219,0.3151533,0.0023829078,0.00019430486,0.0022201284,0.0008160451,0.00094042835,0.0014743069],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.7160695,0.22856334,0.015633417,0.018661618,0.017425347,0.0036468047],"domain_scores_gemma":[0.15345299,0.7829527,0.018099064,0.035756327,0.008712156,0.0010266725],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3364978,0.0017611546,0.0020263514,0.0025564702,0.0021505,0.0040401975,0.0026442998,0.004070569,0.0035158512],"category_scores_gemma":[0.57515115,0.0013997523,0.0041216994,0.003095515,0.0068642073,0.004978411,0.0038381272,0.0065725194,0.00083061197],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.011075338,0.00096488214,0.28110906,0.0032087448,0.0058686347,0.002602655,0.0066778064,0.18528631,0.011500271,0.142403,0.014351997,0.33495128],"study_design_scores_gemma":[0.0011294445,0.0024707748,0.07747065,0.0041280515,0.0037158115,0.0038233872,0.0016076724,0.56530803,0.03665248,0.2874131,0.015101391,0.0011792501],"about_ca_topic_score_codex":0.0037751927,"about_ca_topic_score_gemma":0.0033565706,"teacher_disagreement_score":0.6635022,"about_ca_system_score_codex":0.003546569,"about_ca_system_score_gemma":0.00306644,"threshold_uncertainty_score":0.8182163},"labels":[],"label_agreement":null},{"id":"W2166148588","doi":"10.1002/sim.3828","title":"A nonstationary Markov transition model for computing the relative risk of dementia before death","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"National Institute on Aging; Robert J. Kleberg, Jr. and Helen C. Kleberg Foundation","keywords":"Dementia; Markov model; Transition (genetics); Markov chain; Computer science; Econometrics; Medicine; Mathematics; Disease; Machine learning; Internal medicine","score_opus":0.043449792847671864,"score_gpt":0.3868699731316014,"score_spread":0.3434201802839295,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2166148588","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.016849129,0.00016790834,0.9822594,0.0001751484,0.000027509113,0.000028317567,0.00008931272,0.00013309499,0.0002701977],"genre_scores_gemma":[0.5882517,0.0008847199,0.40610167,0.00022853147,0.00020076812,0.0005101846,0.00081584975,0.000119088065,0.002887395],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9983962,0.00087682693,0.00009211208,0.0003075939,0.00021846357,0.000108706925],"domain_scores_gemma":[0.98096126,0.016858991,0.0009888986,0.0005349564,0.00045643188,0.00019950132],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0063902307,0.00068415113,0.0013218807,0.00166256,0.0006197093,0.0012375071,0.0020571512,0.0016086517,0.0025964787],"category_scores_gemma":[0.030165477,0.00076357706,0.0013409378,0.0012936982,0.0012785784,0.0021002935,0.0012197567,0.0020899563,0.00047763844],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012174545,0.000058388097,0.0033765635,0.00008956978,0.00011941811,0.000117074975,0.00012933338,0.88083655,0.0007175242,0.092290424,0.00067924015,0.021464111],"study_design_scores_gemma":[0.000009772597,0.00002786241,0.00032782683,0.000008689964,0.000016406495,0.00002659895,0.000007880132,0.9710407,0.000108680666,0.028219648,0.00019490249,0.0000109939265],"about_ca_topic_score_codex":0.009126819,"about_ca_topic_score_gemma":0.006442997,"teacher_disagreement_score":0.009126819,"about_ca_system_score_codex":0.0014175214,"about_ca_system_score_gemma":0.0015839692,"threshold_uncertainty_score":0.033795178},"labels":[],"label_agreement":null},{"id":"W2166900700","doi":"10.1002/sim.5970","title":"Age‐period‐cohort models using smoothing splines: a generalized additive model approach","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Insurance, Mortality, Demography, Risk Management","field":"Social Sciences","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Smoothing; Smoothing spline; Identifiability; Box spline; Computer science; Mathematics; Applied mathematics; Dependency (UML); Econometrics; Cohort effect; Mathematical optimization; Cohort; Statistics; Artificial intelligence","score_opus":0.05897676816116972,"score_gpt":0.35134934774278764,"score_spread":0.29237257958161794,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2166900700","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015539243,0.00034444948,0.9821578,0.00041308423,0.00008034291,0.00010064407,0.00041674534,0.00031357756,0.00063409476],"genre_scores_gemma":[0.4521749,0.0024855556,0.5344811,0.00026769043,0.0002712137,0.0011091472,0.0017331517,0.00025325335,0.007223959],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9931497,0.005479135,0.00017478873,0.00048633432,0.0004578537,0.0002522403],"domain_scores_gemma":[0.9840821,0.01249425,0.0008875127,0.0012679952,0.0010235102,0.00024460498],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016605437,0.0010225923,0.0018374309,0.0022825839,0.0006816069,0.0016751448,0.002858999,0.0014795933,0.0029926454],"category_scores_gemma":[0.02969734,0.00066506345,0.0025760136,0.0038256338,0.00078947016,0.0014038454,0.0018303746,0.0025898495,0.0006222433],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00033597377,0.00015486078,0.013336754,0.00028630925,0.0009732657,0.0003645389,0.0006651856,0.70271456,0.0005720243,0.17138068,0.0054490445,0.10376682],"study_design_scores_gemma":[0.000045897024,0.00012266795,0.0016465334,0.00003753716,0.00013878442,0.0000735547,0.000079099234,0.917998,0.00011142022,0.07643248,0.0032665052,0.000047488404],"about_ca_topic_score_codex":0.018388895,"about_ca_topic_score_gemma":0.013203821,"teacher_disagreement_score":0.018388895,"about_ca_system_score_codex":0.0009054487,"about_ca_system_score_gemma":0.0028507796,"threshold_uncertainty_score":0.08781898},"labels":[],"label_agreement":null},{"id":"W2167706052","doi":"10.1002/sim.4015","title":"Interval estimation for the area under the receiver operating characteristic curve when data are subject to error","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"Ontario Ministry of Research and Innovation; Natural Sciences and Engineering Research Council of Canada","keywords":"Receiver operating characteristic; Confidence interval; Statistics; Variance (accounting); Computer science; Observational error; Interval (graph theory); Function (biology); Mathematics; Algorithm","score_opus":0.6034699099556665,"score_gpt":0.5849274279005707,"score_spread":0.018542482055095788,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2167706052","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0027057447,0.0005112213,0.9961349,0.00008225721,0.000040226718,0.000053594693,0.000044648234,0.00018975454,0.00023765917],"genre_scores_gemma":[0.17642082,0.0012810379,0.8193224,0.00027658298,0.0002884611,0.00085214846,0.00051533227,0.00029781618,0.00074531697],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95298654,0.03500459,0.002040213,0.0042008213,0.005304274,0.00046341732],"domain_scores_gemma":[0.66600376,0.30213138,0.011247589,0.013642196,0.006434582,0.0005404805],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06502035,0.0012715576,0.0024188994,0.0041036787,0.00059624296,0.0019446792,0.0032190306,0.0026681686,0.0019556503],"category_scores_gemma":[0.3047051,0.0004862417,0.0019835513,0.0029938535,0.0021501253,0.0027614934,0.0023549173,0.003989939,0.0006257849],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012865759,0.00025066634,0.02062925,0.002027297,0.0023072932,0.0008817475,0.0015722313,0.18385577,0.0072039715,0.12932682,0.0052605476,0.6453979],"study_design_scores_gemma":[0.00020350153,0.0010086566,0.011875066,0.0006953115,0.0007215029,0.0017844503,0.00022675753,0.80910295,0.009647583,0.15411511,0.01035575,0.00026330564],"about_ca_topic_score_codex":0.0011947278,"about_ca_topic_score_gemma":0.0004943058,"teacher_disagreement_score":0.06502035,"about_ca_system_score_codex":0.00081641204,"about_ca_system_score_gemma":0.0012508365,"threshold_uncertainty_score":0.34386462},"labels":[],"label_agreement":null},{"id":"W2168082315","doi":"10.1002/sim.6020","title":"Marginal structural models for skewed outcomes: identifying causal relationships in health care utilization","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Institut National de Santé Publique du Québec; Sante Montreal; McGill University; McGill University Health Centre","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Negative binomial distribution; Causal inference; Generalized linear model; Marginal structural model; Inference; Marginal model; Emergency department; Propensity score matching; Econometrics; Poisson distribution; Psychological intervention; Statistics; Baseline (sea); Generalized linear mixed model; Robustness (evolution); Count data; Medicine; Computer science; Mathematics","score_opus":0.3202027902674543,"score_gpt":0.4968382609247534,"score_spread":0.17663547065729912,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2168082315","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.20636442,0.00066399906,0.7869327,0.002403895,0.000063412735,0.00039864326,0.0007558538,0.00029511194,0.002121967],"genre_scores_gemma":[0.90294313,0.0004986846,0.09377538,0.00024891287,0.00009480736,0.0004995904,0.00048993237,0.00004691515,0.0014025758],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98165417,0.01491744,0.00040733267,0.0014856724,0.0009003903,0.00063497835],"domain_scores_gemma":[0.770366,0.20348348,0.011695347,0.011311071,0.0023251646,0.00081889157],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.040777378,0.00095325836,0.0017639677,0.002094201,0.0008815238,0.0018828103,0.003063973,0.0013840965,0.0062560164],"category_scores_gemma":[0.17871334,0.0007062097,0.0029564647,0.0024610069,0.0035666926,0.003207521,0.0031976264,0.0035754475,0.00038767213],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00069524377,0.00032647725,0.10425092,0.0003813853,0.0014123063,0.0007044756,0.0023292825,0.29670662,0.0005640402,0.4934606,0.0026196684,0.09654895],"study_design_scores_gemma":[0.00014794226,0.00019243965,0.012159452,0.00009294873,0.00023404397,0.00014681157,0.0003964502,0.5704511,0.0003095624,0.41441378,0.0014132606,0.000042210137],"about_ca_topic_score_codex":0.016279882,"about_ca_topic_score_gemma":0.014989972,"teacher_disagreement_score":0.040777378,"about_ca_system_score_codex":0.002572402,"about_ca_system_score_gemma":0.0027413096,"threshold_uncertainty_score":0.2156539},"labels":[],"label_agreement":null},{"id":"W2168158999","doi":"10.1002/sim.3356","title":"Two‐stage design for dose‐finding that accounts for both efficacy and safety","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":47,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Women's Health Research Institute","funders":"","keywords":"Stage (stratigraphy); Computer science; Statistics; Econometrics; Mathematics; Biology","score_opus":0.34766234225019504,"score_gpt":0.5197216396087865,"score_spread":0.17205929735859143,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2168158999","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015171904,0.00023203727,0.9582912,0.00030537165,0.00037975784,0.023476174,0.00025576516,0.0002957148,0.0015920519],"genre_scores_gemma":[0.0955166,0.0002247883,0.7954196,0.00046144548,0.00014319106,0.105677046,0.0002009522,0.00007295831,0.0022833582],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9388827,0.048648424,0.0020176289,0.0039144983,0.005103808,0.0014329491],"domain_scores_gemma":[0.94836116,0.035938844,0.003934215,0.007289604,0.003474448,0.0010016764],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06671376,0.003224384,0.0044829333,0.001399469,0.0010946289,0.001723353,0.0033206535,0.0037556116,0.012282545],"category_scores_gemma":[0.068638474,0.0020053145,0.0030043013,0.0013844016,0.0025360533,0.0019636406,0.0023489292,0.004101384,0.0016364643],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.058586325,0.011056583,0.0105647035,0.004269941,0.0034828912,0.00038098477,0.0013729349,0.1998634,0.024892144,0.35673976,0.0067396243,0.3220507],"study_design_scores_gemma":[0.036276747,0.07594757,0.009888832,0.0003753021,0.002392761,0.0003781227,0.0001363254,0.60998046,0.015272371,0.21032394,0.03834266,0.00068488304],"about_ca_topic_score_codex":0.0006638318,"about_ca_topic_score_gemma":0.0008334265,"teacher_disagreement_score":0.06671376,"about_ca_system_score_codex":0.0024608476,"about_ca_system_score_gemma":0.0051375804,"threshold_uncertainty_score":0.35282028},"labels":[],"label_agreement":null},{"id":"W2168289333","doi":"10.1002/sim.2554","title":"A comparison of methods for the analysis of recurrent health outcome data with environmental covariates","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Air Quality and Health Impacts","field":"Environmental Science","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Population and Public Health; University of Ottawa; University of Windsor","funders":"","keywords":"Covariate; Estimator; Statistics; Poisson distribution; Outcome (game theory); Crossover; Generalized linear model; Poisson regression; Constant (computer programming); Mathematics; Econometrics; Computer science; Medicine; Artificial intelligence","score_opus":0.20357769507771378,"score_gpt":0.5379687256093808,"score_spread":0.334391030531667,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2168289333","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011952981,0.0019701545,0.9847209,0.00019221258,0.00006221522,0.00027104482,0.00016065322,0.00035808946,0.00031175622],"genre_scores_gemma":[0.06533478,0.0021868867,0.9298415,0.00011478571,0.00007335867,0.0010773657,0.00057296216,0.00021549183,0.0005828251],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.94910085,0.040270105,0.002166041,0.0028129763,0.005364767,0.00028525188],"domain_scores_gemma":[0.7749554,0.20300043,0.004464879,0.010693189,0.006329213,0.000556887],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.057819374,0.0013247096,0.001789143,0.0057351333,0.00051572686,0.0017191644,0.0027944127,0.0021833677,0.0017935503],"category_scores_gemma":[0.17259355,0.0011670897,0.003137012,0.0041151163,0.0010824219,0.0026174593,0.0026978643,0.0023018222,0.00061108574],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0027138677,0.00043946508,0.0290474,0.0015871042,0.0043639126,0.00019246763,0.0010632328,0.08533926,0.0041363905,0.056014128,0.0021947934,0.8129079],"study_design_scores_gemma":[0.00088383525,0.0008280612,0.030098312,0.00048872246,0.0007488111,0.0009510487,0.00033769268,0.8799259,0.0039708707,0.0719711,0.009356388,0.00043924767],"about_ca_topic_score_codex":0.004028751,"about_ca_topic_score_gemma":0.0036251855,"teacher_disagreement_score":0.057819374,"about_ca_system_score_codex":0.0012332117,"about_ca_system_score_gemma":0.0020210028,"threshold_uncertainty_score":0.30578178},"labels":[],"label_agreement":null},{"id":"W2169111089","doi":"10.1002/sim.3694","title":"Bayesian analysis of a matched case–control study with expert prior information on both the misclassification of exposure and the exposure–disease association","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Distribution Estimation and Applications","field":"Mathematics","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; University of Alberta; University of Saskatchewan","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Alberta Heritage Foundation for Medical Research","keywords":"Dirichlet distribution; Multinomial distribution; Bayesian probability; Statistics; Association (psychology); Prior probability; Disease; Computer science; A priori and a posteriori; Econometrics; Medicine; Mathematics; Internal medicine; Psychology","score_opus":0.02053217632805361,"score_gpt":0.33561580847107075,"score_spread":0.31508363214301716,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2169111089","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.15348254,0.0033332387,0.83934385,0.0010876363,0.00023107424,0.00037303558,0.00042863612,0.00020667897,0.0015133199],"genre_scores_gemma":[0.8157778,0.0013847543,0.17847365,0.0007413851,0.00039273073,0.00076157606,0.00085292506,0.000045780707,0.0015694712],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9328755,0.053171154,0.0026164046,0.005794026,0.004844801,0.00069811486],"domain_scores_gemma":[0.83723557,0.13759504,0.009529672,0.012302328,0.0025612428,0.0007760886],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.10177682,0.0010404136,0.0030316417,0.0037322503,0.0009222202,0.0020423848,0.0037111046,0.0025480303,0.002894841],"category_scores_gemma":[0.18700081,0.0011627991,0.0028302183,0.0020982479,0.0016414938,0.0012433057,0.001934191,0.0017159913,0.0002809324],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.010472633,0.0014660661,0.20744008,0.0024995795,0.02172595,0.0055987895,0.0026652734,0.15443769,0.007632305,0.2205489,0.0055063744,0.36000642],"study_design_scores_gemma":[0.0020204172,0.0011095045,0.06695975,0.00041253906,0.00837808,0.0020797914,0.000277582,0.69157344,0.0026910573,0.21754381,0.0067510996,0.00020288685],"about_ca_topic_score_codex":0.005614305,"about_ca_topic_score_gemma":0.0032073855,"teacher_disagreement_score":0.10177682,"about_ca_system_score_codex":0.0010822295,"about_ca_system_score_gemma":0.0013974987,"threshold_uncertainty_score":0.5382537},"labels":[],"label_agreement":null},{"id":"W2169258359","doi":"10.1002/sim.2981","title":"Characterizing the functional MRI response using Tikhonov regularization","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced MRI Techniques and Applications","field":"Medicine","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Saskatchewan","funders":"","keywords":"Tikhonov regularization; Regularization (linguistics); Computer science; Artificial intelligence; Mathematics; Inverse problem; Mathematical analysis","score_opus":0.04072377874506911,"score_gpt":0.37546591047137423,"score_spread":0.3347421317263051,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2169258359","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00850892,0.000059998667,0.9909644,0.000053304306,0.000008418545,0.00004104079,0.00003260494,0.00018391959,0.00014741109],"genre_scores_gemma":[0.23405245,0.0002853372,0.76200205,0.00015773755,0.000044997447,0.0009827708,0.000251182,0.00030625134,0.0019172288],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99745005,0.001205316,0.00013916653,0.00048539139,0.0005563737,0.00016371558],"domain_scores_gemma":[0.9950771,0.003482963,0.00047625822,0.00049038196,0.00038434728,0.000089056164],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008522557,0.0014390525,0.0017816578,0.0010878547,0.00034917405,0.0010923435,0.0017691583,0.002662676,0.0011134403],"category_scores_gemma":[0.016366752,0.0006986594,0.0015774432,0.00085452717,0.0013916986,0.0012361393,0.00093709497,0.0019091099,0.00041193978],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00037366635,0.00024273495,0.0016159537,0.0006552272,0.00032202576,0.00033518177,0.00033350513,0.6334498,0.20872557,0.044142283,0.0010857353,0.108718336],"study_design_scores_gemma":[0.000011423339,0.00010782583,0.0008836799,0.000008497009,0.000022810722,0.000059087346,0.000010302151,0.9774541,0.009413405,0.011351118,0.0006510663,0.000026693016],"about_ca_topic_score_codex":0.0026617798,"about_ca_topic_score_gemma":0.002884331,"teacher_disagreement_score":0.008522557,"about_ca_system_score_codex":0.0012622704,"about_ca_system_score_gemma":0.001989011,"threshold_uncertainty_score":0.04507214},"labels":[],"label_agreement":null},{"id":"W2169425545","doi":"10.1002/sim.6276","title":"The use of bootstrapping when using propensity‐score matching without replacement: a simulation study","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":278,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Health Services and Policy Research; University of Toronto; Institute for Work & Health; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Bootstrapping (finance); Statistics; Resampling; Standard error; Matching (statistics); Confidence interval; Sample size determination; Standard deviation; Sampling distribution; Observational study; Econometrics; Mathematics","score_opus":0.38854138632668395,"score_gpt":0.46673056535998564,"score_spread":0.07818917903330169,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2169425545","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5119101,0.0042368346,0.4669323,0.0023386364,0.0002870101,0.0028826233,0.0009938418,0.00029600348,0.010122686],"genre_scores_gemma":[0.81787974,0.0020100994,0.17528425,0.0003949073,0.00012632327,0.002420425,0.00071192434,0.00008999816,0.0010823831],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9766982,0.020502148,0.0007575348,0.0006807402,0.00095544965,0.0004059921],"domain_scores_gemma":[0.7162633,0.26068512,0.0054048337,0.0099404985,0.0068631875,0.00084315095],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.050555594,0.001215734,0.0017522867,0.001828218,0.0012101728,0.0015970112,0.0021306246,0.002479335,0.0032218164],"category_scores_gemma":[0.15563107,0.0006687638,0.002725584,0.0028805838,0.0013004261,0.0026859983,0.0018765338,0.003218769,0.00034245828],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0047195964,0.003925244,0.056405105,0.001437992,0.00201607,0.0012231233,0.0015842897,0.7029054,0.0012146996,0.12480492,0.0071539143,0.09260957],"study_design_scores_gemma":[0.0008707447,0.0012199797,0.003987448,0.00036791514,0.00055677403,0.00028524044,0.00035243962,0.9548794,0.0011498989,0.03293831,0.003290683,0.00010116572],"about_ca_topic_score_codex":0.008527655,"about_ca_topic_score_gemma":0.00665656,"teacher_disagreement_score":0.9494444,"about_ca_system_score_codex":0.0018566762,"about_ca_system_score_gemma":0.0017759546,"threshold_uncertainty_score":0.26736677},"labels":[],"label_agreement":null},{"id":"W2169649310","doi":"10.1002/sim.2892","title":"Accelerated failure time models with covariates subject to measurement error","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":49,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Covariate; Observational error; Estimator; Econometrics; Statistics; Proportional hazards model; Extrapolation; Computer science; Errors-in-variables models; Accelerated failure time model; Data set; Mathematics","score_opus":0.12878516186870306,"score_gpt":0.39882542307849034,"score_spread":0.2700402612097873,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2169649310","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.022906125,0.0010088586,0.97330666,0.00086717267,0.00017027443,0.00013678477,0.00029451327,0.00024082215,0.0010688789],"genre_scores_gemma":[0.6486276,0.0055387057,0.31793147,0.0007227941,0.0011022319,0.001980622,0.0018238872,0.00023076312,0.022041878],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.979752,0.0140390955,0.0007908915,0.0020921524,0.0023437275,0.0009821171],"domain_scores_gemma":[0.8790699,0.09782225,0.011216368,0.0061582364,0.004793033,0.00094025646],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.043434177,0.0020950036,0.00411573,0.0028185775,0.0011332256,0.002878809,0.005720396,0.0036450087,0.004576277],"category_scores_gemma":[0.10392434,0.0015991871,0.003259308,0.0041212356,0.0032906279,0.0037494188,0.0039590723,0.0052000675,0.0011999826],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00025917374,0.00011440573,0.009864296,0.0004099413,0.00042203293,0.0007396145,0.0008000602,0.5062264,0.00055144454,0.43965313,0.00227849,0.038680978],"study_design_scores_gemma":[0.000132908,0.00016639233,0.001707091,0.00010068422,0.00012589262,0.00020676367,0.000056337034,0.7587271,0.00022519368,0.23535354,0.0031280671,0.0000700841],"about_ca_topic_score_codex":0.007541768,"about_ca_topic_score_gemma":0.0041784346,"teacher_disagreement_score":0.043434177,"about_ca_system_score_codex":0.0016518004,"about_ca_system_score_gemma":0.0030210442,"threshold_uncertainty_score":0.22970462},"labels":[],"label_agreement":null},{"id":"W2169900945","doi":"10.1002/sim.2733","title":"Accounting for expected attrition in the planning of community intervention trials","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University; Ottawa Hospital","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Attrition; Sample size determination; Statistics; Randomization; Cohort; Demography; Population; Cluster randomised controlled trial; Sample (material); Cluster (spacecraft); Psychological intervention; Standard error; Econometrics; Medicine; Randomized controlled trial; Mathematics; Computer science; Environmental health; Surgery","score_opus":0.39420011802620225,"score_gpt":0.5438528487911372,"score_spread":0.1496527307649349,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2169900945","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0019387754,0.0010838736,0.99108535,0.0017231061,0.00014272265,0.0020935081,0.00013671993,0.00033051043,0.0014654583],"genre_scores_gemma":[0.07161114,0.0014533702,0.91345716,0.0010872101,0.00024957562,0.01076553,0.00025662428,0.00015044611,0.0009688686],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.607767,0.35690066,0.013179721,0.00740574,0.01340082,0.0013460519],"domain_scores_gemma":[0.29556453,0.6514468,0.02185399,0.019941904,0.01014112,0.0010515712],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3630015,0.002867408,0.0046194345,0.0044585145,0.0012117184,0.004043929,0.0050326074,0.0040011425,0.0045096707],"category_scores_gemma":[0.689842,0.0026758614,0.0033725505,0.0042917766,0.0036198634,0.008783027,0.0040946444,0.0064266864,0.0012755212],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018255045,0.00027967477,0.009941005,0.0055948533,0.0018692637,0.00071128475,0.0020972209,0.13808157,0.00064438937,0.3371189,0.009669697,0.49216667],"study_design_scores_gemma":[0.0010417887,0.0009978279,0.0025691367,0.0023751683,0.0010261402,0.000308994,0.0002034473,0.28029275,0.0026770344,0.69123954,0.017122565,0.00014554914],"about_ca_topic_score_codex":0.0026344678,"about_ca_topic_score_gemma":0.0029938237,"teacher_disagreement_score":0.63699853,"about_ca_system_score_codex":0.003962654,"about_ca_system_score_gemma":0.013576786,"threshold_uncertainty_score":0.7855326},"labels":[],"label_agreement":null},{"id":"W2169987255","doi":"10.1002/sim.6619","title":"Inference on cancer screening exam accuracy using population‐level administrative data","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Global Cancer Incidence and Screening","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McMaster University; Public Health Ontario; University of Toronto; Cancer Care Ontario","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Cancer Society; Cancer Care Ontario","keywords":"Covariate; Markov chain Monte Carlo; Bayesian probability; Statistics; Breast cancer; Inference; Computer science; Population; Cancer registry; Posterior probability; Bayesian inference; Disease; Econometrics; Cancer; Medicine; Artificial intelligence; Mathematics; Internal medicine; Environmental health","score_opus":0.7226447359803319,"score_gpt":0.5708361636674781,"score_spread":0.15180857231285372,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2169987255","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5139944,0.0012205762,0.47502458,0.002367159,0.00004808306,0.00015607165,0.0037553809,0.0005046564,0.0029291206],"genre_scores_gemma":[0.9460327,0.00058245007,0.04926861,0.00028542953,0.000071374874,0.00009588378,0.002960611,0.000032678672,0.0006702668],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98876464,0.0075712833,0.0005452564,0.001241894,0.0014792375,0.0003976102],"domain_scores_gemma":[0.9073459,0.07927152,0.005771391,0.005003476,0.002145688,0.0004620117],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02458669,0.00045655452,0.0015074307,0.0026780982,0.0005658136,0.0019910282,0.0016808145,0.0012136145,0.0009770478],"category_scores_gemma":[0.12691471,0.0010013441,0.001392445,0.0027359596,0.0013806781,0.0015750313,0.0013818272,0.0018209934,0.00029518388],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00041491137,0.00031043438,0.52254534,0.00021011973,0.0012072301,0.00030575765,0.00067788997,0.34116364,0.00069435756,0.045848586,0.0034085214,0.08321319],"study_design_scores_gemma":[0.00011596861,0.0000986659,0.12879539,0.00013910858,0.0003238844,0.00022015811,0.00018608643,0.8181377,0.0009410722,0.04858454,0.0023883379,0.000069019035],"about_ca_topic_score_codex":0.064675346,"about_ca_topic_score_gemma":0.053051136,"teacher_disagreement_score":0.9754133,"about_ca_system_score_codex":0.0025765735,"about_ca_system_score_gemma":0.0026172255,"threshold_uncertainty_score":0.13002843},"labels":[],"label_agreement":null},{"id":"W2170153043","doi":"10.1002/sim.4244","title":"Assessing noninferiority in a three‐arm trial using the Bayesian approach","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"","keywords":"Bayesian probability; Parametric statistics; Computer science; Placebo; Statistical hypothesis testing; Clinical trial; Statistics; Machine learning; Medicine; Artificial intelligence; Mathematics; Alternative medicine","score_opus":0.7454022113351202,"score_gpt":0.6058046770374885,"score_spread":0.13959753429763166,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2170153043","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.036143072,0.0019797499,0.95451444,0.0025643553,0.00017243107,0.0022404373,0.00020246263,0.00016060163,0.002022544],"genre_scores_gemma":[0.4006493,0.0015815685,0.58676785,0.0016206787,0.00031119492,0.00751158,0.00023697535,0.00008102307,0.0012397653],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.658263,0.3182516,0.0066768723,0.0056106374,0.010102325,0.0010955303],"domain_scores_gemma":[0.3726606,0.5962249,0.016025824,0.0102103315,0.0032040442,0.0016742881],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.32666445,0.0020788526,0.007893624,0.003915255,0.0010237772,0.0036544795,0.00379873,0.0057765534,0.006894521],"category_scores_gemma":[0.4188997,0.0010609218,0.0063008284,0.0024776468,0.00541848,0.0052644117,0.00376122,0.006944693,0.00048099845],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.023008723,0.0015132693,0.017540341,0.004561938,0.010070124,0.0010965202,0.0018743272,0.22973599,0.0032828678,0.42092127,0.0038450132,0.28254965],"study_design_scores_gemma":[0.0054327087,0.0036429053,0.004007972,0.0006521712,0.0020505027,0.00035634058,0.00011565166,0.59828234,0.0012848285,0.3812894,0.002714856,0.00017029648],"about_ca_topic_score_codex":0.0014316501,"about_ca_topic_score_gemma":0.0012325018,"teacher_disagreement_score":0.67333555,"about_ca_system_score_codex":0.0020906765,"about_ca_system_score_gemma":0.005868628,"threshold_uncertainty_score":0.83034265},"labels":[],"label_agreement":null},{"id":"W2170987805","doi":"10.1002/sim.5378","title":"Variable selection in semiparametric cure models based on penalized likelihood, with application to breast cancer clinical trials","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"National Science Foundation","keywords":"Breast cancer; Feature selection; Proportional hazards model; Computer science; Logistic regression; Model selection; Semiparametric regression; Accelerated failure time model; Clinical trial; Cure rate; Variable (mathematics); Statistics; Regression analysis; Econometrics; Cancer; Medicine; Mathematics; Artificial intelligence; Machine learning; Internal medicine","score_opus":0.17296653638878065,"score_gpt":0.5074455559875256,"score_spread":0.33447901959874493,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2170987805","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010115837,0.00072872936,0.9877963,0.00064068823,0.000033206958,0.00014393883,0.000047558606,0.0001607617,0.0003329513],"genre_scores_gemma":[0.49097356,0.0015050422,0.5014634,0.000558999,0.00025526504,0.002186834,0.00040365383,0.000265,0.0023883164],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9666588,0.031080622,0.0004756282,0.0006951593,0.0008028247,0.00028701],"domain_scores_gemma":[0.86934096,0.121348605,0.004182687,0.0023854026,0.002043878,0.0006985606],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04374678,0.0011795503,0.0023271965,0.0020617081,0.00063606206,0.0017118126,0.0022674657,0.0018241474,0.0018209969],"category_scores_gemma":[0.100400805,0.0011989794,0.0016809722,0.0021092573,0.002661214,0.0022005779,0.0024697068,0.003072865,0.00030720458],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006135621,0.0001083702,0.004036967,0.00042650217,0.000348616,0.00037111857,0.00035384396,0.8173328,0.0006082894,0.12171949,0.002188275,0.051892128],"study_design_scores_gemma":[0.00007955181,0.00007013008,0.00028250352,0.000032412194,0.00003013506,0.000055444456,0.000014371553,0.9642273,0.00016036411,0.03455881,0.00047015294,0.000018884495],"about_ca_topic_score_codex":0.0018005987,"about_ca_topic_score_gemma":0.0015546541,"teacher_disagreement_score":0.04374678,"about_ca_system_score_codex":0.0012056695,"about_ca_system_score_gemma":0.0017398678,"threshold_uncertainty_score":0.23135787},"labels":[],"label_agreement":null},{"id":"W2171380912","doi":"10.1002/sim.969","title":"A model for foetal growth and diagnosis of intrauterine growth restriction","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Birth, Development, and Health","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Gestational age; Intrauterine growth restriction; Fetal growth; Residual; Pregnancy; Fetus; Growth model; Birth weight; Covariance; Obstetrics; Growth curve (statistics); Medicine; Statistics; Mathematics; Biology; Algorithm","score_opus":0.04616301159606384,"score_gpt":0.34571737856377327,"score_spread":0.29955436696770943,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2171380912","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.077399805,0.0013316827,0.8965586,0.0059857555,0.00034181462,0.00044266696,0.0092729,0.0012086876,0.007458172],"genre_scores_gemma":[0.761033,0.002602376,0.16400579,0.0007918226,0.00049785175,0.003011688,0.011557581,0.00033266743,0.05616718],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99651057,0.0015950279,0.00013504326,0.00090319855,0.00040230437,0.0004537825],"domain_scores_gemma":[0.9882957,0.008845361,0.0012407433,0.00048602276,0.0007951844,0.00033719666],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006380804,0.0017611319,0.0019962508,0.0023544603,0.00084082526,0.0026658017,0.0040014484,0.0036662142,0.010258812],"category_scores_gemma":[0.018012134,0.0010733989,0.0018001205,0.002325,0.0016800305,0.002011054,0.0017548184,0.0035576883,0.0036446983],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007235452,0.00026380163,0.031113738,0.00023244474,0.0006104517,0.001002278,0.0005552716,0.71799237,0.0015174919,0.1899969,0.010525839,0.04546587],"study_design_scores_gemma":[0.00026553465,0.00022216231,0.0061738365,0.00007245329,0.00015465902,0.00046853677,0.00008352471,0.92156965,0.00024375453,0.063881494,0.0067705684,0.00009374748],"about_ca_topic_score_codex":0.025731739,"about_ca_topic_score_gemma":0.012744023,"teacher_disagreement_score":0.025731739,"about_ca_system_score_codex":0.002728067,"about_ca_system_score_gemma":0.0031359522,"threshold_uncertainty_score":0.05116391},"labels":[],"label_agreement":null},{"id":"W2171999085","doi":"10.1002/sim.1336","title":"Improved up‐and‐down designs for phase I trials","year":2002,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":130,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"National Institute of Allergy and Infectious Diseases","keywords":"Estimator; Isotonic regression; Maximum tolerated dose; Sample size determination; Statistics; Computer science; Zero (linguistics); Sample (material); Mathematics; Clinical trial; Medicine","score_opus":0.7965506191140403,"score_gpt":0.6475064483633985,"score_spread":0.1490441707506418,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2171999085","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014424929,0.0009460131,0.97889405,0.0002514004,0.0002343696,0.003856111,0.0001712284,0.00034896264,0.00087302626],"genre_scores_gemma":[0.13176256,0.00070689246,0.85056156,0.00078470196,0.0002354948,0.014186878,0.0002720852,0.00011756804,0.0013723961],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8362265,0.14805037,0.0046348358,0.00493212,0.005388379,0.00076772645],"domain_scores_gemma":[0.80006874,0.16071709,0.011632095,0.020787532,0.0054041184,0.0013904368],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.111774474,0.0022509282,0.003697201,0.0019621227,0.0007390579,0.0015027792,0.0026981735,0.0026370124,0.005094571],"category_scores_gemma":[0.18937479,0.001316645,0.0029352324,0.0015152292,0.0021574404,0.002240918,0.0022745642,0.0042548054,0.0009701784],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.035528123,0.0015847642,0.00472821,0.0037977525,0.0031296746,0.00029638226,0.00089651113,0.20503959,0.009270903,0.20996158,0.0050127157,0.52075386],"study_design_scores_gemma":[0.022720216,0.024732232,0.0046258513,0.0007403603,0.0031575768,0.0003640879,0.00008553105,0.6281479,0.013725307,0.2798264,0.021389173,0.00048533594],"about_ca_topic_score_codex":0.00040132072,"about_ca_topic_score_gemma":0.00043912255,"teacher_disagreement_score":0.88822556,"about_ca_system_score_codex":0.0010912031,"about_ca_system_score_gemma":0.002171398,"threshold_uncertainty_score":0.591127},"labels":[],"label_agreement":null},{"id":"W2172095467","doi":"10.1002/sim.2663","title":"On estimating treatment effects under non‐compliance in randomized clinical trials: are intent‐to‐treat or instrumental variables analyses perfect solutions?","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":46,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"U.S. Food and Drug Administration; Ministère de l'Économie, de la Science et de l'Innovation - Québec","keywords":"Instrumental variable; Estimator; Protocol (science); Econometrics; Estimation; Randomized controlled trial; Average treatment effect; Statistics; Computer science; Medicine; Mathematics; Economics; Surgery; Alternative medicine","score_opus":0.5017714453619644,"score_gpt":0.591935176063639,"score_spread":0.0901637307016746,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2172095467","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015859243,0.0019172671,0.9756988,0.0034195024,0.00016197882,0.00064238417,0.00016213521,0.00019152505,0.001947094],"genre_scores_gemma":[0.505818,0.0029715346,0.48362595,0.0025536423,0.0004005723,0.0032237188,0.0003255128,0.00014375371,0.0009372762],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.57682467,0.3973459,0.007745134,0.005454746,0.011212034,0.0014174707],"domain_scores_gemma":[0.19561014,0.7655203,0.01688059,0.017087542,0.0043987427,0.00050273555],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.36633128,0.0017249624,0.004264637,0.0022652585,0.0009281942,0.0036303592,0.0030148332,0.004673433,0.003521246],"category_scores_gemma":[0.6860341,0.0013869896,0.0033373146,0.0027472654,0.0074878954,0.008798009,0.0037468267,0.0055041607,0.0005647525],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0026415628,0.00040868946,0.012253825,0.0026414078,0.003219709,0.0001744989,0.0013791532,0.11978162,0.0005239825,0.6504811,0.0026432246,0.20385137],"study_design_scores_gemma":[0.001196514,0.0012342306,0.003132682,0.0010067172,0.00078437047,0.00015766076,0.00024213539,0.19176745,0.0019353031,0.7949737,0.0034450286,0.00012422646],"about_ca_topic_score_codex":0.0008940801,"about_ca_topic_score_gemma":0.00058502646,"teacher_disagreement_score":0.6336687,"about_ca_system_score_codex":0.0019777787,"about_ca_system_score_gemma":0.007409425,"threshold_uncertainty_score":0.78142637},"labels":[],"label_agreement":null},{"id":"W2178694734","doi":"10.1002/sim.6803","title":"Bias due to composite reference standards in diagnostic accuracy studies","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":64,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre","funders":"Canadian Institutes of Health Research; Nederlandse Organisatie voor Wetenschappelijk Onderzoek","keywords":"Statistics; Computer science; Data mining; Mathematics","score_opus":0.7570256031050993,"score_gpt":0.6499421358043234,"score_spread":0.10708346730077589,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2178694734","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00538046,0.012067956,0.97142315,0.005064161,0.0010218306,0.00041439492,0.00030053715,0.0003294343,0.003998045],"genre_scores_gemma":[0.33534032,0.012647918,0.63683724,0.0051229065,0.0027832128,0.0032002002,0.0006947012,0.00040137974,0.0029720648],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5832873,0.34070992,0.018074963,0.019137885,0.037037827,0.0017521827],"domain_scores_gemma":[0.2912189,0.6240479,0.031494047,0.038528156,0.013965499,0.00074558327],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.36755636,0.0021456582,0.0044811126,0.0069867806,0.0012208406,0.0083218245,0.006108151,0.0051190816,0.0037851017],"category_scores_gemma":[0.65105593,0.0016932018,0.0042382334,0.008336945,0.011621968,0.008457511,0.007573344,0.007863456,0.0013458263],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00046857866,0.00005332281,0.011252496,0.0025125607,0.0018920604,0.0004167345,0.0012543218,0.034687012,0.0004907755,0.8114863,0.005549075,0.12993672],"study_design_scores_gemma":[0.00016009869,0.00026021202,0.0025539172,0.0011505118,0.00046679142,0.0005518842,0.00011867731,0.055989847,0.0012608707,0.9265573,0.010813682,0.00011621896],"about_ca_topic_score_codex":0.002830513,"about_ca_topic_score_gemma":0.0015885003,"teacher_disagreement_score":0.63244367,"about_ca_system_score_codex":0.0052106013,"about_ca_system_score_gemma":0.004673799,"threshold_uncertainty_score":0.77991563},"labels":[],"label_agreement":null},{"id":"W2180200726","doi":"10.1002/sim.6808","title":"2 × 2 Tables: a note on Campbell's recommendation","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Essar Steel Algoma (Canada); St. Joseph's Care Group; NOSM University; Lakehead University","funders":"","keywords":"Statistic; Pearson's chi-squared test; Statistics; Chi-square test; Pearson product-moment correlation coefficient; Mathematics; Mean squared error; Test statistic; Statistical hypothesis testing","score_opus":0.20562809733330492,"score_gpt":0.5035358811184455,"score_spread":0.2979077837851406,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2180200726","genre_codex":"commentary","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00049651234,0.009140072,0.16364096,0.48070997,0.27937248,0.0040939073,0.011041163,0.015347425,0.036157496],"genre_scores_gemma":[0.0046261246,0.005351248,0.41555572,0.48164573,0.038646895,0.008883215,0.002766312,0.006897604,0.035627212],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.73960924,0.12104594,0.043442212,0.01285905,0.079927795,0.0031157865],"domain_scores_gemma":[0.15980594,0.46891835,0.023782372,0.07552254,0.2628662,0.009104539],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.22052303,0.0023385389,0.0056643193,0.01304775,0.003944927,0.012812839,0.016979778,0.025384557,0.06533879],"category_scores_gemma":[0.71934015,0.0047316984,0.0075087645,0.018934863,0.011331748,0.013736157,0.0065877712,0.043412328,0.07101862],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000081409664,0.000057470035,0.00026084902,0.000572027,0.00006057467,0.00006408525,0.00019384173,0.00016166855,0.00009067434,0.009499536,0.9601923,0.028765535],"study_design_scores_gemma":[0.0001754457,0.00004632266,0.00076919724,0.0030433906,0.00007604146,0.00026122743,0.00014726787,0.0008543682,0.00030426827,0.025537081,0.9686309,0.00015454469],"about_ca_topic_score_codex":0.017769294,"about_ca_topic_score_gemma":0.021807607,"teacher_disagreement_score":0.779477,"about_ca_system_score_codex":0.0053596674,"about_ca_system_score_gemma":0.032209832,"threshold_uncertainty_score":0.96123385},"labels":[],"label_agreement":null},{"id":"W2211643105","doi":"10.1002/1097-0258(20000915/30)19:17/18<2437::aid-sim580>3.0.co;2-j","title":"Methodological issues in the development of theCanadian Cancer Incidence Atlas","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Colorectal Cancer Screening and Detection","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Saskatchewan Cancer Agency; Cancer Care Ontario; McMaster University; BC Cancer Agency; Health Canada","funders":"","keywords":"Poisson regression; Confidence interval; Statistics; Poisson distribution; Atlas (anatomy); Demography; Cancer registry; Population; Correlation; Census; Cartography; Geography; Medicine; Mathematics","score_opus":0.12879942501858482,"score_gpt":0.4343008291345362,"score_spread":0.3055014041159514,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2211643105","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07182824,0.010885193,0.6914282,0.032226775,0.011565821,0.046574246,0.03488882,0.0026659928,0.09793673],"genre_scores_gemma":[0.15046138,0.0040694713,0.74064064,0.0053611184,0.0012058853,0.06989968,0.0089497,0.0017878184,0.017624214],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.7557989,0.16763568,0.02654931,0.007375863,0.039301943,0.0033383123],"domain_scores_gemma":[0.6844426,0.11285642,0.012513391,0.043284714,0.14367937,0.0032235703],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27817667,0.0012246193,0.002578736,0.013487728,0.006131951,0.008280827,0.00755962,0.00113761,0.0067639025],"category_scores_gemma":[0.43293297,0.0018383687,0.0020017351,0.022725767,0.0038753415,0.0026261678,0.0069802497,0.0037378538,0.0016264353],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012050569,0.0002548206,0.067418836,0.007121831,0.0011826443,0.00080513925,0.031326238,0.018540146,0.0018704595,0.348475,0.15345332,0.36834645],"study_design_scores_gemma":[0.00023735607,0.00036085982,0.121580735,0.0051740403,0.00059005147,0.0004895348,0.01707942,0.015939757,0.0044619734,0.058003105,0.77563,0.00045322473],"about_ca_topic_score_codex":0.4007948,"about_ca_topic_score_gemma":0.54131037,"teacher_disagreement_score":0.9784277,"about_ca_system_score_codex":0.021572314,"about_ca_system_score_gemma":0.04428454,"threshold_uncertainty_score":0.89013666},"labels":[],"label_agreement":null},{"id":"W2226573656","doi":"10.1002/sim.6845","title":"A comparison of change point models with application to longitudinal lung function measurements in children with cystic fibrosis","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Cystic Fibrosis Research Advances","field":"Medicine","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"Cystic Fibrosis Foundation","keywords":"Cystic fibrosis; Random effects model; Lung function; Demography; Population; Lung disease; Time point; Medicine; Pulmonary function testing; Lung; Statistics; Mathematics; Internal medicine; Physics","score_opus":0.07383692652110245,"score_gpt":0.37621764581914946,"score_spread":0.302380719298047,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2226573656","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5556169,0.0047743246,0.42982504,0.0030939588,0.0003610659,0.000898594,0.0013790554,0.00093284075,0.0031182233],"genre_scores_gemma":[0.87692696,0.0015716734,0.11609247,0.00043090823,0.00017281098,0.00092184317,0.001845287,0.0002503763,0.0017877461],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9711151,0.025426488,0.00063916633,0.0014488163,0.00084117363,0.000529291],"domain_scores_gemma":[0.66030216,0.32247782,0.0058547016,0.0048765875,0.005067165,0.0014216468],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.083148554,0.0014798315,0.0027014986,0.0026166504,0.0012380936,0.002423224,0.0040506525,0.0030121885,0.0033461398],"category_scores_gemma":[0.17247337,0.000997568,0.0046760333,0.0023473653,0.0018214197,0.0031346553,0.002868926,0.004731231,0.0004639173],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0023114847,0.00040583452,0.04750191,0.00030918827,0.001338484,0.0002558735,0.00092333165,0.87645894,0.00018778416,0.029906973,0.0019724567,0.038427662],"study_design_scores_gemma":[0.000114102295,0.00036388947,0.0045515345,0.000050300885,0.00012509013,0.000058337802,0.00019436535,0.98348653,0.000072622104,0.010364154,0.0005720975,0.000047040765],"about_ca_topic_score_codex":0.032938797,"about_ca_topic_score_gemma":0.016485538,"teacher_disagreement_score":0.083148554,"about_ca_system_score_codex":0.0035685182,"about_ca_system_score_gemma":0.0031785136,"threshold_uncertainty_score":0.43973684},"labels":[],"label_agreement":null},{"id":"W2242307981","doi":"10.1002/sim.6818","title":"The missing cause approach to unmeasured confounding in pharmacoepidemiology","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; Bruyère; University of Ottawa; McGill University; Ottawa Public Health; McGill University Health Centre","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Confounding; Pharmacoepidemiology; Spurious relationship; Statistics; Observational study; Missing data; Econometrics; Medicine; Instrumental variable; Average treatment effect; Variance (accounting); Mathematics; Propensity score matching; Medical prescription","score_opus":0.2739779879782596,"score_gpt":0.5135117581148383,"score_spread":0.23953377013657867,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2242307981","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008158916,0.00061149226,0.9885725,0.0010937416,0.00016405922,0.00025358668,0.0003263622,0.00013188276,0.00068753655],"genre_scores_gemma":[0.5144419,0.0011539853,0.4770767,0.0019146041,0.0006648949,0.0020441995,0.00070630223,0.000111724046,0.00188566],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8986723,0.08281395,0.003617767,0.0063249627,0.007622997,0.0009480208],"domain_scores_gemma":[0.7318843,0.21736205,0.016251352,0.028284447,0.0053083817,0.0009095534],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08874108,0.0014461303,0.0030313737,0.0034481648,0.001473757,0.002759468,0.005448657,0.0030028366,0.0028176648],"category_scores_gemma":[0.27506724,0.0012442191,0.0043447763,0.003607826,0.0038899102,0.0028056325,0.004462811,0.0048369234,0.0002814917],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013970678,0.00034126325,0.10000933,0.0023365424,0.005385956,0.0015104335,0.0021843014,0.09902861,0.0015371697,0.57372344,0.0061689713,0.20637685],"study_design_scores_gemma":[0.0007535802,0.00062802836,0.015339574,0.0006006208,0.0017501423,0.00094388874,0.0002847465,0.2893139,0.0028778925,0.6757121,0.011585166,0.00021026727],"about_ca_topic_score_codex":0.004592464,"about_ca_topic_score_gemma":0.002673947,"teacher_disagreement_score":0.08874108,"about_ca_system_score_codex":0.0017136876,"about_ca_system_score_gemma":0.0041395547,"threshold_uncertainty_score":0.46931326},"labels":[],"label_agreement":null},{"id":"W2268470382","doi":"10.1002/sim.6812","title":"Methods to adjust for misclassification in the quantiles for the generalized linear model with measurement error in continuous exposures","year":2015,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical and numerical algorithms","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Institute of Environmental Health Sciences; National Institute of Mental Health; National Cancer Institute; National Heart, Lung, and Blood Institute","keywords":"Quantile; Statistics; Generalized linear model; Observational error; Linear model; Econometrics; Computer science; Mathematics","score_opus":0.37371390643757485,"score_gpt":0.4800672643962178,"score_spread":0.10635335795864292,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2268470382","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0014833339,0.00019019791,0.9974698,0.00019935137,0.000044765704,0.000100478675,0.00003287603,0.00030644573,0.00017284685],"genre_scores_gemma":[0.10491685,0.0008140087,0.8890956,0.00060423784,0.00022067464,0.0013433477,0.00032740668,0.00040500506,0.0022728774],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98495436,0.011655148,0.00053812144,0.0013678132,0.0012504113,0.00023421444],"domain_scores_gemma":[0.9446003,0.04224737,0.0050092516,0.0053001363,0.002536634,0.00030635588],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03949323,0.0014840927,0.002096671,0.0027549593,0.0007796335,0.0012499671,0.0033877061,0.0018983588,0.0047868886],"category_scores_gemma":[0.12446405,0.00096562516,0.0025172255,0.002299154,0.0018317552,0.0021436175,0.0034416807,0.004022352,0.0011359585],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00035860777,0.00027672766,0.01876353,0.00077190925,0.0014232615,0.00031977228,0.0008177099,0.1931752,0.002110161,0.16566189,0.007608974,0.6087122],"study_design_scores_gemma":[0.0002765084,0.00033341118,0.007818727,0.0002744042,0.00034606678,0.00046607025,0.0001795037,0.77070737,0.0024395827,0.2062118,0.010800178,0.00014631187],"about_ca_topic_score_codex":0.0041339523,"about_ca_topic_score_gemma":0.003211767,"teacher_disagreement_score":0.03949323,"about_ca_system_score_codex":0.0014596727,"about_ca_system_score_gemma":0.0025299366,"threshold_uncertainty_score":0.20886266},"labels":[],"label_agreement":null},{"id":"W2280398094","doi":"10.1002/sim.6904","title":"A latent process model for forecasting multiple time series in environmental public health surveillance","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Air Quality and Health Impacts","field":"Environmental Science","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"BC Centre for Disease Control; University of British Columbia; McGill University; McGill University Health Centre","funders":"Canadian Institutes of Health Research; Health Canada; British Columbia Centre for Disease Control; University of British Columbia","keywords":"Univariate; Bivariate analysis; Environmental epidemiology; Laplace's method; Computer science; Bayesian probability; Outcome (game theory); Econometrics; Statistics; Multivariate statistics; Environmental health; Machine learning; Mathematics; Medicine; Artificial intelligence","score_opus":0.09898888752182965,"score_gpt":0.34136953729052444,"score_spread":0.24238064976869478,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2280398094","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0141252335,0.00044632575,0.9827759,0.00071619963,0.00007002073,0.00005279175,0.0003997926,0.00025929735,0.0011543855],"genre_scores_gemma":[0.76324934,0.0020853877,0.22161135,0.00045100017,0.00040085329,0.0008441163,0.0022050568,0.00015872715,0.00899411],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9972485,0.0013766042,0.00015684128,0.0005954515,0.0003959249,0.00022681503],"domain_scores_gemma":[0.9882295,0.009487463,0.0010085651,0.00039689467,0.00068506756,0.00019249093],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008746388,0.0011652596,0.001762956,0.0019559332,0.0006516892,0.0022459123,0.0032845766,0.0029060263,0.003798618],"category_scores_gemma":[0.028504089,0.00093641615,0.0019513023,0.0027901677,0.0015497338,0.0034533232,0.0016300877,0.003395952,0.000717935],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000056892863,0.000047197696,0.00372169,0.00007605763,0.00008083423,0.00012423936,0.00017711894,0.88055915,0.00030386902,0.099503465,0.0009597852,0.014389695],"study_design_scores_gemma":[0.000010570848,0.000011643712,0.0002379379,0.000011718171,0.000012160323,0.000017138162,0.000012856055,0.97751564,0.00005340704,0.021668207,0.00043562707,0.0000131283105],"about_ca_topic_score_codex":0.02254449,"about_ca_topic_score_gemma":0.014842808,"teacher_disagreement_score":0.02254449,"about_ca_system_score_codex":0.0021526641,"about_ca_system_score_gemma":0.0015992989,"threshold_uncertainty_score":0.046255887},"labels":[],"label_agreement":null},{"id":"W2288666460","doi":"10.1002/sim.6907","title":"Inference on treatment‐covariate interaction based on a nonparametric measure of treatment effects and censored survival data","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"","keywords":"Covariate; Nonparametric statistics; Statistics; Inference; Parametric statistics; Econometrics; Data set; Confidence interval; Survival analysis; Estimator; Computer science; Mathematics; Artificial intelligence","score_opus":0.5481851531616271,"score_gpt":0.5778375376790365,"score_spread":0.02965238451740937,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2288666460","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.030028675,0.0012257108,0.9665139,0.0005559667,0.000096708165,0.0002615202,0.00028149813,0.00022753495,0.0008084837],"genre_scores_gemma":[0.69856554,0.0012994448,0.29378027,0.0009193185,0.000455675,0.00185693,0.0010770799,0.00014039874,0.0019053436],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.90211755,0.08462438,0.0023524894,0.0048029446,0.005100304,0.0010023435],"domain_scores_gemma":[0.51380515,0.452354,0.016697211,0.014198246,0.0018044943,0.0011409597],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09993556,0.0017357483,0.0054023787,0.003217131,0.00068226684,0.0030525548,0.0033439635,0.0036532746,0.003289341],"category_scores_gemma":[0.27669856,0.0009671689,0.0035085916,0.0029720936,0.004690472,0.0036196487,0.0033437936,0.0059375307,0.00042915592],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0071463184,0.0011623593,0.042328674,0.0024022479,0.0054575847,0.0014392053,0.0009834231,0.35916013,0.004119232,0.26868838,0.003937378,0.30317503],"study_design_scores_gemma":[0.0006032124,0.002240131,0.0137114255,0.00026735498,0.00091762457,0.00082239445,0.00010597812,0.6761618,0.002598719,0.29940253,0.00299853,0.00017036841],"about_ca_topic_score_codex":0.000725428,"about_ca_topic_score_gemma":0.00052866957,"teacher_disagreement_score":0.09993556,"about_ca_system_score_codex":0.0016167584,"about_ca_system_score_gemma":0.0024981021,"threshold_uncertainty_score":0.52851605},"labels":[],"label_agreement":null},{"id":"W2300752603","doi":"10.1002/sim.6909","title":"Type‐II generalized family‐wise error rate formulas with application to sample size determination","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Type I and type II errors; Sample size determination; False discovery rate; Null hypothesis; Statistical power; Word error rate; Statistics; Multiple comparisons problem; Statistical hypothesis testing; Computer science; Mathematics; Nominal level; Algorithm; Artificial intelligence; Confidence interval","score_opus":0.30340492714075973,"score_gpt":0.5395663056930371,"score_spread":0.23616137855227737,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2300752603","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00036166812,0.001040244,0.9965391,0.0003072385,0.00024084789,0.00048556246,0.00013553457,0.0003912062,0.0004986441],"genre_scores_gemma":[0.0149425715,0.0017933138,0.97249043,0.0008536871,0.0005048517,0.0065662195,0.000338828,0.0009516746,0.0015585023],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8644354,0.10561527,0.0069284136,0.004874371,0.017197933,0.00094866374],"domain_scores_gemma":[0.5008769,0.44503415,0.014758389,0.023834923,0.014643547,0.0008520612],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.16203114,0.0044390555,0.005201057,0.0061906106,0.0010430048,0.0032511435,0.007166119,0.0054445164,0.012688289],"category_scores_gemma":[0.4720957,0.0015709095,0.0043881116,0.0061734365,0.005677131,0.0052106604,0.0043658754,0.013243738,0.004260613],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008027045,0.00021456978,0.0051352168,0.00293195,0.0013274006,0.0008893517,0.0014928045,0.08406947,0.0020646893,0.4175005,0.045783192,0.43778813],"study_design_scores_gemma":[0.00052854884,0.00075996114,0.0022388133,0.0016592193,0.00060658995,0.0013297105,0.00019164316,0.40296644,0.005332352,0.5298415,0.054268435,0.00027676608],"about_ca_topic_score_codex":0.0023713952,"about_ca_topic_score_gemma":0.0016285853,"teacher_disagreement_score":0.8379688,"about_ca_system_score_codex":0.0030171315,"about_ca_system_score_gemma":0.004887142,"threshold_uncertainty_score":0.85691285},"labels":[],"label_agreement":null},{"id":"W2304400665","doi":"10.1002/sim.7278","title":"Estimating cross‐validatory predictive <i>p</i>‐values with integrated importance sampling for disease mapping models","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Saskatchewan Health Authority; University of Saskatchewan","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Posterior probability; Markov chain Monte Carlo; Markov chain; Sampling (signal processing); Bayesian probability; Cross-validation; Pattern recognition (psychology)","score_opus":0.18728641016175812,"score_gpt":0.46136737297218855,"score_spread":0.27408096281043043,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2304400665","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.045358054,0.0016641543,0.94998944,0.00037028798,0.00008534901,0.00018983167,0.00043765214,0.0007834079,0.001121813],"genre_scores_gemma":[0.6286863,0.00093245646,0.36437264,0.00053711055,0.00028377576,0.00048992,0.0030936257,0.00038493963,0.0012193149],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98290956,0.011223887,0.00071525463,0.0028470585,0.0019393981,0.00036485487],"domain_scores_gemma":[0.8294375,0.15133965,0.004993892,0.009793458,0.0035120621,0.0009234759],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.041573882,0.0022427826,0.0028990116,0.004799285,0.0014830747,0.0031771632,0.0036612274,0.0020867933,0.002563224],"category_scores_gemma":[0.15582791,0.00094482966,0.0026092653,0.004268141,0.0040057953,0.0035591691,0.0035570713,0.00469637,0.00055737194],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008833374,0.00046340856,0.13002661,0.0010850035,0.0031743653,0.0009959961,0.001094837,0.4187075,0.0031369696,0.067031,0.0076002576,0.36580062],"study_design_scores_gemma":[0.00010426369,0.00016571487,0.011112703,0.00016323161,0.0003582706,0.0003375934,0.00011735892,0.8809727,0.002002617,0.102068245,0.0025133924,0.00008388151],"about_ca_topic_score_codex":0.0069363057,"about_ca_topic_score_gemma":0.0060535613,"teacher_disagreement_score":0.041573882,"about_ca_system_score_codex":0.001423012,"about_ca_system_score_gemma":0.002700367,"threshold_uncertainty_score":0.21986634},"labels":[],"label_agreement":null},{"id":"W2343906356","doi":"10.1002/sim.6969","title":"Handling incomplete correlated continuous and binary outcomes in meta‐analysis of individual participant data","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; U.S. Food and Drug Administration; Hamilton Health Sciences Foundation","keywords":"Binary data; Meta-analysis; Computer science; Binary number; Statistics; Mathematics; Medicine; Internal medicine","score_opus":0.8426618781836268,"score_gpt":0.5676160556797044,"score_spread":0.2750458225039224,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2343906356","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01057417,0.2154103,0.7603713,0.0042023975,0.0014939762,0.0026984753,0.0029953918,0.0008853273,0.0013685924],"genre_scores_gemma":[0.32859644,0.06345804,0.58257335,0.0047117076,0.0013067396,0.0137533,0.003923933,0.00066743355,0.0010091092],"study_design_codex":"meta_analysis","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.48715487,0.4652075,0.024416232,0.0111265825,0.011244338,0.00085051084],"domain_scores_gemma":[0.41835168,0.52838343,0.02139694,0.0281212,0.0032439854,0.0005027685],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.34832382,0.0052096895,0.017787378,0.013698681,0.0015871827,0.009159093,0.0063655847,0.0066003124,0.003283064],"category_scores_gemma":[0.54021066,0.0037613607,0.041594524,0.01492139,0.0027777068,0.0068065836,0.0067821257,0.008000114,0.00054022676],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0030299416,0.00016538154,0.019110898,0.09437192,0.62204033,0.0008211592,0.0011764807,0.08268503,0.000843226,0.033803016,0.006445393,0.13550732],"study_design_scores_gemma":[0.0038951302,0.0018794833,0.011075744,0.031799693,0.48701537,0.001136851,0.00047807742,0.10748068,0.0031364853,0.32149717,0.029788513,0.00081682345],"about_ca_topic_score_codex":0.004259446,"about_ca_topic_score_gemma":0.005023037,"teacher_disagreement_score":0.6516762,"about_ca_system_score_codex":0.0041531837,"about_ca_system_score_gemma":0.0070521664,"threshold_uncertainty_score":0.80363274},"labels":[],"label_agreement":null},{"id":"W2345136754","doi":"10.1002/sim.6955","title":"Linear models of coregionalization for multivariate lattice data: a general framework for coregionalized multivariate CAR models","year":2016,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":39,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Multivariate statistics; Univariate; Autoregressive model; Covariance; Bayesian probability; Computer science; Multivariate analysis; Gaussian; Conditional independence; Multivariate normal distribution; Econometrics; Statistics; Mathematics; Artificial intelligence; Machine learning","score_opus":0.39298559087138896,"score_gpt":0.5257090736074782,"score_spread":0.13272348273608925,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2345136754","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0012895917,0.010101403,0.9838971,0.001234504,0.00009870349,0.000032225325,0.00024392498,0.0001399755,0.002962547],"genre_scores_gemma":[0.2427621,0.075343885,0.65983194,0.0023898801,0.002208867,0.0011372307,0.0020456843,0.00045298896,0.013827358],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99686,0.001819974,0.00015096627,0.00054831733,0.000497482,0.00012334547],"domain_scores_gemma":[0.989291,0.008208826,0.00075293117,0.0007088016,0.0009259962,0.00011247635],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008053275,0.0014886857,0.0020269926,0.002241831,0.0004372006,0.002767139,0.0032636686,0.0018833831,0.0037514765],"category_scores_gemma":[0.014683959,0.0010499336,0.0027387792,0.003152762,0.0031226687,0.0039920546,0.0021409222,0.003948721,0.001480186],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000016774346,0.000022713075,0.00087569555,0.0005383132,0.00018346614,0.00008447941,0.00017427719,0.08528703,0.0002306289,0.8252121,0.003872116,0.08350234],"study_design_scores_gemma":[0.000011568322,0.000035673744,0.00068916765,0.000299583,0.00007729994,0.00015338017,0.000047750178,0.2437448,0.00023503976,0.7277112,0.026940309,0.000054275486],"about_ca_topic_score_codex":0.006599281,"about_ca_topic_score_gemma":0.0039492706,"teacher_disagreement_score":0.008053275,"about_ca_system_score_codex":0.00247329,"about_ca_system_score_gemma":0.0022328014,"threshold_uncertainty_score":0.04259026},"labels":[],"label_agreement":null},{"id":"W2378268168","doi":"10.1002/sim.6979","title":"A flexible parametric approach for estimating continuous‐time inverse probability of treatment and censoring weights","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cancer Care Ontario; Public Health Ontario; University of Toronto","funders":"","keywords":"Inverse probability; Censoring (clinical trials); Marginal structural model; Inverse probability weighting; Parametric statistics; Statistics; Weighting; Event (particle physics); Computer science; Marginal distribution; Proportional hazards model; Confounding; Parametric model; Econometrics; Mathematics; Posterior probability; Medicine; Estimator; Random variable","score_opus":0.13772868312824896,"score_gpt":0.40989090075438367,"score_spread":0.2721622176261347,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2378268168","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.001207697,0.00006354825,0.9983053,0.00006471417,0.000013227569,0.000042398722,0.000051561532,0.00007233198,0.00017924103],"genre_scores_gemma":[0.2208025,0.0006319204,0.77255946,0.00029618983,0.00020211955,0.0014538016,0.0006732758,0.00021378581,0.0031669207],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9826099,0.012403051,0.00056551,0.0018910888,0.0019574936,0.00057291565],"domain_scores_gemma":[0.94067377,0.04522385,0.0035570175,0.008331578,0.0017398513,0.00047404965],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026620982,0.0015705214,0.0024545453,0.0030514302,0.0007352648,0.0025629338,0.0053960998,0.0028129807,0.005372311],"category_scores_gemma":[0.1031211,0.0016912814,0.0037633162,0.0038305868,0.0028752913,0.0030668352,0.004557059,0.0067307632,0.0008420428],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002727301,0.00021553338,0.007579504,0.00039459608,0.0009826559,0.00059786113,0.00078042527,0.32593244,0.0019346174,0.4392083,0.0030149126,0.21908635],"study_design_scores_gemma":[0.00006838751,0.00014215552,0.0019957304,0.000078146586,0.00015239095,0.00029219786,0.0000720383,0.7112534,0.0009311364,0.2805478,0.0043859514,0.00008053501],"about_ca_topic_score_codex":0.0034474863,"about_ca_topic_score_gemma":0.0029629485,"teacher_disagreement_score":0.026620982,"about_ca_system_score_codex":0.0015839443,"about_ca_system_score_gemma":0.0026729591,"threshold_uncertainty_score":0.14078689},"labels":[],"label_agreement":null},{"id":"W2400385987","doi":"10.1002/sim.6986","title":"Quantifying the impact of different approaches for handling continuous predictors on the performance of a prognostic model","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":177,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Population Health Research Institute","funders":"Medical Research Council; National Institute for Health and Care Research; Cancer Research UK","keywords":"Computer science; Outcome (game theory); Ignorance; Calibration; Statistics; Econometrics; Machine learning; Mathematics","score_opus":0.6090490742207253,"score_gpt":0.5329850486829449,"score_spread":0.07606402553778036,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2400385987","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.25960585,0.0139262965,0.7102767,0.006599986,0.000547694,0.0010931797,0.0017370143,0.0007261062,0.005487072],"genre_scores_gemma":[0.7334692,0.003137388,0.25878513,0.000932119,0.00024196159,0.00091480615,0.0016120969,0.000270162,0.0006371347],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8925464,0.086613305,0.0061342665,0.0051504,0.008396836,0.0011588169],"domain_scores_gemma":[0.47304648,0.4878442,0.01104025,0.020448953,0.0062713525,0.0013487393],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1710393,0.002825563,0.0032107257,0.0039718794,0.0012525384,0.0058260933,0.0023741245,0.0032551263,0.0017762055],"category_scores_gemma":[0.37294564,0.000867778,0.0044974126,0.0043319236,0.0040472453,0.005243704,0.004891267,0.0070146048,0.00066624535],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0053660558,0.0005530912,0.18642066,0.0024638595,0.009841758,0.0005390573,0.001757809,0.47370234,0.0032530513,0.02829788,0.003941333,0.28386316],"study_design_scores_gemma":[0.00067817676,0.0048296615,0.06508611,0.002078689,0.003133206,0.0012235703,0.001187027,0.75592667,0.007873443,0.15055986,0.0067396103,0.00068405137],"about_ca_topic_score_codex":0.0027455583,"about_ca_topic_score_gemma":0.0027414502,"teacher_disagreement_score":0.1710393,"about_ca_system_score_codex":0.0018834476,"about_ca_system_score_gemma":0.0028904974,"threshold_uncertainty_score":0.90455306},"labels":[],"label_agreement":null},{"id":"W2401995712","doi":"10.1002/sim.6998","title":"Estimating the cumulative mean function for history process with time‐dependent covariates and censoring mechanism","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Regina","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Censoring (clinical trials); Covariate; Estimator; Inverse probability weighting; Computer science; Inverse probability; Weighting; Statistics; Consistency (knowledge bases); Cumulative distribution function; Econometrics; Mathematics; Probability density function; Artificial intelligence; Medicine; Posterior probability","score_opus":0.07753858154408844,"score_gpt":0.37604131994293466,"score_spread":0.2985027383988462,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2401995712","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0054287324,0.00017742293,0.9940726,0.00007499439,0.00001302177,0.000021591937,0.000029312203,0.000061397055,0.000121035664],"genre_scores_gemma":[0.3911597,0.002062042,0.60090774,0.0002538838,0.0002634997,0.0006489619,0.00087925384,0.00013675414,0.0036881967],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9970386,0.0015662201,0.00014569491,0.0006105929,0.00044225279,0.0001966179],"domain_scores_gemma":[0.97933334,0.016817352,0.0015142829,0.0013040652,0.0008525302,0.00017841693],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012585169,0.0010580891,0.0020814894,0.002565097,0.0005332375,0.0013998956,0.0025168208,0.0019259773,0.0019465411],"category_scores_gemma":[0.03542356,0.000609922,0.0017303054,0.0018986173,0.0013259989,0.0030099843,0.0015867453,0.002454069,0.00039540028],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00020324757,0.00017713271,0.024451466,0.00054291694,0.00061931554,0.0004975825,0.0004494752,0.49821556,0.0035313342,0.2761758,0.002311716,0.19282448],"study_design_scores_gemma":[0.000032689233,0.00010181379,0.0032425288,0.00006812839,0.00011171003,0.00019970992,0.000053902986,0.88794947,0.0013428659,0.10513279,0.001701808,0.000062442035],"about_ca_topic_score_codex":0.0026225732,"about_ca_topic_score_gemma":0.0016275389,"teacher_disagreement_score":0.012585169,"about_ca_system_score_codex":0.00087087427,"about_ca_system_score_gemma":0.0019398697,"threshold_uncertainty_score":0.06655753},"labels":[],"label_agreement":null},{"id":"W2407438980","doi":"10.1002/sim.6994","title":"Developing points‐based risk‐scoring systems in the presence of competing risks","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":122,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University Health Network; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Event (particle physics); Computer science; Risk assessment; Set (abstract data type); Medicine; Intensive care medicine; Risk analysis (engineering)","score_opus":0.4536856419470451,"score_gpt":0.48069930320661425,"score_spread":0.027013661259569166,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2407438980","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.013177479,0.00091104314,0.9803432,0.0012093488,0.00015321221,0.00085088314,0.0006903517,0.00085478055,0.0018096541],"genre_scores_gemma":[0.16344343,0.0009154206,0.8306636,0.0003298243,0.00026079387,0.0015480146,0.0015118469,0.00022589702,0.0011010177],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9649872,0.024103353,0.004528297,0.0016743154,0.0041719135,0.0005349851],"domain_scores_gemma":[0.8409515,0.124947056,0.012918411,0.0064960048,0.0132848825,0.0014021485],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.055444077,0.0018599709,0.0021754773,0.005077873,0.0007955741,0.004554222,0.0032471393,0.0018909745,0.0058929343],"category_scores_gemma":[0.19610715,0.0011347656,0.002610725,0.0037735775,0.0013063562,0.004039643,0.0046535353,0.004444749,0.0021671578],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00086876954,0.00034327566,0.08049824,0.0014198939,0.001923872,0.0011553474,0.0012962277,0.23838751,0.0013410326,0.099651314,0.028869025,0.54424554],"study_design_scores_gemma":[0.00018486404,0.00043533862,0.013580568,0.00056991325,0.00037318928,0.0007951426,0.00031460368,0.7407589,0.0014853742,0.2275821,0.01364929,0.0002707284],"about_ca_topic_score_codex":0.0034196475,"about_ca_topic_score_gemma":0.0028759474,"teacher_disagreement_score":0.055444077,"about_ca_system_score_codex":0.0014387298,"about_ca_system_score_gemma":0.0024427753,"threshold_uncertainty_score":0.2932198},"labels":[],"label_agreement":null},{"id":"W2410992638","doi":"10.1002/sim.6996","title":"Bayesian adjustment for the misclassification in both dependent and independent variables with application to a breast cancer study","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; University of Saskatchewan","funders":"Natural Sciences and Engineering Research Council of Canada; American Cancer Society","keywords":"Concordance; Covariate; Logistic regression; Breast cancer; Statistics; Medicine; Econometrics; Internal medicine; Oncology; Cancer; Mathematics","score_opus":0.18434440924207227,"score_gpt":0.501049341478635,"score_spread":0.3167049322365627,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2410992638","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0036152338,0.0011337294,0.9931249,0.0008982344,0.0002083805,0.00029529296,0.00007796784,0.0002066538,0.00043958848],"genre_scores_gemma":[0.11078091,0.0011345119,0.88229626,0.0012235008,0.0006283693,0.0019049937,0.00031062687,0.00019714954,0.0015236998],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8396282,0.1400161,0.0051497133,0.0071847886,0.007059355,0.00096185226],"domain_scores_gemma":[0.7150938,0.23431914,0.016374715,0.02215192,0.010163133,0.0018973737],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.17651467,0.0019181424,0.0040583815,0.0037768448,0.0019052658,0.003067729,0.006519815,0.004077201,0.0034397102],"category_scores_gemma":[0.33611846,0.0018722926,0.0048580607,0.004044221,0.002698149,0.0026863944,0.006096594,0.0064330595,0.0006626504],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0021202536,0.0004183072,0.05037316,0.0021249894,0.009185577,0.0010972873,0.0023122574,0.16660762,0.0017777596,0.20700112,0.012340587,0.5446411],"study_design_scores_gemma":[0.00085201475,0.00073631667,0.01731691,0.0011138468,0.0021203423,0.00090129965,0.00023464777,0.62652016,0.0012877277,0.32345214,0.025200207,0.00026444526],"about_ca_topic_score_codex":0.009059257,"about_ca_topic_score_gemma":0.01049574,"teacher_disagreement_score":0.17651467,"about_ca_system_score_codex":0.0022283727,"about_ca_system_score_gemma":0.0050523076,"threshold_uncertainty_score":0.93350995},"labels":[],"label_agreement":null},{"id":"W2421687783","doi":"10.1002/sim.7606","title":"A threshold‐free summary index of prediction accuracy for censored time to event data","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University; University of Alberta","funders":"National Cancer Institute; Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Estimator; Event (particle physics); Cutoff; Nonparametric statistics; Inference; Index (typography); Risk assessment; Measure (data warehouse); Ordinal regression","score_opus":0.10933726131374259,"score_gpt":0.44081132255934224,"score_spread":0.33147406124559964,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2421687783","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.16161802,0.0027257209,0.8257052,0.001349865,0.00034075553,0.00033173282,0.0034699617,0.0007100956,0.0037487596],"genre_scores_gemma":[0.84000874,0.0008461848,0.15307055,0.00031948127,0.00027719,0.00051844295,0.00382252,0.000096616735,0.0010402347],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98268616,0.008582137,0.0026050063,0.0018567872,0.0038341794,0.00043573836],"domain_scores_gemma":[0.76357687,0.19377315,0.015438699,0.016017715,0.010111524,0.0010820003],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03488912,0.0008350206,0.0015754093,0.005577907,0.000468911,0.002435356,0.0016340875,0.0018879387,0.002010526],"category_scores_gemma":[0.22133638,0.00026384613,0.001751783,0.0039057042,0.001144681,0.0035761236,0.0012551867,0.0017049165,0.0005213957],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001323851,0.0003489844,0.3158287,0.0011183332,0.0031649189,0.00036715344,0.0007291283,0.17435874,0.002662549,0.051261343,0.008800763,0.44003558],"study_design_scores_gemma":[0.00015104271,0.0027345766,0.18817082,0.00051933265,0.0011290434,0.0014726119,0.00044915595,0.70548874,0.0048382855,0.08767493,0.0070412485,0.0003303076],"about_ca_topic_score_codex":0.0011333077,"about_ca_topic_score_gemma":0.0010571616,"teacher_disagreement_score":0.03488912,"about_ca_system_score_codex":0.0011021649,"about_ca_system_score_gemma":0.0010281692,"threshold_uncertainty_score":0.18451351},"labels":[],"label_agreement":null},{"id":"W2466391332","doi":"10.1002/sim.7012","title":"Martingale residual‐based method to control for confounders measured only in a validation sample in time‐to‐event analysis","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Statistics; Residual; Confounding; Computer science; Econometrics; Control chart; Event (particle physics); Sample (material); Sample size determination; Martingale (probability theory); Mathematics; Algorithm; Chromatography","score_opus":0.1189508705834617,"score_gpt":0.47092695879781293,"score_spread":0.3519760882143512,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2466391332","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0015062285,0.00014102011,0.997743,0.00012092024,0.000035430832,0.000056833844,0.00007391022,0.0001459414,0.00017683516],"genre_scores_gemma":[0.1412878,0.0005127842,0.85335857,0.00041596647,0.00020014078,0.0010632874,0.00080981496,0.0002512819,0.002100332],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9863914,0.010984069,0.00045247687,0.00097574445,0.0009756394,0.00022069612],"domain_scores_gemma":[0.92596084,0.06026468,0.003325701,0.007800108,0.0023245318,0.00032412572],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04667363,0.0012612771,0.0016357572,0.0025470427,0.00048550233,0.0010295876,0.003387843,0.0015826554,0.0038115182],"category_scores_gemma":[0.10791395,0.0006414644,0.0021480261,0.0019913048,0.0019214482,0.0017376421,0.0023639088,0.0033467633,0.00074389036],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00052324025,0.00021313455,0.01677709,0.0008168312,0.0014774281,0.00038114053,0.000557814,0.22081551,0.0024202531,0.47245747,0.007922661,0.2756375],"study_design_scores_gemma":[0.0002760594,0.00028487557,0.003454768,0.00017307265,0.0002441891,0.0002508487,0.00004247233,0.75625074,0.0023441631,0.22576381,0.010843721,0.00007124114],"about_ca_topic_score_codex":0.00308412,"about_ca_topic_score_gemma":0.0024588772,"teacher_disagreement_score":0.04667363,"about_ca_system_score_codex":0.00079953676,"about_ca_system_score_gemma":0.0027747445,"threshold_uncertainty_score":0.24683666},"labels":[],"label_agreement":null},{"id":"W2471145407","doi":"10.1002/sim.7011","title":"Statistical issues related to dietary intake as the response variable in intervention trials","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Nutritional Studies and Diet","field":"Medicine","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"National Cancer Institute; National Heart, Lung, and Blood Institute; Medical Research Council; National Institutes of Health","keywords":"Reliability (semiconductor); Biomarker; Context (archaeology); Calibration; Observational error; Statistics; Computer science; Intervention (counseling); Clinical trial; Medicine; Mathematics; Internal medicine; Biology","score_opus":0.05277491442254351,"score_gpt":0.41833186008813683,"score_spread":0.3655569456655933,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2471145407","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0035215819,0.02288473,0.9292612,0.025408728,0.007167035,0.005169074,0.000829028,0.0009882565,0.0047703516],"genre_scores_gemma":[0.19260103,0.010008483,0.7100677,0.034908712,0.00800218,0.03835521,0.0008703722,0.0009931156,0.00419333],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.09230244,0.85270005,0.023221334,0.010798089,0.019946402,0.0010317502],"domain_scores_gemma":[0.052844096,0.88463825,0.02127378,0.034675352,0.0059274463,0.0006410846],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.69588387,0.0038852606,0.01028316,0.0066539375,0.0030746593,0.009394005,0.0097036315,0.013670171,0.006036328],"category_scores_gemma":[0.89517885,0.003018209,0.010846423,0.013662972,0.023965709,0.009879567,0.008553137,0.020610476,0.0016978525],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0040896754,0.000334572,0.01894968,0.016224464,0.015580507,0.0015152949,0.0042389827,0.04263328,0.00075697503,0.56389135,0.049287744,0.28249747],"study_design_scores_gemma":[0.0014294444,0.001955379,0.0058454806,0.0063878526,0.0024473951,0.0010553531,0.0005604978,0.10275841,0.0016967905,0.8299318,0.045632217,0.00029940574],"about_ca_topic_score_codex":0.0038777662,"about_ca_topic_score_gemma":0.0024421692,"teacher_disagreement_score":0.30411613,"about_ca_system_score_codex":0.009101218,"about_ca_system_score_gemma":0.010377384,"threshold_uncertainty_score":0.37502939},"labels":[],"label_agreement":null},{"id":"W2516585541","doi":"10.1002/sim.3925","title":"Comments on ‘Some methodological issues in biosurveillance’","year":2011,"lang":"en","type":"letter","venue":"Statistics in Medicine","topic":"Data-Driven Disease Surveillance","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Canada Research Chairs","keywords":"Biostatistics; Library science; Citation; Computer science; Epidemiology; Medicine","score_opus":0.14166287861213966,"score_gpt":0.42258403642489334,"score_spread":0.2809211578127537,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2516585541","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0001079537,0.00025414152,0.00009112347,0.9910851,0.007913481,0.000008482038,0.000039307193,0.000013338075,0.0004869896],"genre_scores_gemma":[0.00048884336,0.00008718511,0.0001464456,0.9881207,0.0102709085,0.000020129635,0.000010212631,0.000012004192,0.000843556],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9434904,0.022115732,0.010277221,0.0062973113,0.011562067,0.0062572146],"domain_scores_gemma":[0.7864398,0.1550493,0.014670648,0.004346675,0.026442945,0.013050756],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05128352,0.0014979244,0.0030391845,0.002074778,0.008154189,0.013330203,0.0060085543,0.15547656,0.007919998],"category_scores_gemma":[0.2381365,0.0030041616,0.0034246761,0.0030934908,0.01006583,0.008776267,0.0055046286,0.10308329,0.007318242],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000044118027,0.000013598351,0.00062457396,0.00007181634,0.000025790716,0.0005478294,0.00033945745,0.00007255245,0.000082218656,0.0021084559,0.99269557,0.0033738785],"study_design_scores_gemma":[0.00018704528,0.000081583916,0.0036392547,0.0011941798,0.00011223878,0.0013091356,0.0013809674,0.00086309353,0.00030564412,0.01096874,0.97969645,0.00026163555],"about_ca_topic_score_codex":0.026808193,"about_ca_topic_score_gemma":0.041274093,"teacher_disagreement_score":0.94871646,"about_ca_system_score_codex":0.011315749,"about_ca_system_score_gemma":0.013785385,"threshold_uncertainty_score":0.27121645},"labels":[],"label_agreement":null},{"id":"W2516839318","doi":"10.1002/sim.7079","title":"Estimation for zero-inflated over-dispersed count data model with missing response","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Windsor","funders":"","keywords":"Count data; Negative binomial distribution; Missing data; Poisson distribution; Zero-inflated model; Expectation–maximization algorithm; Statistics; Estimator; Maximum likelihood; Mathematics; Zero (linguistics); Overdispersion; Mixture model; Censoring (clinical trials); Poisson regression; Applied mathematics; Econometrics; Population","score_opus":0.1297925260737236,"score_gpt":0.4484117459185901,"score_spread":0.3186192198448665,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2516839318","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01090461,0.00023366768,0.9883609,0.00010949226,0.000019463685,0.000038781465,0.00008595367,0.0000761341,0.00017098413],"genre_scores_gemma":[0.28454894,0.0010301464,0.7094812,0.00035346468,0.00020446006,0.00048413075,0.001469418,0.0001326173,0.0022956668],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9864564,0.008505342,0.00075370073,0.0027234838,0.0012338314,0.00032728232],"domain_scores_gemma":[0.9505313,0.038168658,0.0045069885,0.004538804,0.001886643,0.00036762768],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02467386,0.0011485312,0.002913336,0.0018791854,0.0006729885,0.0018523935,0.0051715714,0.0023842733,0.0030099542],"category_scores_gemma":[0.075900644,0.0011121384,0.0017909254,0.002665121,0.0020018513,0.003596988,0.0028159919,0.0033666366,0.0008581369],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005307899,0.00024101044,0.03650434,0.0016571262,0.0010264324,0.002766629,0.002275781,0.38298514,0.004788875,0.34410346,0.003364151,0.21975625],"study_design_scores_gemma":[0.000072585965,0.00017864909,0.004176084,0.00017648125,0.00016776401,0.00093638327,0.00028275614,0.8143304,0.001695183,0.17414823,0.0037418578,0.00009354952],"about_ca_topic_score_codex":0.0017133084,"about_ca_topic_score_gemma":0.0011316291,"teacher_disagreement_score":0.02467386,"about_ca_system_score_codex":0.0008492529,"about_ca_system_score_gemma":0.0010838603,"threshold_uncertainty_score":0.13048941},"labels":[],"label_agreement":null},{"id":"W2517825627","doi":"10.1002/sim.7074","title":"Trivariate mover‐stayer counting process models for investigating joint damage in psoriatic arthritis","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Spondyloarthritis Studies and Treatments","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Connaught Fund; University of Toronto","keywords":"Psoriatic arthritis; Covariate; Econometrics; Computer science; Joint (building); Feature (linguistics); Event (particle physics); Statistics; Population; Mathematics; Arthritis; Demography; Medicine; Engineering","score_opus":0.044325215750139216,"score_gpt":0.3280716948314211,"score_spread":0.2837464790812819,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2517825627","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06329437,0.0024128386,0.9284788,0.0013638455,0.0002923041,0.00025754477,0.0012725883,0.00045929616,0.0021683583],"genre_scores_gemma":[0.82246745,0.0038762337,0.13868138,0.00062607316,0.00054251845,0.0014681031,0.0030837436,0.0001782384,0.02907622],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9953452,0.0027853749,0.00019704767,0.0009546645,0.00035418774,0.0003635231],"domain_scores_gemma":[0.9822973,0.013628069,0.0021707288,0.00091232563,0.0005688755,0.00042264757],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011986144,0.0017916706,0.0030395845,0.0022328736,0.00069077616,0.0021803416,0.0047438354,0.004239741,0.0069980607],"category_scores_gemma":[0.023554495,0.0009774403,0.0030794484,0.0024546573,0.0022141563,0.0023827483,0.0025758774,0.004646492,0.0016553821],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005060008,0.00023348682,0.017644834,0.0003591575,0.00083313097,0.00069476926,0.0005912038,0.7542826,0.0010669904,0.18361947,0.0031134381,0.03705498],"study_design_scores_gemma":[0.00004935523,0.00018021914,0.001870084,0.000041217078,0.00014037435,0.00011246473,0.00007547487,0.94380325,0.0001777506,0.05159554,0.0019059064,0.000048478174],"about_ca_topic_score_codex":0.009957658,"about_ca_topic_score_gemma":0.006854686,"teacher_disagreement_score":0.011986144,"about_ca_system_score_codex":0.0015785649,"about_ca_system_score_gemma":0.0014036287,"threshold_uncertainty_score":0.0633896},"labels":[],"label_agreement":null},{"id":"W2527117517","doi":"10.1002/sim.7137","title":"Choosing appropriate analysis methods for cluster randomised cross‐over trials with a binary outcome","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":57,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"London Health Sciences Centre; Western University","funders":"National Institute for Health and Care Research; National Institute on Handicapped Research","keywords":"Statistics; Type I and type II errors; Cluster (spacecraft); Correlation; Cluster analysis; Outcome (game theory); Hierarchical clustering; Random effects model; Binary number; Cluster randomised controlled trial; Regression; Mathematics; Computer science; Meta-analysis; Medicine; Randomized controlled trial","score_opus":0.5289356418669582,"score_gpt":0.6556294378967649,"score_spread":0.12669379602980668,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2527117517","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0018669625,0.0020250326,0.97891104,0.0011043652,0.0007473104,0.013834167,0.00026442466,0.0006769052,0.0005697483],"genre_scores_gemma":[0.026028624,0.0008119258,0.90763533,0.0007661883,0.00018823899,0.06365851,0.00016373588,0.00025937767,0.0004880965],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.38433394,0.5727393,0.018121162,0.01034199,0.013237865,0.0012256603],"domain_scores_gemma":[0.33870426,0.58441705,0.031976473,0.029592551,0.014174054,0.0011356028],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.44531882,0.0046688486,0.008220669,0.0073081073,0.0011322869,0.005404534,0.006203635,0.0073891203,0.009227587],"category_scores_gemma":[0.66078943,0.0028461788,0.0098217325,0.0055617746,0.003870718,0.005701883,0.0045600687,0.0097631,0.002342563],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.015973281,0.0011036077,0.007916351,0.030621085,0.02181519,0.00074765465,0.003412574,0.12480299,0.0029045744,0.2348842,0.028010005,0.5278084],"study_design_scores_gemma":[0.014299968,0.0044609867,0.0042408435,0.009824464,0.0055231885,0.00042617274,0.0005861985,0.5057977,0.0051518218,0.42065004,0.028302465,0.0007361354],"about_ca_topic_score_codex":0.0013311913,"about_ca_topic_score_gemma":0.001278503,"teacher_disagreement_score":0.5546812,"about_ca_system_score_codex":0.0048497505,"about_ca_system_score_gemma":0.005278251,"threshold_uncertainty_score":0.6840207},"labels":[],"label_agreement":null},{"id":"W2531230258","doi":"10.1002/sim.7148","title":"Understanding the effects of conditional dependence in research studies involving imperfect diagnostic tests","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Covariate; Econometrics; Covariance; Statistics; Mathematics; Conditional variance; Autoregressive conditional heteroskedasticity","score_opus":0.8096974943345971,"score_gpt":0.6540817036197748,"score_spread":0.15561579071482234,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2531230258","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.106215194,0.0031984183,0.87936264,0.0050950176,0.00018124921,0.0005567221,0.000858732,0.00043277215,0.004099173],"genre_scores_gemma":[0.8414842,0.0012744458,0.15179294,0.0020411336,0.0003040716,0.00091164565,0.000622002,0.00010280937,0.0014667382],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.87330097,0.10005733,0.0071991105,0.011527369,0.0059815203,0.0019336566],"domain_scores_gemma":[0.29733083,0.6118535,0.046726838,0.037943583,0.004849052,0.0012961957],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.20199272,0.0012759531,0.002430981,0.0024136363,0.0014358487,0.0037019008,0.0030463848,0.004033035,0.0035212073],"category_scores_gemma":[0.46585417,0.0015506868,0.004162167,0.0036708154,0.009171274,0.006142588,0.0053992276,0.005527957,0.00032557268],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019747221,0.0002685512,0.26371947,0.002072754,0.005372341,0.0020924653,0.0041386015,0.14280754,0.0033007455,0.40991685,0.0037797873,0.16055615],"study_design_scores_gemma":[0.00031481354,0.0008421301,0.0921562,0.00088081637,0.002306224,0.0014236734,0.00035802354,0.21952245,0.0040372065,0.6708415,0.007041127,0.00027580798],"about_ca_topic_score_codex":0.005613469,"about_ca_topic_score_gemma":0.005984348,"teacher_disagreement_score":0.79800725,"about_ca_system_score_codex":0.002844991,"about_ca_system_score_gemma":0.004228454,"threshold_uncertainty_score":0.984085},"labels":[],"label_agreement":null},{"id":"W2545262557","doi":"10.1002/sim.7153","title":"Goodness‐of‐fit test for monotone proportional subdistribution hazards assumptions based on weighted residuals","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Institute of Cancer Research; Institut National Du Cancer","keywords":"Goodness of fit; Mathematics; Monotone polygon; Statistics; Regression analysis; Monte Carlo method; Regression; Proportional hazards model; Applied mathematics","score_opus":0.11451533146660954,"score_gpt":0.4452264803293031,"score_spread":0.3307111488626936,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2545262557","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.18277983,0.0005953649,0.8121374,0.0005545316,0.00011325889,0.0003084909,0.0005594098,0.00059570087,0.0023559604],"genre_scores_gemma":[0.90882444,0.00017213509,0.08825973,0.0002507268,0.0000681432,0.00054530083,0.00090219814,0.00017779428,0.0007995515],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96654886,0.024328828,0.0016500893,0.0033931558,0.0032587205,0.0008203543],"domain_scores_gemma":[0.6948715,0.27518415,0.009740038,0.012601253,0.0062041925,0.0013989552],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.048477117,0.0013038849,0.0026765983,0.004267959,0.001001037,0.0017481644,0.0034207138,0.002430604,0.005930333],"category_scores_gemma":[0.25502002,0.00061632216,0.00356921,0.003149389,0.0058062053,0.004296939,0.0028220022,0.003312999,0.0008051497],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004513015,0.00062911597,0.21990456,0.0012622786,0.0047791908,0.001920092,0.0019109289,0.24043263,0.005027214,0.27688566,0.004238138,0.23849724],"study_design_scores_gemma":[0.00063969364,0.0028115262,0.055470027,0.00019471392,0.00076374796,0.0014441297,0.00076019665,0.68207645,0.0051496257,0.24547048,0.00496633,0.00025302087],"about_ca_topic_score_codex":0.0015687055,"about_ca_topic_score_gemma":0.0008259947,"teacher_disagreement_score":0.048477117,"about_ca_system_score_codex":0.0010644352,"about_ca_system_score_gemma":0.0021272686,"threshold_uncertainty_score":0.25637454},"labels":[],"label_agreement":null},{"id":"W2558645513","doi":"10.1002/sim.7189","title":"Estimation of state occupancy probabilities in multistate models with dependent intermittent observation, with application to HIV viral rebounds","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Waterloo; Lunenfeld-Tanenbaum Research Institute","funders":"Canadian Institutes of Health Research","keywords":"Smoothing; Context (archaeology); Estimation; Occupancy; Nonparametric statistics; Statistics; Observational study; Econometrics; Inverse probability; Cohort; Human immunodeficiency virus (HIV); Computer science; Medicine; Mathematics; Biology; Virology; Bayesian probability; Economics","score_opus":0.04233541563109173,"score_gpt":0.35590498400765236,"score_spread":0.31356956837656064,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2558645513","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.026104072,0.00040094825,0.97236633,0.00038348322,0.00003098115,0.00006172247,0.0001403516,0.00014090643,0.0003711997],"genre_scores_gemma":[0.62802786,0.0016077087,0.36346558,0.00021203021,0.00021872325,0.00055933185,0.0008980533,0.00017672683,0.0048340214],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99610126,0.0027323756,0.00014473032,0.0005467298,0.00028709674,0.00018777874],"domain_scores_gemma":[0.9570879,0.037771445,0.0023154037,0.0014680132,0.00096510624,0.0003921725],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01599074,0.00095763104,0.0019249904,0.0013633241,0.0009926902,0.0016958937,0.003249154,0.0018856819,0.0017092575],"category_scores_gemma":[0.058534585,0.0011895286,0.0019567832,0.0018151565,0.0023981323,0.0021950745,0.0027955782,0.0031177416,0.00023323501],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016011423,0.000104815175,0.013263705,0.00017345214,0.00028868672,0.00022885289,0.000444381,0.83156145,0.00037314938,0.10931454,0.0012850463,0.042801842],"study_design_scores_gemma":[0.000015419138,0.000018153898,0.0009350148,0.000016953953,0.000024645658,0.000023403709,0.000037492475,0.9576242,0.00008171768,0.04073598,0.00046944796,0.000017492113],"about_ca_topic_score_codex":0.040730514,"about_ca_topic_score_gemma":0.03427812,"teacher_disagreement_score":0.040730514,"about_ca_system_score_codex":0.0016144948,"about_ca_system_score_gemma":0.0025825766,"threshold_uncertainty_score":0.08456808},"labels":[],"label_agreement":null},{"id":"W2559470641","doi":"10.1002/sim.7188","title":"The median hazard ratio: a useful measure of variance and general contextual effects in multilevel survival analysis","year":2016,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health disparities and outcomes","field":"Social Sciences","cited_by":103,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto; Institute for Work & Health; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Medicinska Forskningsrådet; Ontario Ministry of Health and Long-Term Care","keywords":"Multilevel model; Statistics; Proportional hazards model; Hazard ratio; Random effects model; Econometrics; Regression analysis; Cluster (spacecraft); Hierarchical database model; Marginal model; Hazard; Demography; Medicine; Mathematics; Computer science; Confidence interval; Data mining; Internal medicine; Meta-analysis","score_opus":0.032975172927863575,"score_gpt":0.36317528383708475,"score_spread":0.3302001109092212,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2559470641","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.020608678,0.008023457,0.9486946,0.0029502055,0.00056702393,0.00036674118,0.0106599955,0.0017239763,0.006405355],"genre_scores_gemma":[0.4545846,0.005645135,0.51749945,0.0018859182,0.0014691412,0.0021083036,0.009002624,0.0024457555,0.0053590285],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9858795,0.008920078,0.0008684732,0.0020242564,0.0020034662,0.00030419248],"domain_scores_gemma":[0.9433844,0.044494353,0.0046975934,0.004504112,0.0023996641,0.0005198955],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020797264,0.0009803426,0.0019089753,0.0037766832,0.0006078937,0.0019018321,0.0022095912,0.0013386492,0.009225544],"category_scores_gemma":[0.111319825,0.00044875042,0.0024288914,0.004996499,0.001514689,0.0023671372,0.0024367315,0.0032583135,0.0014610195],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00065371493,0.0001899177,0.14006528,0.0035759069,0.00453655,0.0011820429,0.0027663868,0.051234223,0.0018611021,0.19704798,0.10066018,0.49622676],"study_design_scores_gemma":[0.00016968437,0.0007735821,0.11991982,0.0019517218,0.0014377314,0.0022398965,0.0012345335,0.10660986,0.0028513926,0.6187303,0.14365467,0.00042676667],"about_ca_topic_score_codex":0.0068043633,"about_ca_topic_score_gemma":0.0054367348,"teacher_disagreement_score":0.020797264,"about_ca_system_score_codex":0.0013641234,"about_ca_system_score_gemma":0.0019075912,"threshold_uncertainty_score":0.109987795},"labels":[],"label_agreement":null},{"id":"W2579663487","doi":"10.1002/sim.7215","title":"Accounting for competing risks in randomized controlled trials: a review and recommendations for improvement","year":2017,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":152,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Randomized controlled trial; Cumulative incidence; Incidence (geometry); Outcome (game theory); Event (particle physics); Survival analysis; Medicine; Relative risk; Actuarial science; Confidence interval; Cohort; Internal medicine","score_opus":0.8315391831852751,"score_gpt":0.7107421776711227,"score_spread":0.1207970055141524,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2579663487","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00013658765,0.94542706,0.024946166,0.023486264,0.0028032637,0.0017635835,0.00040492904,0.0003140544,0.00071811577],"genre_scores_gemma":[0.0051692,0.8178201,0.15439053,0.010874607,0.0023115866,0.008116662,0.00065044634,0.00014800193,0.0005188343],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.7053757,0.19280425,0.07002151,0.00586092,0.024670053,0.0012676059],"domain_scores_gemma":[0.23743142,0.6514736,0.04511706,0.013251445,0.05076016,0.0019664178],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.35134336,0.0056226137,0.021836072,0.018602444,0.0016708894,0.0103307655,0.01192326,0.009319861,0.010318176],"category_scores_gemma":[0.5422609,0.004306477,0.023382785,0.018139008,0.0055046645,0.016132629,0.0051343227,0.01334883,0.003187943],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00045594078,0.00009593748,0.0009696681,0.37558475,0.008218248,0.00016110661,0.00080508075,0.0014534568,0.00018857054,0.017596323,0.045664717,0.54880625],"study_design_scores_gemma":[0.0012166885,0.00030601243,0.0016669398,0.7318075,0.020138212,0.0003647647,0.0005884843,0.0035554203,0.00040024635,0.0559381,0.18363214,0.00038555503],"about_ca_topic_score_codex":0.010374872,"about_ca_topic_score_gemma":0.01738652,"teacher_disagreement_score":0.6486566,"about_ca_system_score_codex":0.013468137,"about_ca_system_score_gemma":0.047621436,"threshold_uncertainty_score":0.7999091},"labels":[],"label_agreement":null},{"id":"W2582712477","doi":"10.1002/sim.7234","title":"Estimation of parametric failure time distributions based on interval‐censored data with irregular dependent follow‐up","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Truncation (statistics); Statistics; Observational study; Inverse probability; Estimator; Confidence interval; Parametric statistics; Confounding; Mathematics; Interval (graph theory); Event (particle physics); Marginal structural model; Medicine; Computer science; Econometrics; Bayesian probability","score_opus":0.0851947569602915,"score_gpt":0.4119459126927523,"score_spread":0.32675115573246083,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2582712477","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.041311886,0.00049549516,0.95682746,0.00018223695,0.00003177188,0.00016804569,0.00021398267,0.00019260065,0.0005763972],"genre_scores_gemma":[0.7358167,0.0015677536,0.25472257,0.00015765998,0.00014020376,0.001244639,0.0018692784,0.0001776501,0.0043035382],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99107516,0.00630343,0.0003826742,0.0010120709,0.0007951752,0.00043155072],"domain_scores_gemma":[0.85578394,0.12326619,0.009428986,0.008082252,0.0026270228,0.00081163616],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.035664845,0.0012615144,0.002425768,0.0023989072,0.00050284737,0.0019139623,0.0039834334,0.0017310121,0.0029116673],"category_scores_gemma":[0.12132042,0.00086764485,0.002708673,0.0021819905,0.0020243293,0.0022707884,0.002367591,0.0031591724,0.00051255437],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007462028,0.00022872654,0.031205632,0.00059603894,0.00063163176,0.0006754879,0.0009870564,0.6972361,0.0016512439,0.1386285,0.0022032117,0.1252101],"study_design_scores_gemma":[0.000044756667,0.00011110995,0.0034085256,0.00007191251,0.00007501525,0.00012877799,0.00009010779,0.9586732,0.0005863314,0.03593575,0.0008337464,0.00004072938],"about_ca_topic_score_codex":0.0037655672,"about_ca_topic_score_gemma":0.0021344665,"teacher_disagreement_score":0.035664845,"about_ca_system_score_codex":0.0012928753,"about_ca_system_score_gemma":0.0013911686,"threshold_uncertainty_score":0.18861598},"labels":[],"label_agreement":null},{"id":"W2587412035","doi":"10.1002/sim.7242","title":"Quantifying the bias in the estimated treatment effect in randomized trials having interim analyses and a rule for early stopping for futility","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Interim; Interim analysis; Early stopping; Treatment effect; Econometrics; Randomized controlled trial; Statistics; Computer science; Medicine; Mathematics; Law; Artificial intelligence; Surgery","score_opus":0.9144447875273284,"score_gpt":0.717800957027186,"score_spread":0.19664383050014234,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2587412035","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.012297551,0.004797205,0.9779759,0.0014502353,0.00022295197,0.0005117801,0.0001263442,0.0001909597,0.0024270026],"genre_scores_gemma":[0.373019,0.0039719017,0.6153886,0.0024075876,0.00045870995,0.0024387669,0.00035702004,0.00036624057,0.0015921673],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.81474596,0.14958385,0.010021287,0.006024349,0.018065827,0.0015588182],"domain_scores_gemma":[0.2938363,0.6501151,0.027972152,0.01892511,0.00850546,0.00064579525],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.24813014,0.0022501198,0.005214498,0.0036973257,0.000793407,0.0057980553,0.003976836,0.005885266,0.0029895667],"category_scores_gemma":[0.5817924,0.0011015066,0.0040141633,0.0031500268,0.004595312,0.0070565916,0.003358514,0.006792852,0.0008874992],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001828983,0.00025105372,0.01826561,0.0044600265,0.0028902846,0.0011530714,0.0019617756,0.25918683,0.0030931644,0.41800457,0.004239901,0.28466478],"study_design_scores_gemma":[0.00062506704,0.0011065053,0.006672855,0.0029283036,0.0015705801,0.001648916,0.0002516471,0.43320882,0.006216778,0.5346067,0.0108095985,0.00035421966],"about_ca_topic_score_codex":0.0006940888,"about_ca_topic_score_gemma":0.0004104101,"teacher_disagreement_score":0.75186986,"about_ca_system_score_codex":0.002564598,"about_ca_system_score_gemma":0.0034096797,"threshold_uncertainty_score":0.9271894},"labels":[],"label_agreement":null},{"id":"W2589124067","doi":"10.1002/sim.7251","title":"Binocular sensitivity and specificity of screening tests in cross‐sectional diagnostic studies of paired organs","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Systemic Lupus Erythematosus Research","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta; University of Calgary","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Copula (linguistics); Medical diagnosis; Probit model; Computer science; Artificial intelligence; Correlation; Statistics; Mathematics; Optometry; Medicine; Pathology; Econometrics","score_opus":0.09399225451713349,"score_gpt":0.414452970517695,"score_spread":0.3204607160005615,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2589124067","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7538706,0.005777555,0.22988145,0.0005092133,0.00017224682,0.00024250672,0.001969821,0.00025081035,0.007325752],"genre_scores_gemma":[0.98706704,0.00027452304,0.012016333,0.00008547021,0.000048955473,0.00007900715,0.0002661266,0.000012521578,0.00015000244],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9673758,0.02257069,0.002775695,0.003237322,0.00305372,0.0009868355],"domain_scores_gemma":[0.82019114,0.1409891,0.02280498,0.010744102,0.003729476,0.0015412373],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04337935,0.000959569,0.0013800984,0.0049539055,0.00039034535,0.0021081173,0.0010874723,0.0012958316,0.0018523874],"category_scores_gemma":[0.15468529,0.00052140787,0.002073042,0.0030094278,0.002792959,0.002038882,0.004014964,0.0009807728,0.00026110566],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009344669,0.00007835768,0.91277266,0.0005432238,0.0018387386,0.00048038922,0.0008587647,0.0109782005,0.0016906379,0.011778419,0.000616879,0.0574293],"study_design_scores_gemma":[0.000103813036,0.0018683542,0.828131,0.00037213133,0.0014998001,0.0046349587,0.001296731,0.1064781,0.0050713303,0.04670622,0.0036535976,0.00018402416],"about_ca_topic_score_codex":0.0016466664,"about_ca_topic_score_gemma":0.0015162476,"teacher_disagreement_score":0.95662063,"about_ca_system_score_codex":0.0006791098,"about_ca_system_score_gemma":0.0009023087,"threshold_uncertainty_score":0.2294147},"labels":[],"label_agreement":null},{"id":"W2604188292","doi":"10.1002/sim.7289","title":"Correcting covariate‐dependent measurement error with non‐zero mean","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Biotechnology Research Institute; McGill University","funders":"","keywords":"Covariate; Statistics; Observational error; Extrapolation; Regression; Imputation (statistics); Econometrics; Calibration; Standard error; Mathematics; Computer science; Missing data","score_opus":0.07450561637566287,"score_gpt":0.3482761329384653,"score_spread":0.27377051656280244,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2604188292","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010467781,0.0003269692,0.9880112,0.00029593485,0.00007819747,0.0000985457,0.00012397411,0.0003354769,0.0002618192],"genre_scores_gemma":[0.40580705,0.0006801316,0.588562,0.00046276322,0.00016212171,0.0007158175,0.00082878844,0.00015853117,0.0026229175],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9664338,0.025388405,0.0015041346,0.0041998043,0.0019131218,0.0005606076],"domain_scores_gemma":[0.883484,0.0811681,0.00811538,0.022399686,0.0043856413,0.0004472701],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06263764,0.0015006312,0.0025718343,0.002023918,0.0015797572,0.0019727554,0.0040315012,0.0034230212,0.002229604],"category_scores_gemma":[0.18281718,0.0012278844,0.00304383,0.004140587,0.0027506903,0.0021067115,0.0032361492,0.0032867643,0.0007304832],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008118874,0.0002816792,0.1230872,0.00088246027,0.0036613254,0.0010158839,0.0020435222,0.28133494,0.0037870859,0.1832129,0.0066978224,0.39318338],"study_design_scores_gemma":[0.00023821485,0.00040734562,0.028975613,0.00036449984,0.0008055274,0.0007432932,0.0002477971,0.7518696,0.0061292425,0.1978084,0.012210542,0.00019981553],"about_ca_topic_score_codex":0.010101991,"about_ca_topic_score_gemma":0.010063369,"teacher_disagreement_score":0.06263764,"about_ca_system_score_codex":0.0016383453,"about_ca_system_score_gemma":0.0033889972,"threshold_uncertainty_score":0.33126348},"labels":[],"label_agreement":null},{"id":"W2605130264","doi":"10.1002/sim.7298","title":"A comparison of Bayesian and Monte Carlo sensitivity analysis for unmeasured confounding","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":47,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; Simon Fraser University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Frequentist inference; Bayes' theorem; Prior probability; Sensitivity (control systems); Bayesian probability; Econometrics; Monte Carlo method; Statistics; Confounding; Posterior probability; Contrast (vision); Computer science; Mathematics; Bayesian inference; Artificial intelligence","score_opus":0.1480890340324776,"score_gpt":0.49242406635217234,"score_spread":0.34433503231969476,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2605130264","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03258826,0.015560908,0.9323462,0.0050230455,0.0007363323,0.0020770724,0.0008165822,0.00046393814,0.01038773],"genre_scores_gemma":[0.6829131,0.0076300134,0.2989583,0.003361765,0.0005462538,0.004114117,0.000610673,0.00027533813,0.0015904163],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.669489,0.3056965,0.005266724,0.004786397,0.0132854255,0.0014759016],"domain_scores_gemma":[0.2607548,0.70614845,0.01082846,0.012571695,0.009003098,0.00069349376],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.24223456,0.0023586405,0.0042798296,0.006811631,0.001337885,0.004689804,0.0035947652,0.004330519,0.0060982965],"category_scores_gemma":[0.5311684,0.0015265058,0.009391157,0.005135446,0.004001137,0.008129991,0.0050995545,0.004694429,0.00041023036],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002994785,0.00033124804,0.013972437,0.0058169314,0.01160135,0.00049933343,0.0016112841,0.4412928,0.00067889754,0.3782015,0.006826407,0.13617298],"study_design_scores_gemma":[0.0008101885,0.00075702317,0.008339328,0.00300017,0.0034092078,0.0006754105,0.00048386937,0.5008111,0.0010236605,0.4673674,0.012859143,0.00046341371],"about_ca_topic_score_codex":0.0054178247,"about_ca_topic_score_gemma":0.0036986414,"teacher_disagreement_score":0.7577654,"about_ca_system_score_codex":0.0050918935,"about_ca_system_score_gemma":0.0058103716,"threshold_uncertainty_score":0.9344597},"labels":[],"label_agreement":null},{"id":"W2609432052","doi":"10.1002/sim.7300","title":"Constructing longitudinal disease progression curves using sparse, short‐term individual data with an application to Alzheimer's disease","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Neuroimaging Techniques and Applications","field":"Medicine","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Institute on Aging; National Institute of Biomedical Imaging and Bioengineering; Canadian Institutes of Health Research; Genentech; National Institutes of Health; IXICO; H. Lundbeck A/S; Servier; Eisai; Department of Health, Government of Western Australia; Northern California Institute for Research and Education; Pfizer; Biogen; BioClinica; Eli Lilly and Company; U.S. Department of Defense; Alzheimer's Disease Neuroimaging Initiative; Meso Scale Diagnostics; University of Southern California; F. Hoffmann-La Roche; Australian Government; Novartis Pharmaceuticals Corporation; Government of Western Australia; Bristol-Myers Squibb; Alzheimer's Drug Discovery Foundation; Australian Institute of Health and Welfare, Australian Government; Foundation for the National Institutes of Health","keywords":"Term (time); Trajectory; Construct (python library); Regression; Computer science; Longitudinal data; Disease; Regression analysis; Statistics; Algorithm; Mathematics; Applied mathematics; Machine learning; Medicine; Data mining; Pathology","score_opus":0.2365476554198065,"score_gpt":0.48871271654198184,"score_spread":0.25216506112217535,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2609432052","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09368449,0.00028023202,0.904322,0.0003072903,0.000021401765,0.00011510728,0.00033347745,0.0005167742,0.00041926425],"genre_scores_gemma":[0.5495142,0.0005056954,0.44766057,0.000076611825,0.000027758475,0.00024716355,0.0010025135,0.00016139784,0.00080413016],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9971283,0.0018792021,0.00012865434,0.00040588286,0.00035996595,0.00009806604],"domain_scores_gemma":[0.97205293,0.020446666,0.0025283983,0.0031068716,0.0015628062,0.00030220515],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011088007,0.00060372864,0.0007400894,0.0016970607,0.0005732649,0.0011725371,0.0012599701,0.0011977553,0.0012826596],"category_scores_gemma":[0.04872779,0.0005450719,0.0016428712,0.0021027036,0.00083952997,0.0015163168,0.001583755,0.0018180808,0.00041136175],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00025863256,0.00014438282,0.030876556,0.00031837964,0.00027689835,0.00026166064,0.0014898926,0.77173495,0.0035968476,0.025288355,0.0008538575,0.16489957],"study_design_scores_gemma":[0.000020650852,0.0001605863,0.010223957,0.00006936382,0.00003444411,0.00014558114,0.00018393286,0.96258074,0.0012751501,0.022883287,0.0023402611,0.00008209865],"about_ca_topic_score_codex":0.011049513,"about_ca_topic_score_gemma":0.0103985155,"teacher_disagreement_score":0.011088007,"about_ca_system_score_codex":0.0008395015,"about_ca_system_score_gemma":0.0015067841,"threshold_uncertainty_score":0.058639705},"labels":[],"label_agreement":null},{"id":"W2618058430","doi":"10.1002/sim.7336","title":"Intermediate and advanced topics in multilevel logistic regression analysis","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":657,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sunnybrook Hospital; University of Toronto; Institute for Work & Health; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Vetenskapsrådet; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Marginal model; Logistic regression; Covariate; Multilevel model; Statistics; Hierarchical clustering; Econometrics; Regression analysis; Population; Odds ratio; Regression; Cluster analysis; Computer science; Mathematics; Demography","score_opus":0.1141231773850907,"score_gpt":0.47828034943776926,"score_spread":0.36415717205267856,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2618058430","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.004759708,0.042782024,0.90992975,0.022421474,0.0029564255,0.00032621494,0.0016682983,0.0014545217,0.013701613],"genre_scores_gemma":[0.1237067,0.048938278,0.78533006,0.0072334465,0.014445159,0.0023805648,0.00446634,0.0022663285,0.011233218],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.964048,0.025929125,0.0021216215,0.0026724956,0.0046835863,0.00054511917],"domain_scores_gemma":[0.8925722,0.09010526,0.0054546753,0.0060950844,0.0047205086,0.0010523826],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.028343363,0.0014705481,0.002011997,0.005637617,0.0010886597,0.00432164,0.0024974607,0.0025269564,0.017196676],"category_scores_gemma":[0.13324374,0.0009230289,0.0036031443,0.009889791,0.0030119696,0.0044966484,0.005159437,0.008467041,0.0058280467],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016895018,0.00015490218,0.014345222,0.0040883957,0.0011576968,0.0006751585,0.0014489723,0.011388416,0.0008708251,0.4339261,0.12571655,0.4060588],"study_design_scores_gemma":[0.00004978274,0.00018303789,0.0060369424,0.0021817584,0.00025287428,0.00065927615,0.00033182086,0.046344817,0.00065013237,0.7539292,0.18925026,0.00013015997],"about_ca_topic_score_codex":0.0019302104,"about_ca_topic_score_gemma":0.0017646108,"teacher_disagreement_score":0.028343363,"about_ca_system_score_codex":0.00169445,"about_ca_system_score_gemma":0.0035671857,"threshold_uncertainty_score":0.14989585},"labels":[],"label_agreement":null},{"id":"W2625352524","doi":"10.1002/sim.7346","title":"Analysis of panel data under hidden mover-stayer models","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Inference; Expectation–maximization algorithm; Econometrics; Latent variable; Maximum likelihood; Panel data; Maximization; Population; Statistics; Data mining; Artificial intelligence; Mathematical optimization; Mathematics","score_opus":0.31928921598111903,"score_gpt":0.479378919571194,"score_spread":0.16008970359007496,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2625352524","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04638058,0.0003430246,0.95156866,0.000517051,0.000037231664,0.000089478155,0.00045722115,0.00017321196,0.00043345775],"genre_scores_gemma":[0.7587324,0.00090329075,0.2331556,0.00040845512,0.00019643713,0.00053212617,0.002203884,0.00009425534,0.0037736187],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9818288,0.014237037,0.00052262685,0.0020877162,0.00083863724,0.00048525142],"domain_scores_gemma":[0.88969964,0.09445087,0.007528656,0.006762402,0.0011453732,0.00041305934],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04159486,0.00089296437,0.0025997174,0.001546739,0.0008748227,0.0021070843,0.0034080448,0.0024474103,0.0029321457],"category_scores_gemma":[0.085176095,0.0009512958,0.0025080233,0.00235207,0.0019452382,0.002721296,0.0022610656,0.0034813609,0.00045480125],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006220526,0.00020924465,0.056686476,0.00043948094,0.002496127,0.0011345658,0.0010876957,0.5697369,0.00135653,0.27953306,0.0030642715,0.08363363],"study_design_scores_gemma":[0.000070064765,0.000092664544,0.005636436,0.00004032993,0.00021074177,0.00008932879,0.00008749006,0.8474089,0.00048263682,0.14486879,0.0009779644,0.000034774894],"about_ca_topic_score_codex":0.006918287,"about_ca_topic_score_gemma":0.0050343815,"teacher_disagreement_score":0.04159486,"about_ca_system_score_codex":0.0011411307,"about_ca_system_score_gemma":0.0013680239,"threshold_uncertainty_score":0.21997732},"labels":[],"label_agreement":null},{"id":"W2736786428","doi":"10.1002/sim.8673","title":"Developing biomarker combinations in multicenter studies via direct maximization and penalization","year":2020,"lang":"en","type":"preprint","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Pritzker School of Medicine; National Institutes of Health; National Institute of Diabetes and Digestive and Kidney Diseases; National Heart, Lung, and Blood Institute; Institute for Clinical Evaluative Sciences; U.S. Department of Veterans Affairs","keywords":"Biomarker; Logistic regression; Maximization; Computer science; Data mining; Machine learning; Mathematics; Mathematical optimization; Biology","score_opus":0.24837774175540903,"score_gpt":0.4778237775230547,"score_spread":0.2294460357676457,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2736786428","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014160304,0.0007136057,0.98291403,0.0007611173,0.000053138225,0.000350319,0.00017679547,0.00026557813,0.0006051576],"genre_scores_gemma":[0.17580669,0.0004862783,0.81999797,0.00041320035,0.00021291507,0.0015264108,0.0005760869,0.00022247928,0.0007579716],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.956499,0.037844747,0.0013574619,0.0026135226,0.0012408472,0.0004445357],"domain_scores_gemma":[0.8728163,0.10776208,0.008517985,0.005866003,0.003414342,0.0016233017],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0836506,0.0027368744,0.0038909193,0.0037071242,0.0010671533,0.0027919076,0.0031477383,0.0023739224,0.002357585],"category_scores_gemma":[0.13505076,0.0024316092,0.003542298,0.003556642,0.0020742964,0.0028013678,0.0054961615,0.0028876984,0.0005567574],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011987326,0.00030435764,0.019262012,0.001059279,0.0031269751,0.00082859903,0.00044741877,0.77098054,0.0031902187,0.05108631,0.0044332515,0.14408225],"study_design_scores_gemma":[0.00022716554,0.00037655004,0.0018020857,0.00013526385,0.00038039443,0.0001923925,0.000050362818,0.9193212,0.001372368,0.07373974,0.0023508796,0.000051536786],"about_ca_topic_score_codex":0.0013726426,"about_ca_topic_score_gemma":0.0018295837,"teacher_disagreement_score":0.0836506,"about_ca_system_score_codex":0.0013656522,"about_ca_system_score_gemma":0.0031203316,"threshold_uncertainty_score":0.44239193},"labels":[],"label_agreement":null},{"id":"W2739069932","doi":"10.1002/sim.7397","title":"Dynamic classification using credible intervals in longitudinal discriminant analysis","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council Canada; Belgian Federal Science Policy Office; Institut National de la Santé et de la Recherche Médicale; Medical Research Council; National Institute for Health and Care Research","keywords":"Linear discriminant analysis; False positive paradox; Bayesian probability; Multivariate statistics; Context (archaeology); Time point; Discriminant; Computer science; Confidence interval; Bayes' theorem; Statistics; Data mining; Set (abstract data type); Data set; Artificial intelligence; Machine learning; Mathematics","score_opus":0.10170984138903208,"score_gpt":0.42324950014943274,"score_spread":0.3215396587604007,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2739069932","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007722948,0.00089791673,0.99032557,0.00030794123,0.000032396336,0.000034098644,0.000053670086,0.00014531249,0.00048015066],"genre_scores_gemma":[0.44526175,0.0017393511,0.54951537,0.00025166263,0.0003649736,0.0005803675,0.0006301218,0.00023085382,0.00142553],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96937704,0.022414172,0.0011050458,0.003648857,0.0028661834,0.0005885927],"domain_scores_gemma":[0.79027945,0.18827741,0.008336385,0.007929224,0.0042663934,0.00091101613],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.050719783,0.0016908104,0.0030991223,0.006302904,0.0013812003,0.0037927562,0.003936781,0.002772613,0.001868399],"category_scores_gemma":[0.20804022,0.001247905,0.0021058936,0.0039713453,0.0049630506,0.004185031,0.004491737,0.005815393,0.0005642368],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00042529675,0.000096640004,0.007893928,0.00031970302,0.0004206956,0.00032419842,0.0007736991,0.56722957,0.0008965173,0.29410887,0.0016177336,0.12589309],"study_design_scores_gemma":[0.0000360289,0.000042256237,0.0008910893,0.000093187,0.00004090318,0.00006120854,0.000052963907,0.8153654,0.00046549313,0.18182236,0.0010566906,0.00007233304],"about_ca_topic_score_codex":0.0055839857,"about_ca_topic_score_gemma":0.0023582648,"teacher_disagreement_score":0.050719783,"about_ca_system_score_codex":0.0024946574,"about_ca_system_score_gemma":0.0019565374,"threshold_uncertainty_score":0.26823503},"labels":[],"label_agreement":null},{"id":"W2743866054","doi":"10.1002/sim.7401","title":"A joint modeling and estimation method for multivariate longitudinal data with mixed types of responses to analyze physical activity data generated by accelerometers","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Urban Transport and Accessibility","field":"Social Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"National Cancer Institute; Natural Sciences and Engineering Research Council of Canada","keywords":"Multivariate statistics; Accelerometer; Computer science; Longitudinal data; Estimation; Multivariate analysis; Joint (building); Statistics; Econometrics; Data mining; Mathematics; Machine learning","score_opus":0.2269431800474397,"score_gpt":0.4798321457310576,"score_spread":0.25288896568361785,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2743866054","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00046328933,0.000040909516,0.99931467,0.000026500065,0.000010594998,0.000016381317,0.000029892046,0.00005937162,0.00003832348],"genre_scores_gemma":[0.04540367,0.0003236648,0.9509679,0.000108661516,0.00012852403,0.000946959,0.00056447525,0.00014613252,0.001410003],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9918283,0.0059237615,0.0003446075,0.0010244164,0.00065773574,0.00022113346],"domain_scores_gemma":[0.9863512,0.010191622,0.000913281,0.0012397734,0.0010925706,0.00021159754],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.015227971,0.0015277585,0.0021623229,0.0019838987,0.0007222556,0.0013156247,0.0035397843,0.0011989772,0.0035618034],"category_scores_gemma":[0.025779031,0.0013215534,0.003992851,0.0024146687,0.001051912,0.0020757518,0.0026695724,0.0028462922,0.0010127332],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00037923115,0.000324291,0.012384629,0.0006460214,0.0018838765,0.00037998267,0.00065878156,0.39326265,0.0051352535,0.25792533,0.0050041117,0.32201573],"study_design_scores_gemma":[0.000031040017,0.00010365122,0.0007685739,0.0000375169,0.00011285718,0.0000958851,0.000031868884,0.9634904,0.0007239752,0.03161496,0.0029476841,0.000041607775],"about_ca_topic_score_codex":0.004737266,"about_ca_topic_score_gemma":0.0044548293,"teacher_disagreement_score":0.015227971,"about_ca_system_score_codex":0.00072200544,"about_ca_system_score_gemma":0.0021483891,"threshold_uncertainty_score":0.08053416},"labels":[],"label_agreement":null},{"id":"W2744856250","doi":"10.1002/sim.7427","title":"Bayesian analysis of pair‐matched case‐control studies subject to outcome misclassification","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"British Columbia Centre of Excellence for Women's Health; Vancouver Coastal Health; University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada; National Multiple Sclerosis Society","keywords":"Outcome (game theory); Bayesian probability; Computer science; Subject (documents); Statistics; Econometrics; Artificial intelligence; Mathematics","score_opus":0.17087201834420754,"score_gpt":0.48955020471738364,"score_spread":0.31867818637317613,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2744856250","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.075069174,0.004863337,0.91697127,0.0010479082,0.0001330315,0.0004979185,0.00042056333,0.00024675598,0.00075019104],"genre_scores_gemma":[0.7722879,0.0027253747,0.22012594,0.0007716999,0.00025594994,0.0015867767,0.0010664922,0.0000840265,0.001095872],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.87371933,0.10995962,0.004077927,0.0068347678,0.0047209235,0.00068748207],"domain_scores_gemma":[0.42688504,0.529922,0.021181954,0.016968032,0.00427773,0.0007652608],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.19141617,0.0018841908,0.0037267066,0.0043434254,0.0010807028,0.0034049351,0.0048772967,0.0033301406,0.0022317956],"category_scores_gemma":[0.45363492,0.001624711,0.0036168003,0.0031905049,0.0025518239,0.0022186441,0.0028485954,0.0027018378,0.00031782498],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004110856,0.00047999463,0.17461297,0.0026050669,0.019044297,0.0024676982,0.0018611858,0.3635143,0.0020136079,0.15759765,0.0035728307,0.26811954],"study_design_scores_gemma":[0.0007990337,0.00058517733,0.023070136,0.0005480581,0.004304119,0.0012065256,0.00015690601,0.73282814,0.0012297039,0.23220672,0.0029296898,0.0001357858],"about_ca_topic_score_codex":0.0054367036,"about_ca_topic_score_gemma":0.0027702646,"teacher_disagreement_score":0.19141617,"about_ca_system_score_codex":0.0014529562,"about_ca_system_score_gemma":0.0014567329,"threshold_uncertainty_score":0.99712783},"labels":[],"label_agreement":null},{"id":"W2752127454","doi":"10.1002/sim.7440","title":"Algorithms for evaluating reference scaled average bioequivalence: power, bias, and consumer risk","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Bioequivalence; Computer science; Statistics; Econometrics; Sample size determination; Statistical power; Mathematics; Medicine","score_opus":0.7233084269583646,"score_gpt":0.6410520296377792,"score_spread":0.08225639732058543,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2752127454","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0036479814,0.0005696586,0.9946542,0.00017836693,0.000027665992,0.000114422844,0.0000340707,0.00026400192,0.00050972315],"genre_scores_gemma":[0.06363994,0.0006438229,0.9337115,0.00022134431,0.00007555824,0.000710386,0.000150329,0.0001756398,0.0006714005],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9733979,0.017351897,0.0017744269,0.0016786923,0.0054337326,0.00036326642],"domain_scores_gemma":[0.85917926,0.1191738,0.005254732,0.0067331,0.009222351,0.00043666328],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04258199,0.0017737192,0.0024617456,0.0045425957,0.0006541864,0.0028087771,0.0027713466,0.0025002384,0.0037553585],"category_scores_gemma":[0.15263236,0.0009577601,0.0019896054,0.00278481,0.002514002,0.0040935054,0.0028947887,0.002974856,0.0010644712],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00059369736,0.00011775219,0.005860494,0.000528619,0.00047733047,0.00009095886,0.00032193423,0.2615975,0.0022996461,0.08181633,0.0033278528,0.6429678],"study_design_scores_gemma":[0.00023773617,0.00038325135,0.0029427179,0.00027387257,0.00014904316,0.00037193423,0.00007985652,0.8690595,0.004968295,0.115821965,0.0055664103,0.00014541352],"about_ca_topic_score_codex":0.0019701594,"about_ca_topic_score_gemma":0.0013735252,"teacher_disagreement_score":0.04258199,"about_ca_system_score_codex":0.00175751,"about_ca_system_score_gemma":0.0030012608,"threshold_uncertainty_score":0.22519779},"labels":[],"label_agreement":null},{"id":"W2752310745","doi":"10.1002/sim.7449","title":"Modeling conditional dependence among multiple diagnostic tests","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Distribution Estimation and Applications","field":"Mathematics","cited_by":44,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Natural Sciences and Engineering Research Council of Canada; National Institutes of Health; Canadian Institutes of Health Research; Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada; Medical Research Council; South African Medical Research Council","keywords":"Diagnostic test; Conditional dependence; Statistics; Pairwise comparison; Bayesian probability; Conditional independence; Statistical hypothesis testing; Econometrics; Random effects model; Mathematics; Computer science; Medicine","score_opus":0.10206920322916589,"score_gpt":0.42708298095350533,"score_spread":0.3250137777243394,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2752310745","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1300975,0.0012046683,0.8619766,0.0018562658,0.00014405051,0.0003360949,0.0009670181,0.00037330523,0.0030445457],"genre_scores_gemma":[0.883192,0.0007809606,0.10899806,0.00046457408,0.00020105368,0.00056684576,0.0010649131,0.00008841003,0.004643137],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9782931,0.013475976,0.0008820845,0.004031658,0.0020612187,0.0012560218],"domain_scores_gemma":[0.8003184,0.17100927,0.013470821,0.009118557,0.004884211,0.0011987122],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03972425,0.0015591,0.002657986,0.0028063287,0.0010042585,0.0026547797,0.0043489966,0.0032568227,0.0045870435],"category_scores_gemma":[0.11352914,0.0018274027,0.0026073237,0.0026393346,0.004068169,0.0030901693,0.00352299,0.0043073357,0.00063207664],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012942639,0.000365207,0.096326254,0.0005044766,0.0015482836,0.0021222453,0.0015373208,0.5745779,0.0018556173,0.22003923,0.0028697252,0.09695942],"study_design_scores_gemma":[0.0001604811,0.0002530201,0.014738003,0.00018625065,0.0007673917,0.0005494117,0.000096094125,0.8532652,0.0012176032,0.12643763,0.0022165289,0.00011240875],"about_ca_topic_score_codex":0.014527986,"about_ca_topic_score_gemma":0.012110844,"teacher_disagreement_score":0.03972425,"about_ca_system_score_codex":0.0027809567,"about_ca_system_score_gemma":0.0030721743,"threshold_uncertainty_score":0.21008438},"labels":[],"label_agreement":null},{"id":"W2752545986","doi":"10.1002/sim.7453","title":"Improving phase II oncology trials using best observed RECIST response as an endpoint by modelling continuous tumour measurements","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada; Cancer Research UK; AstraZeneca","keywords":"Response Evaluation Criteria in Solid Tumors; Sample size determination; Clinical endpoint; Clinical trial; Estimator; Medicine; Oncology; Outcome (game theory); Complete response; Computer science; Statistics; Internal medicine; Phases of clinical research; Mathematics; Chemotherapy","score_opus":0.8504567930226415,"score_gpt":0.6474845982302354,"score_spread":0.20297219479240614,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2752545986","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01361829,0.0014834335,0.9814689,0.0010406919,0.00017890015,0.0006709261,0.00018136094,0.00041947645,0.0009381004],"genre_scores_gemma":[0.31468058,0.00085811835,0.6793344,0.0008046667,0.00023490164,0.0024013736,0.00050466316,0.00019904727,0.0009822878],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8937006,0.097051315,0.0024196613,0.0023388974,0.0040638386,0.00042569108],"domain_scores_gemma":[0.76639724,0.20322953,0.013821479,0.0114297215,0.004280506,0.00084155117],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09093987,0.0016481369,0.0029794162,0.0012980409,0.0004859234,0.0024489276,0.0019799958,0.0024214103,0.0022178115],"category_scores_gemma":[0.22431561,0.00094612414,0.002419698,0.0015972255,0.0011867513,0.0024794328,0.0021606404,0.0030720353,0.00062095467],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.007315574,0.0006309902,0.016453808,0.002925557,0.0030696013,0.00029817355,0.0006654452,0.5012247,0.005192518,0.048592053,0.007423808,0.4062078],"study_design_scores_gemma":[0.0019189059,0.0023351756,0.0041188262,0.0003060665,0.00059925456,0.00027056175,0.000044258606,0.90927863,0.003072994,0.06867177,0.009247239,0.00013642403],"about_ca_topic_score_codex":0.0007358298,"about_ca_topic_score_gemma":0.00076393306,"teacher_disagreement_score":0.9090601,"about_ca_system_score_codex":0.0008417588,"about_ca_system_score_gemma":0.0024226971,"threshold_uncertainty_score":0.48094177},"labels":[],"label_agreement":null},{"id":"W2754160328","doi":"10.1002/sim.7501","title":"Practical recommendations for reporting<scp>F</scp>ine‐<scp>G</scp>ray model analyses for competing risk data","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":1176,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Heart and Stroke Foundation of Canada","keywords":"Covariate; Proportional hazards model; Statistics; Regression analysis; Econometrics; Regression; Event (particle physics); Hazard; Mathematics; Biology","score_opus":0.7862327183202007,"score_gpt":0.6125248362078489,"score_spread":0.17370788211235177,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2754160328","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0021059033,0.024978895,0.38319826,0.30803224,0.105760135,0.026243763,0.060629804,0.021977501,0.06707347],"genre_scores_gemma":[0.009412916,0.018415036,0.733041,0.113855064,0.013958609,0.053575937,0.023969973,0.008519219,0.025252303],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.58296514,0.26938564,0.088712804,0.00601859,0.04878734,0.0041305306],"domain_scores_gemma":[0.13285784,0.4924123,0.061205357,0.06263549,0.24402368,0.0068654134],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.30931008,0.0039831134,0.0072899507,0.020954732,0.0037502293,0.017159112,0.015517524,0.020799479,0.097985044],"category_scores_gemma":[0.758992,0.005966932,0.013667555,0.026375612,0.0062874286,0.017288165,0.00960404,0.021227974,0.075301945],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00020839492,0.000172651,0.001023936,0.0064524016,0.000286545,0.00020826381,0.0004874475,0.0014411526,0.00029378783,0.008188965,0.9224579,0.058778517],"study_design_scores_gemma":[0.00078483723,0.0002107025,0.0024845307,0.029378833,0.00043070363,0.0004834707,0.0011158452,0.004822148,0.0014190751,0.036714017,0.9217334,0.00042254166],"about_ca_topic_score_codex":0.015132575,"about_ca_topic_score_gemma":0.013929063,"teacher_disagreement_score":0.6906899,"about_ca_system_score_codex":0.0077144667,"about_ca_system_score_gemma":0.043844227,"threshold_uncertainty_score":0.85174364},"labels":[],"label_agreement":null},{"id":"W2755786236","doi":"10.1002/sim.8013","title":"One‐sample aggregate data meta‐analysis of medians","year":2018,"lang":"en","type":"preprint","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Fonds de Recherche du Québec - Santé","keywords":"Median; Meta-analysis; Statistics; Weighting; Sample size determination; Pooled variance; Standard deviation; Variance (accounting); Sample (material); Outcome (game theory); Mathematics; Data set; Computer science; Confidence interval; Medicine","score_opus":0.327383040627426,"score_gpt":0.4826645932734693,"score_spread":0.15528155264604326,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2755786236","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.025737053,0.17690244,0.77733845,0.0029563387,0.0020788312,0.002457805,0.0072623244,0.0019017801,0.003365029],"genre_scores_gemma":[0.55521744,0.04056615,0.38505524,0.0025396042,0.0012330242,0.007346069,0.0048692,0.0008044326,0.0023688234],"study_design_codex":"meta_analysis","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.90149504,0.08572168,0.0039560604,0.0053071612,0.0031002618,0.00041969548],"domain_scores_gemma":[0.8698329,0.11179847,0.0045736097,0.011122853,0.0023760705,0.00029610674],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08599191,0.0032044125,0.009793665,0.0067839082,0.0008007033,0.004286417,0.002925723,0.0023204528,0.0055986736],"category_scores_gemma":[0.18966226,0.0012193094,0.026977796,0.005911604,0.0011533425,0.0033676415,0.0023013547,0.004024423,0.00053828745],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0070102704,0.00020858005,0.014027215,0.045922942,0.6601808,0.0005231783,0.00034849066,0.06816944,0.0014711774,0.031995684,0.009502305,0.16063985],"study_design_scores_gemma":[0.0055820504,0.0019692678,0.009966072,0.007757929,0.6101407,0.00057858473,0.00026376214,0.113625355,0.003960551,0.21081045,0.03503544,0.0003099427],"about_ca_topic_score_codex":0.002156603,"about_ca_topic_score_gemma":0.002176486,"teacher_disagreement_score":0.9140081,"about_ca_system_score_codex":0.0017630551,"about_ca_system_score_gemma":0.0021537705,"threshold_uncertainty_score":0.45477414},"labels":[],"label_agreement":null},{"id":"W2756610413","doi":"10.1002/sim.7532","title":"Measures of clustering and heterogeneity in multilevel <scp>P</scp>oisson regression analyses of rates/count data","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":206,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Prince Edward Island; University of Toronto; Institute for Work & Health; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Vetenskapsrådet; Heart and Stroke Foundation of Canada","keywords":"Statistics; Poisson regression; Multilevel model; Poisson distribution; Hazard ratio; Proportional hazards model; Count data; Random effects model; Regression analysis; Cluster analysis; Confidence interval; Mathematics; Odds ratio; Rate ratio; Linear regression; Hierarchical clustering; Cluster (spacecraft); Econometrics; Medicine; Meta-analysis; Computer science; Internal medicine; Population","score_opus":0.4042531616015181,"score_gpt":0.5285301629397802,"score_spread":0.12427700133826208,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2756610413","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.23211281,0.003102074,0.73889863,0.004013534,0.0003837951,0.0007307314,0.008288891,0.0009012612,0.011568237],"genre_scores_gemma":[0.89734703,0.0005351271,0.09524191,0.00046047088,0.00022302534,0.0011801878,0.0030531508,0.00021192418,0.0017471695],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9649846,0.02073218,0.003008292,0.004096145,0.006469579,0.000709293],"domain_scores_gemma":[0.81461823,0.1374074,0.024037508,0.018035416,0.005137029,0.0007643289],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02922282,0.0004972554,0.001091739,0.0047212453,0.0007556032,0.0019752579,0.0017615971,0.0010450681,0.004373983],"category_scores_gemma":[0.17080471,0.00038010243,0.0023389815,0.006826541,0.002038914,0.0020467814,0.0021587948,0.0017541035,0.00053573685],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003745402,0.00018061147,0.5183472,0.0013663252,0.006040739,0.001478373,0.0031401527,0.083669215,0.0013700634,0.18639962,0.02021203,0.17742112],"study_design_scores_gemma":[0.000089017994,0.00034026854,0.3625159,0.0008440107,0.0008011307,0.0016999057,0.0020073825,0.24733981,0.0022809163,0.35853666,0.02327658,0.0002684401],"about_ca_topic_score_codex":0.0057877763,"about_ca_topic_score_gemma":0.0038820233,"teacher_disagreement_score":0.02922282,"about_ca_system_score_codex":0.0015634194,"about_ca_system_score_gemma":0.001049616,"threshold_uncertainty_score":0.15454686},"labels":[],"label_agreement":null},{"id":"W2765200233","doi":"10.1002/sim.7534","title":"Three‐part joint modeling methods for complex functional data mixed with zero‐and‐one–inflated proportions and zero‐inflated continuous outcomes with skewness","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Mental Health Research Topics","field":"Psychology","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"National Cancer Institute; National Heart, Lung, and Blood Institute; Natural Sciences and Engineering Research Council of Canada; National Institutes of Health","keywords":"Bivariate analysis; Skewness; Statistics; Mathematics; Zero (linguistics); Econometrics; Interval (graph theory); Combinatorics","score_opus":0.4264507123200055,"score_gpt":0.5347583986790945,"score_spread":0.108307686359089,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2765200233","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0031255146,0.00019365996,0.9957066,0.00020658252,0.00005207572,0.0001267118,0.00015715041,0.00023488949,0.00019681943],"genre_scores_gemma":[0.20427887,0.0008015917,0.782418,0.0006771627,0.00042997664,0.0039568776,0.001636932,0.00047155604,0.005329074],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9704367,0.022576874,0.0010523781,0.0036497202,0.0015607122,0.00072359113],"domain_scores_gemma":[0.9292927,0.054661613,0.0047792424,0.007466478,0.0030117591,0.00078815396],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.052314896,0.0031350353,0.004010221,0.0032407746,0.0015390422,0.0042317463,0.007215412,0.0033906503,0.00729189],"category_scores_gemma":[0.07976369,0.0021220017,0.007651708,0.0033047162,0.0034878,0.004426064,0.006630806,0.0051810844,0.0016713922],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007731468,0.00034525202,0.025771715,0.00072976673,0.0027438367,0.0009815417,0.00225942,0.37557635,0.0016301204,0.3976441,0.004805706,0.18673909],"study_design_scores_gemma":[0.000062137646,0.00018403122,0.0013611171,0.00009681288,0.00017190131,0.00010163092,0.00013344795,0.87843764,0.00032827893,0.11556321,0.0034748013,0.000084941],"about_ca_topic_score_codex":0.008066226,"about_ca_topic_score_gemma":0.006230119,"teacher_disagreement_score":0.052314896,"about_ca_system_score_codex":0.0019440702,"about_ca_system_score_gemma":0.0033760446,"threshold_uncertainty_score":0.27667087},"labels":[],"label_agreement":null},{"id":"W2765966791","doi":"10.1002/sim.7515","title":"A mechanistic nonlinear model for censored and mismeasured covariates in longitudinal models, with application in AIDS studies","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; HIV Legal Network; St. Paul's Hospital","funders":"","keywords":"Covariate; Censoring (clinical trials); Econometrics; Inference; Observational error; Computer science; Statistics; Mathematics; Artificial intelligence","score_opus":0.1472845397461962,"score_gpt":0.438050019936488,"score_spread":0.2907654801902918,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2765966791","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015098332,0.0005765319,0.9810176,0.0013800988,0.0000755975,0.00009045713,0.0004345001,0.00013474858,0.0011921722],"genre_scores_gemma":[0.64307255,0.0032573119,0.33165094,0.001354425,0.0005343599,0.0012954314,0.001499011,0.00015960823,0.017176423],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.995849,0.002739799,0.00015013527,0.00069452374,0.0003740783,0.0001923784],"domain_scores_gemma":[0.9798877,0.015670743,0.0021754864,0.0011940547,0.0007663831,0.00030563763],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014261661,0.0011620473,0.0013809123,0.0017157625,0.0011377551,0.0019417568,0.0036771162,0.002575042,0.003916624],"category_scores_gemma":[0.035910927,0.00093828724,0.002039733,0.0024297035,0.0023917968,0.0036019154,0.0022101672,0.0035748833,0.0006312877],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013561398,0.00010953444,0.013033066,0.0002493682,0.0002161053,0.00060812844,0.0007214098,0.33918306,0.0007708935,0.6039257,0.0032546595,0.03779257],"study_design_scores_gemma":[0.000048040718,0.00007514743,0.002145808,0.00006828379,0.00009365092,0.00027585635,0.00008541109,0.72517616,0.00021281731,0.26717827,0.0045811865,0.00005931564],"about_ca_topic_score_codex":0.007938186,"about_ca_topic_score_gemma":0.009036611,"teacher_disagreement_score":0.014261661,"about_ca_system_score_codex":0.0018109004,"about_ca_system_score_gemma":0.0022257196,"threshold_uncertainty_score":0.07542378},"labels":[],"label_agreement":null},{"id":"W2766219675","doi":"10.1002/sim.7527","title":"Collaborative targeted learning using regression shrinkage","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Université de Montréal","keywords":"Covariate; Causal inference; Propensity score matching; Estimator; Statistics; Regression; Regression analysis; Inference; Computer science; Stepwise regression; Mean squared error; Model selection; Outcome (game theory); Econometrics; Mathematics; Artificial intelligence","score_opus":0.11734741745116913,"score_gpt":0.4830962113690352,"score_spread":0.3657487939178661,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2766219675","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0041919677,0.00011239319,0.994827,0.00016529742,0.000015387788,0.000044293953,0.000027819131,0.00020231411,0.0004135144],"genre_scores_gemma":[0.42744938,0.0005461033,0.5663723,0.0005946813,0.00019626589,0.0007871857,0.00047583197,0.00018346153,0.0033947672],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.994111,0.0037989526,0.0002501636,0.00081649894,0.0007785208,0.0002449131],"domain_scores_gemma":[0.97129244,0.023069518,0.0013589837,0.0025465605,0.0014007158,0.000331811],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012519174,0.0012159772,0.002591243,0.0014622249,0.0006373856,0.001326297,0.0027924068,0.0018456973,0.0026737473],"category_scores_gemma":[0.045459352,0.0008906438,0.0017998462,0.0014900245,0.001661999,0.0021143458,0.0035215975,0.002635943,0.0006716912],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00027610004,0.00017440849,0.003977405,0.00024354381,0.0003035618,0.00021196334,0.00025902045,0.71141034,0.0016482292,0.08624688,0.0032996996,0.1919488],"study_design_scores_gemma":[0.000042657895,0.000050219256,0.00019703942,0.000018570672,0.000026018148,0.000026186874,0.0000097768125,0.96500236,0.00052780844,0.033476878,0.0006136841,0.000008824043],"about_ca_topic_score_codex":0.0030087996,"about_ca_topic_score_gemma":0.0031987578,"teacher_disagreement_score":0.012519174,"about_ca_system_score_codex":0.0011313069,"about_ca_system_score_gemma":0.0021582157,"threshold_uncertainty_score":0},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2766434410","doi":"10.1002/sim.7522","title":"Weighted estimation for confounded binary outcomes subject to misclassification","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Institute of Population and Public Health; University of Ottawa","funders":"","keywords":"Covariate; Estimator; Statistics; Logistic regression; Confounding; Computer science; Estimation; Inverse probability; Econometrics; Inverse probability weighting; Consistency (knowledge bases); Identification (biology); Mathematics; Artificial intelligence; Posterior probability","score_opus":0.19441443568362424,"score_gpt":0.500652691376323,"score_spread":0.3062382556926987,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2766434410","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0047178743,0.0002752085,0.99426335,0.00015432778,0.000041483592,0.00007511307,0.000075943564,0.00007945159,0.00031720893],"genre_scores_gemma":[0.20456441,0.0009835096,0.78989214,0.00045627652,0.00018588672,0.00095159066,0.00074988505,0.00011172391,0.0021045362],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9837286,0.011059581,0.0008953827,0.0021357762,0.0018089074,0.0003718163],"domain_scores_gemma":[0.9492394,0.03868092,0.003629277,0.0063361423,0.0018947155,0.00021950509],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03513706,0.0010486506,0.0019387688,0.0024950563,0.0004942969,0.0017052678,0.003452485,0.001986597,0.0031022027],"category_scores_gemma":[0.15084349,0.0008042077,0.0019817925,0.0026524856,0.0014185072,0.0027165636,0.0022942405,0.0022269546,0.00044533153],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00026810754,0.00020474396,0.017549343,0.00079173123,0.0013741428,0.00042475102,0.0007823051,0.13162714,0.0020566883,0.585344,0.0031434968,0.25643358],"study_design_scores_gemma":[0.000092324226,0.00013615156,0.0037021216,0.00026065382,0.00030761186,0.00028121623,0.00010349945,0.34536955,0.0015703975,0.6400019,0.008122146,0.00005237319],"about_ca_topic_score_codex":0.002268892,"about_ca_topic_score_gemma":0.001853879,"teacher_disagreement_score":0.03513706,"about_ca_system_score_codex":0.00082732196,"about_ca_system_score_gemma":0.0012872762,"threshold_uncertainty_score":0.18582475},"labels":[],"label_agreement":null},{"id":"W2768383839","doi":"10.1002/sim.7502","title":"A state transition framework for patient‐level modeling of engagement and retention in HIV care using longitudinal cohort data","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"HIV/AIDS Research and Interventions","field":"Medicine","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"National Institute of Allergy and Infectious Diseases; National Institute of Mental Health; Eunice Kennedy Shriver National Institute of Child Health and Human Development; Center for AIDS Research, University of Washington; National Institutes of Health","keywords":"Cascade; Disengagement theory; Health care; Computer science; Human immunodeficiency virus (HIV); Observational study; Econometrics; Medicine; Gerontology; Statistics; Mathematics; Family medicine; Engineering; Political science","score_opus":0.2299358649159377,"score_gpt":0.4522708781184758,"score_spread":0.22233501320253812,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2768383839","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.033041194,0.0006736925,0.9577189,0.0019658413,0.00017382082,0.0002634102,0.002687312,0.00047739703,0.0029984494],"genre_scores_gemma":[0.83677655,0.0016171127,0.13847588,0.00054466334,0.0002857834,0.0022972256,0.004390018,0.0001487279,0.015464016],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9977847,0.0012216993,0.000109969405,0.0004692352,0.00017063334,0.0002437243],"domain_scores_gemma":[0.99153894,0.006675663,0.0005430172,0.00025156472,0.0007222082,0.00026857594],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00807938,0.0013635961,0.0014261343,0.0018476943,0.0009932801,0.00217233,0.0029221172,0.0022800951,0.006181194],"category_scores_gemma":[0.01590796,0.0010312683,0.0023499026,0.001544944,0.0013513609,0.0018838492,0.0024314946,0.0035178228,0.0006953514],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000080943435,0.000079172896,0.0076436526,0.00007665457,0.0001487936,0.00014774004,0.00026948334,0.9097047,0.000214765,0.07225386,0.0015204618,0.007859809],"study_design_scores_gemma":[0.000014151127,0.000020355383,0.00048798954,0.000013198681,0.000024715688,0.000011020584,0.000032963977,0.9835955,0.00003058151,0.015058016,0.0006972524,0.0000142060935],"about_ca_topic_score_codex":0.070738114,"about_ca_topic_score_gemma":0.037546817,"teacher_disagreement_score":0.070738114,"about_ca_system_score_codex":0.0028593892,"about_ca_system_score_gemma":0.0032328195,"threshold_uncertainty_score":0.14065272},"labels":[],"label_agreement":null},{"id":"W2768417485","doi":"10.1002/sim.7539","title":"Exponential decay for binary time‐varying covariates in Cox models","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Guelph","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Covariate; Proportional hazards model; Statistics; Event (particle physics); Econometrics; Exponential growth; Binary data; Computer science; Mathematics; Binary number","score_opus":0.5055243770677599,"score_gpt":0.5754043878517637,"score_spread":0.06988001078400385,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2768417485","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0029902942,0.001239293,0.9934776,0.00056799443,0.000114849965,0.000092679664,0.0002117396,0.00017049,0.001135056],"genre_scores_gemma":[0.37215817,0.009735477,0.5880233,0.0019186817,0.0013563955,0.0031069838,0.0023730635,0.00070240715,0.020625368],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9859278,0.008963312,0.00064583006,0.001828839,0.0020363103,0.0005978796],"domain_scores_gemma":[0.92470855,0.06510936,0.00331117,0.0042192303,0.002263704,0.00038801777],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.040377334,0.0014176348,0.0018814524,0.0018163768,0.00067271077,0.002520104,0.0038653552,0.0027824382,0.0069668205],"category_scores_gemma":[0.12569362,0.0010051586,0.002730937,0.0025171954,0.0025513554,0.0049395883,0.0024047603,0.006506115,0.002057909],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003326113,0.00008703104,0.0061547356,0.0007630845,0.00026840394,0.0007180444,0.0006849694,0.19127905,0.0010066094,0.71902484,0.007507082,0.07217364],"study_design_scores_gemma":[0.00009971271,0.00018156109,0.0015234934,0.0002494208,0.00018869169,0.00059687725,0.00010987128,0.53980356,0.0005964883,0.44328457,0.013263783,0.00010197216],"about_ca_topic_score_codex":0.0036831428,"about_ca_topic_score_gemma":0.0024938877,"teacher_disagreement_score":0.040377334,"about_ca_system_score_codex":0.0018341804,"about_ca_system_score_gemma":0.00247032,"threshold_uncertainty_score":0.21353829},"labels":[],"label_agreement":null},{"id":"W2774445575","doi":"10.1002/sim.7555","title":"Bayesian inference for unidirectional misclassification of a binary response trait","year":2017,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Identifiability; Covariate; Computer science; Inference; Poisson distribution; Identification (biology); Binary data; Bayesian probability; Binary number; Statistics; Econometrics; Bayesian inference; Binary classification; Bayes' theorem; Machine learning; Artificial intelligence; Mathematics","score_opus":0.1210735242408162,"score_gpt":0.4597409341384469,"score_spread":0.3386674098976307,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2774445575","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.028872812,0.00032606896,0.9690048,0.00074811454,0.000051972493,0.00008034617,0.000108349006,0.00010474401,0.00070284895],"genre_scores_gemma":[0.6271313,0.0008394304,0.3671009,0.0009900165,0.00022315387,0.0006701166,0.0006103146,0.00010303971,0.0023316543],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96004456,0.029468434,0.0015960581,0.0059611653,0.0022525238,0.0006772484],"domain_scores_gemma":[0.7811842,0.18807313,0.011958734,0.014922787,0.0030541618,0.0008069668],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.090335384,0.0013044203,0.0026152509,0.0022285297,0.0014350347,0.0030012925,0.0037399195,0.0027965007,0.002274116],"category_scores_gemma":[0.2870932,0.0013508507,0.0019476764,0.0017631878,0.0044069467,0.0049870396,0.0037338766,0.005413296,0.0004780133],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006030508,0.00025108614,0.042619556,0.0007315519,0.0009994911,0.00057205366,0.0023836764,0.14644358,0.0018826657,0.6197959,0.0027140344,0.18100345],"study_design_scores_gemma":[0.00008108968,0.0000989811,0.0047192206,0.00023223492,0.00022288426,0.00025089874,0.00015915094,0.39781615,0.0014704022,0.59342563,0.0014576446,0.00006570689],"about_ca_topic_score_codex":0.0039612222,"about_ca_topic_score_gemma":0.00373733,"teacher_disagreement_score":0.090335384,"about_ca_system_score_codex":0.0021187016,"about_ca_system_score_gemma":0.0021288553,"threshold_uncertainty_score":0.47774488},"labels":[],"label_agreement":null},{"id":"W2784307041","doi":"10.1002/sim.7553","title":"Modeling clustering and treatment effect heterogeneity in parallel and stepped‐wedge cluster randomized trials","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":47,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Canada","funders":"National Health and Medical Research Council; Medical Research Council; National Institute for Health and Care Research","keywords":"Cluster analysis; Cluster (spacecraft); Computer science; Random effects model; Econometrics; Wedge (geometry); Statistics; Data mining; Mathematics; Medicine; Artificial intelligence; Meta-analysis","score_opus":0.1332079247993277,"score_gpt":0.4520500531089921,"score_spread":0.3188421283096644,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2784307041","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.016776472,0.0007667121,0.9769644,0.0016101344,0.00015137767,0.0015033328,0.00025162273,0.00021738326,0.0017585281],"genre_scores_gemma":[0.3816721,0.0009967551,0.6042442,0.0015525295,0.00016971707,0.007803386,0.00048616756,0.00012613577,0.0029490027],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.87187153,0.11169243,0.0039465753,0.005861364,0.0053986777,0.0012293897],"domain_scores_gemma":[0.7256604,0.23939161,0.014234611,0.015914598,0.0038047,0.0009940489],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.16432996,0.0017744015,0.004289493,0.0024385485,0.0010088673,0.002975023,0.006592457,0.0056446325,0.0059832335],"category_scores_gemma":[0.28702456,0.0019958757,0.005132027,0.0031810466,0.0047082095,0.0045033707,0.00399515,0.0060943323,0.00074813166],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0023369875,0.0002677782,0.0053563938,0.0015990123,0.0014706888,0.0007764974,0.0010431898,0.35277832,0.0006088744,0.57228124,0.0035121671,0.057968825],"study_design_scores_gemma":[0.0013607263,0.00055217603,0.0008784479,0.0003943764,0.000512552,0.00024831272,0.00011824726,0.42268834,0.00057110674,0.5692797,0.0033051101,0.00009096246],"about_ca_topic_score_codex":0.0026182886,"about_ca_topic_score_gemma":0.002167927,"teacher_disagreement_score":0.16432996,"about_ca_system_score_codex":0.0035397373,"about_ca_system_score_gemma":0.004114038,"threshold_uncertainty_score":0.8690703},"labels":[],"label_agreement":null},{"id":"W2785825193","doi":"10.1002/sim.7589","title":"Meta‐analysis of Gaussian individual patient data: Two‐stage or not two‐stage?","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":39,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada","keywords":"Stage (stratigraphy); Meta-analysis; Computer science; Flexibility (engineering); Econometrics; Random effects model; Outcome (game theory); Aggregate data; Sample size determination; Aggregate (composite); Range (aeronautics); Statistics; Mathematics; Medicine","score_opus":0.8323632235708937,"score_gpt":0.5893490509411681,"score_spread":0.24301417262972558,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2785825193","genre_codex":"review","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00412534,0.5121614,0.407842,0.05610751,0.010632165,0.0043619736,0.001671857,0.0007754337,0.00232229],"genre_scores_gemma":[0.16985579,0.17513357,0.54538006,0.07552425,0.012337452,0.015819993,0.001778139,0.0008798269,0.0032910027],"study_design_codex":"meta_analysis","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.68721765,0.26705012,0.019884834,0.011034405,0.013577043,0.0012359474],"domain_scores_gemma":[0.5563137,0.38512793,0.015750498,0.032097463,0.009498679,0.0012117488],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27721924,0.0034227415,0.02157163,0.0049860487,0.0013084612,0.0099956095,0.00761372,0.00848317,0.0044624913],"category_scores_gemma":[0.44678023,0.0034689493,0.02852739,0.007071104,0.0038371542,0.013061833,0.0041687237,0.0095251165,0.0011464467],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.005919388,0.00034796775,0.006186002,0.16903765,0.35260263,0.000716821,0.001838083,0.011421298,0.0013721954,0.10606517,0.046127595,0.29836515],"study_design_scores_gemma":[0.007890715,0.0014026499,0.0049209665,0.050198305,0.20961821,0.0010215706,0.00040035817,0.033756223,0.0019704853,0.6177817,0.07002862,0.0010103356],"about_ca_topic_score_codex":0.003684011,"about_ca_topic_score_gemma":0.006178565,"teacher_disagreement_score":0.72278076,"about_ca_system_score_codex":0.0043283445,"about_ca_system_score_gemma":0.009191948,"threshold_uncertainty_score":0.89131737},"labels":[],"label_agreement":null},{"id":"W2790095428","doi":"10.1002/sim.7611","title":"Discrimination surfaces with application to region‐specific brain asymmetry analysis","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Morphological variations and asymmetry","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Fundação para a Ciência e a Tecnologia; University of British Columbia; University of South Carolina; National Science Foundation","keywords":"Estimator; Statistic; Asymmetry; Brain asymmetry; Mathematics; Statistics; Confidence interval; Pattern recognition (psychology); Psychology; Computer science; Artificial intelligence; Cognitive psychology; Physics","score_opus":0.04038446706239159,"score_gpt":0.3465978518653931,"score_spread":0.30621338480300153,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2790095428","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.038715616,0.0001641138,0.9594985,0.000097639626,0.00001903384,0.000046515048,0.00011586183,0.00044337864,0.00089927454],"genre_scores_gemma":[0.6669688,0.00025013342,0.3309373,0.00007405927,0.000106652056,0.0001522461,0.00043075244,0.00028245125,0.0007976193],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99881256,0.00043667125,0.00008230302,0.0001842048,0.00039438333,0.000089930196],"domain_scores_gemma":[0.9899149,0.0072579486,0.00074982556,0.00087900995,0.0009828395,0.00021546736],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0033292784,0.00049824774,0.0007265245,0.0032080547,0.00035979922,0.0012077529,0.0006587257,0.00072493736,0.001547883],"category_scores_gemma":[0.021528607,0.0002473489,0.0007375586,0.0016273048,0.0013387384,0.0009716256,0.0016160775,0.0010262331,0.0003220614],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00044927778,0.00013965696,0.023848338,0.00032017007,0.00017598803,0.0006385245,0.00056956644,0.23187454,0.057136033,0.17114034,0.0029485677,0.510759],"study_design_scores_gemma":[0.00002116754,0.00011459303,0.011415858,0.000025985591,0.000028648583,0.00061494904,0.000088675275,0.8707053,0.009109817,0.10512425,0.0026935684,0.000057140584],"about_ca_topic_score_codex":0.0006570256,"about_ca_topic_score_gemma":0.00030755097,"teacher_disagreement_score":0.0033292784,"about_ca_system_score_codex":0.00047062134,"about_ca_system_score_gemma":0.00046894499,"threshold_uncertainty_score":0.017607093},"labels":[],"label_agreement":null},{"id":"W2791924225","doi":"10.1002/sim.7615","title":"Assessing the performance of the generalized propensity score for estimating the effect of quantitative or continuous exposures on binary outcomes","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":85,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Work & Health; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Covariate; Statistics; Confounding; Mathematics; Ordinary least squares; Medicine","score_opus":0.215693518484249,"score_gpt":0.4913806844402001,"score_spread":0.2756871659559511,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2791924225","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.102251515,0.0011898255,0.8924076,0.00062188454,0.000069698355,0.0005410864,0.000581627,0.00028603972,0.0020507684],"genre_scores_gemma":[0.49070236,0.0013095593,0.50521857,0.00023675297,0.0000951063,0.0009372741,0.00083358947,0.00019520268,0.00047159696],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9321754,0.059082,0.002059848,0.0026273031,0.003679029,0.00037659853],"domain_scores_gemma":[0.67107356,0.29337218,0.0119911805,0.018121459,0.004803966,0.0006376775],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.100310445,0.0012252225,0.0014141351,0.0035023615,0.00062723015,0.0016515984,0.0013976856,0.0016302787,0.002778472],"category_scores_gemma":[0.32012492,0.00052581384,0.0026960429,0.004267148,0.0019919209,0.0023367738,0.0022196209,0.0015477283,0.00031175508],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016258927,0.00047164888,0.19902515,0.002091007,0.007928977,0.00043297268,0.001756356,0.38145775,0.0025026938,0.12017827,0.004563851,0.27796546],"study_design_scores_gemma":[0.00060255383,0.0022450192,0.087037854,0.00075415114,0.0013726538,0.00063427695,0.00052027224,0.7284478,0.0033573029,0.16633888,0.008441861,0.0002473652],"about_ca_topic_score_codex":0.005430653,"about_ca_topic_score_gemma":0.0039596516,"teacher_disagreement_score":0.89968956,"about_ca_system_score_codex":0.0010992242,"about_ca_system_score_gemma":0.0029491223,"threshold_uncertainty_score":0.5304987},"labels":[],"label_agreement":null},{"id":"W2794650963","doi":"10.1002/sim.7641","title":"Design of cancer trials based on progression‐free survival with intermittent assessment","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Censoring (clinical trials); Sample size determination; Medicine; Clinical trial; Randomized controlled trial; Clinical endpoint; Progression-free survival; Survival analysis; Confidence interval; Cancer; Clinical study design; Research design; Intensive care medicine; Oncology; Overall survival; Statistics; Internal medicine; Mathematics","score_opus":0.6283440379913406,"score_gpt":0.6537779914408499,"score_spread":0.025433953449509272,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2794650963","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06339659,0.003464429,0.8596599,0.0025559487,0.0013588737,0.064632304,0.00045787022,0.00035740813,0.0041166386],"genre_scores_gemma":[0.4159437,0.0023614457,0.44806948,0.0015509428,0.00042181739,0.12973393,0.00031197318,0.000041393512,0.0015653028],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.81374633,0.17153502,0.0049462034,0.004021693,0.0048909197,0.00085977296],"domain_scores_gemma":[0.88188756,0.09374291,0.01339691,0.0072210594,0.0022645555,0.0014869891],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.13034278,0.0016219444,0.0040101428,0.0019136956,0.00051987934,0.002393748,0.0019186492,0.00351856,0.003467031],"category_scores_gemma":[0.17728399,0.000850161,0.002417351,0.0013248124,0.0027820691,0.0027770188,0.00207247,0.0032006085,0.00048800194],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.08905823,0.004530494,0.015996065,0.010631212,0.007074427,0.00062454015,0.0009949167,0.17660901,0.0067326142,0.35431406,0.004628681,0.3288057],"study_design_scores_gemma":[0.0666351,0.057984192,0.005941248,0.0018410729,0.004152314,0.00037122957,0.00021576474,0.42077124,0.0057258857,0.4193903,0.016696317,0.0002753584],"about_ca_topic_score_codex":0.00014612275,"about_ca_topic_score_gemma":0.00018696145,"teacher_disagreement_score":0.8696572,"about_ca_system_score_codex":0.0012649063,"about_ca_system_score_gemma":0.0030513448,"threshold_uncertainty_score":0.68932676},"labels":[],"label_agreement":null},{"id":"W2799472749","doi":"10.1002/sim.7658","title":"Exposure, hazard, and survival analysis of diffusion on social networks","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Complex Network Analysis Techniques","field":"Physics and Astronomy","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Yale Center for Clinical Investigation, Yale School of Medicine; Eunice Kennedy Shriver National Institute of Child Health and Human Development; National Center for Advancing Translational Sciences; National Institute of Mental Health; Canadian Institutes of Health Research; National Institute on Aging; National Institutes of Health; Yale University; Bill and Melinda Gates Foundation","keywords":"Social network (sociolinguistics); Psychological intervention; Realization (probability); Diffusion; Computer science; Population; Social network analysis; Relevance (law); Diffusion of innovations; Intervention (counseling); Process (computing); Econometrics; Actuarial science; Data science; Psychology; Statistics; Business; Sociology; Economics; Marketing; Demography; Mathematics; Political science","score_opus":0.0185726490180078,"score_gpt":0.32509458966867677,"score_spread":0.30652194065066896,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2799472749","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19598553,0.00079991564,0.79827005,0.0014160458,0.00007917254,0.00021582186,0.0006726509,0.00028476943,0.002275975],"genre_scores_gemma":[0.9483206,0.001101765,0.046144668,0.00012344366,0.00014347397,0.00042781874,0.00055402395,0.0000497529,0.003134465],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9975177,0.0016058917,0.00008710233,0.0003902535,0.00024009023,0.00015907465],"domain_scores_gemma":[0.9747482,0.020167805,0.0026605062,0.0011921372,0.0007183767,0.0005129822],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00889286,0.0007939273,0.0007786649,0.0033708694,0.00063100137,0.0011365279,0.0012153416,0.0012458664,0.0025825053],"category_scores_gemma":[0.030723037,0.0003531072,0.0014136342,0.001437524,0.002092927,0.0028157278,0.0015654095,0.0015893369,0.00029634053],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017945761,0.00013005087,0.041026745,0.00021239334,0.0003419444,0.0003769975,0.0011866448,0.43404755,0.0015696355,0.47411373,0.0016868759,0.04512797],"study_design_scores_gemma":[0.000018572231,0.00006884403,0.006276779,0.00003129762,0.000051749634,0.00012529641,0.00013757392,0.850018,0.00025075645,0.14168267,0.0013035988,0.000034783076],"about_ca_topic_score_codex":0.005515624,"about_ca_topic_score_gemma":0.0018409324,"teacher_disagreement_score":0.00889286,"about_ca_system_score_codex":0.00143585,"about_ca_system_score_gemma":0.00068528,"threshold_uncertainty_score":0.04703051},"labels":[],"label_agreement":null},{"id":"W2800193779","doi":"10.1002/sim.7675","title":"Dynamic thresholds and a summary ROC curve: Assessing prognostic accuracy of longitudinal markers","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Acute Ischemic Stroke Management","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada","keywords":"Receiver operating characteristic; Computer science; Statistics; Mathematics; Machine learning","score_opus":0.020997980025216547,"score_gpt":0.35293424950437674,"score_spread":0.3319362694791602,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2800193779","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.16217762,0.00855524,0.8206948,0.0017293914,0.00026897277,0.00035740493,0.002030707,0.0012991928,0.0028866157],"genre_scores_gemma":[0.8298152,0.0017493132,0.16468339,0.0004004193,0.00039717363,0.00042570848,0.0016383679,0.00022435667,0.00066605565],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9721361,0.019434473,0.0019802677,0.0029351441,0.0029905324,0.000523567],"domain_scores_gemma":[0.7984022,0.15953188,0.016605925,0.015480395,0.008457548,0.0015221402],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04102547,0.0016599748,0.002027692,0.0069894637,0.00039394494,0.0039091045,0.0019053383,0.0022979698,0.0010612609],"category_scores_gemma":[0.20332246,0.0005551147,0.002235044,0.004445211,0.0019264311,0.004407196,0.0017369278,0.0018465847,0.00050468725],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002347931,0.00027827447,0.26437682,0.0013193729,0.0037985353,0.0006536744,0.0011959468,0.2375974,0.0051712766,0.04834987,0.008946593,0.42596427],"study_design_scores_gemma":[0.00011974322,0.0026776805,0.08230238,0.00047447335,0.0008781241,0.0026051637,0.00053971546,0.79269713,0.0045107324,0.10425298,0.008546391,0.00039546803],"about_ca_topic_score_codex":0.0011878291,"about_ca_topic_score_gemma":0.00052107946,"teacher_disagreement_score":0.95897454,"about_ca_system_score_codex":0.0010949011,"about_ca_system_score_gemma":0.0010762617,"threshold_uncertainty_score":0.21696603},"labels":[],"label_agreement":null},{"id":"W2801071792","doi":"10.1002/sim.7646","title":"Tracing studies in cohorts with attrition: Selection models for efficient sampling","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Liver Disease Diagnosis and Treatment","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Cohort; Estimator; Statistics; Sampling (signal processing); Attrition; Tracing; Selection (genetic algorithm); Simple random sample; Computer science; Medicine; Cohort study; Econometrics; Mathematics; Machine learning; Environmental health","score_opus":0.11554361333381112,"score_gpt":0.40594295539019143,"score_spread":0.2903993420563803,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2801071792","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.004280933,0.000775707,0.9925794,0.00084712123,0.00008231062,0.00044815126,0.0002108754,0.00018094225,0.0005945635],"genre_scores_gemma":[0.17934056,0.003521896,0.79942673,0.0010990789,0.00058470346,0.00658925,0.001389521,0.00027035267,0.007777863],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9346287,0.057450008,0.0014335344,0.0030259101,0.002524466,0.00093733106],"domain_scores_gemma":[0.737497,0.23278053,0.009765485,0.014462296,0.004102172,0.0013925289],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.12977345,0.002312404,0.004623475,0.0032662675,0.0017133548,0.0036106708,0.0063085207,0.0050339997,0.0056498004],"category_scores_gemma":[0.22036736,0.0024014877,0.0035235928,0.004610242,0.004468435,0.0050379084,0.0051414794,0.006281217,0.0013052095],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007066893,0.00022394366,0.015078015,0.0007917149,0.0009085608,0.0008737165,0.0015283849,0.29185584,0.00070876133,0.5777027,0.0054831575,0.104138635],"study_design_scores_gemma":[0.00038255772,0.00022912314,0.0012679051,0.00024013205,0.00022194038,0.00019753289,0.000098318786,0.6659604,0.0003213901,0.32559612,0.0054120114,0.00007260913],"about_ca_topic_score_codex":0.008534125,"about_ca_topic_score_gemma":0.007827286,"teacher_disagreement_score":0.87022656,"about_ca_system_score_codex":0.0029458979,"about_ca_system_score_gemma":0.0037405274,"threshold_uncertainty_score":0.6863158},"labels":[],"label_agreement":null},{"id":"W2801251405","doi":"10.1002/sim.7680","title":"Time series analysis of fMRI data: Spatial modelling and Bayesian computation","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"National Institute of Neurological Disorders and Stroke; Natural Sciences and Engineering Research Council of Canada; National Institutes of Health","keywords":"Computer science; Computation; Approximate Bayesian computation; Bayesian probability; Bayes' theorem; Statistical inference; Algorithm; Inference; Bayesian inference; Statistical parametric mapping; Parametric statistics; Artificial intelligence; Machine learning; Statistics; Mathematics","score_opus":0.07775302964608646,"score_gpt":0.4042363175240131,"score_spread":0.3264832878779267,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2801251405","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0018245026,0.00037423384,0.99681807,0.0002949865,0.00001452315,0.000013894877,0.000039764618,0.00012684174,0.0004931798],"genre_scores_gemma":[0.17896879,0.0030942112,0.8143055,0.00033651898,0.0001871323,0.0002829375,0.00039669216,0.0002705154,0.002157648],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9983272,0.00079880666,0.000092482784,0.00024590586,0.00047046968,0.00006511794],"domain_scores_gemma":[0.99164104,0.0069227265,0.000555041,0.0003949227,0.00038828028,0.000097989905],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0039559784,0.0007332933,0.0012520392,0.0014801592,0.0005380902,0.002197801,0.0019400045,0.0019151843,0.0019167254],"category_scores_gemma":[0.021325225,0.00102496,0.0012992647,0.002093979,0.0017884022,0.003012197,0.0016387634,0.0024568522,0.0006027047],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00004126248,0.000035414643,0.0015165083,0.00022450191,0.000102219645,0.000102772115,0.0001770339,0.6934745,0.0020738703,0.22493665,0.002085742,0.07522952],"study_design_scores_gemma":[0.0000030357542,0.000005586323,0.00018431457,0.000018574943,0.000006941545,0.000029825718,0.000009214402,0.9402241,0.00024817875,0.058391698,0.0008684739,0.000010128614],"about_ca_topic_score_codex":0.011621141,"about_ca_topic_score_gemma":0.00861286,"teacher_disagreement_score":0.011621141,"about_ca_system_score_codex":0.0015702372,"about_ca_system_score_gemma":0.0017440801,"threshold_uncertainty_score":0.023106992},"labels":[],"label_agreement":null},{"id":"W2801930703","doi":"10.1002/sim.7655","title":"Quantile regression with nominated samples: An application to a bone mineral density study","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Manitoba","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Quantile; Statistics; Quantile regression; Covariate; Sampling (signal processing); Mathematics; Simple random sample; Maxima and minima; Conditional probability distribution; Maxima; Econometrics; Population; Computer science","score_opus":0.09116800202283325,"score_gpt":0.4430228777902052,"score_spread":0.35185487576737196,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2801930703","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008423635,0.0003107559,0.99016577,0.0002642991,0.000038103284,0.000105141145,0.000039959366,0.00006851995,0.0005837622],"genre_scores_gemma":[0.20672973,0.0007722161,0.7906725,0.0002335985,0.00011740852,0.00043902034,0.00010803201,0.000071354625,0.0008561591],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9723412,0.024869334,0.0003267819,0.0009381232,0.0013112697,0.00021329158],"domain_scores_gemma":[0.9333612,0.058521144,0.0023008124,0.0033732764,0.0020481225,0.00039538406],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02987732,0.0008171686,0.0012264076,0.0012039274,0.00094860233,0.0012958206,0.0020566091,0.0014273089,0.002732921],"category_scores_gemma":[0.084091775,0.00055020006,0.0014306342,0.0020932795,0.001777728,0.001240809,0.0022786723,0.002518272,0.00032779924],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011928559,0.00045426708,0.02373977,0.00085245207,0.00038084626,0.0013256671,0.0012503878,0.39200652,0.0051109586,0.2826241,0.003773123,0.28728902],"study_design_scores_gemma":[0.00030980463,0.0006189357,0.0066210474,0.00016013486,0.00012570713,0.00044043057,0.00029127413,0.85418147,0.0029867087,0.12343907,0.0107357735,0.000089641646],"about_ca_topic_score_codex":0.0045561744,"about_ca_topic_score_gemma":0.0047849473,"teacher_disagreement_score":0.02987732,"about_ca_system_score_codex":0.0010085726,"about_ca_system_score_gemma":0.0014163043,"threshold_uncertainty_score":0.15800828},"labels":[],"label_agreement":null},{"id":"W2802284801","doi":"10.1002/sim.7628","title":"Targeted maximum likelihood estimation for a binary treatment: A tutorial","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":134,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"Instituto de Salud Carlos III; Canadian Institutes of Health Research; Cancer Research UK","keywords":"Computer science; Outcome (game theory); Nonparametric statistics; Contrast (vision); Propensity score matching; Code (set theory); Parametric statistics; Binary number; Estimation; Specification; Statistics; Machine learning; Artificial intelligence; Programming language; Mathematics","score_opus":0.08412398744132396,"score_gpt":0.43726232599602827,"score_spread":0.35313833855470433,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2802284801","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0004310138,0.008905142,0.95930916,0.0026118972,0.0009918371,0.00025526687,0.0021145435,0.0053397506,0.020041471],"genre_scores_gemma":[0.0066937697,0.0157833,0.9348167,0.0026146304,0.0019809469,0.0011891208,0.0036618137,0.0036382878,0.02962148],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99892265,0.00044853982,0.000101570455,0.00018848095,0.0003013434,0.000037375805],"domain_scores_gemma":[0.99437135,0.004674042,0.00015952776,0.00019946524,0.00049408624,0.000101522965],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0025557468,0.0022686492,0.0010925521,0.001695089,0.0003664037,0.0015554589,0.001314588,0.001973995,0.07930793],"category_scores_gemma":[0.014318027,0.0010379327,0.0013215591,0.0015640566,0.00056420907,0.0025576404,0.0013400046,0.0036088186,0.04035589],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00005726251,0.00016190844,0.00047519666,0.0025179328,0.00010222867,0.0005548483,0.0002888383,0.007934279,0.0022707877,0.11432061,0.35977745,0.5115387],"study_design_scores_gemma":[0.00005677262,0.00007028812,0.0006696883,0.0008946168,0.00003918998,0.00089690165,0.000056273413,0.025923692,0.0016735424,0.21909972,0.7505281,0.000091265356],"about_ca_topic_score_codex":0.0014243354,"about_ca_topic_score_gemma":0.001966785,"teacher_disagreement_score":0.07930793,"about_ca_system_score_codex":0.000808661,"about_ca_system_score_gemma":0.0011522713,"threshold_uncertainty_score":0.26531148},"labels":[],"label_agreement":null},{"id":"W2804840181","doi":"10.1002/sim.7805","title":"Nonparametric analysis of dependently interval‐censored failure time data","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Nonparametric statistics; Estimator; Statistics; Confidence interval; Parametric statistics; Weighting; Inverse probability weighting; Econometrics; Interval (graph theory); Accelerated failure time model; Observational study; Mathematics; Computer science; Covariate; Medicine","score_opus":0.10310703289422077,"score_gpt":0.4403236091353544,"score_spread":0.33721657624113366,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2804840181","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.059601612,0.00023459762,0.9392998,0.0001313517,0.000021011321,0.000055993223,0.00015110907,0.0001187617,0.0003856674],"genre_scores_gemma":[0.81089365,0.0006392994,0.18567504,0.00008067806,0.000083672814,0.00038011777,0.0007624664,0.00007620175,0.001408767],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9941772,0.0039914865,0.00022930482,0.0005955871,0.0008055695,0.00020083618],"domain_scores_gemma":[0.93567085,0.0517519,0.005060989,0.0056900517,0.0014807157,0.00034562484],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01635149,0.0004458799,0.0009890873,0.0012944482,0.00029148257,0.00093280536,0.002136389,0.00071116764,0.0012598094],"category_scores_gemma":[0.07362329,0.00031255325,0.0010309126,0.0012470868,0.0014037357,0.0015680006,0.0012367424,0.0015700968,0.00015167784],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00040998266,0.00029612347,0.056000173,0.0005347712,0.00077320467,0.000671655,0.0011398615,0.36316276,0.0040688976,0.30656955,0.0021053753,0.26426768],"study_design_scores_gemma":[0.000037285285,0.00019859493,0.019135043,0.00005447337,0.000064128595,0.00018297255,0.00014756,0.85859245,0.0012052384,0.1184246,0.0019083003,0.00004944277],"about_ca_topic_score_codex":0.0020002925,"about_ca_topic_score_gemma":0.0014450233,"teacher_disagreement_score":0.01635149,"about_ca_system_score_codex":0.0005364681,"about_ca_system_score_gemma":0.001162654,"threshold_uncertainty_score":0.08647597},"labels":[],"label_agreement":null},{"id":"W2805176315","doi":"10.1002/sim.7837","title":"A single‐index threshold Cox proportional hazard model for identifying a treatment‐sensitive subset based on multiple biomarkers","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University; Canadian Cancer Society","funders":"Natural Sciences and Engineering Research Council of Canada; National Natural Science Foundation of China","keywords":"Proportional hazards model; Estimator; Statistics; Hazard; Hazard ratio; Mathematics; Likelihood function; Computer science; Medicine; Oncology; Maximum likelihood; Biology; Confidence interval","score_opus":0.5433031768834886,"score_gpt":0.5514168099636382,"score_spread":0.008113633080149563,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2805176315","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008830747,0.00036131413,0.9897087,0.00035238636,0.000039933497,0.0001372691,0.00015035571,0.000099256875,0.00032005514],"genre_scores_gemma":[0.46684486,0.0015049393,0.5249485,0.0006109865,0.00037746466,0.0015858434,0.0009981046,0.00008163121,0.003047758],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99734575,0.0013384949,0.00013001698,0.00051320624,0.00053902075,0.0001334751],"domain_scores_gemma":[0.994495,0.0040109707,0.00056020543,0.00044792978,0.00032377554,0.00016216429],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010003341,0.0010059365,0.0019045525,0.0014050965,0.00033516827,0.0013272674,0.0027814116,0.0011796262,0.0015841826],"category_scores_gemma":[0.017822826,0.00058539346,0.0013828705,0.0016137596,0.0009545211,0.002262625,0.0015052676,0.0021250877,0.00038586618],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011424101,0.00036447277,0.029866477,0.00094584835,0.0008413705,0.0008203868,0.0005574636,0.46700504,0.0076023266,0.16323608,0.0051490176,0.32246915],"study_design_scores_gemma":[0.00015529848,0.00044133718,0.00242749,0.00004856818,0.00026853394,0.00042203575,0.00004507765,0.914054,0.0014125482,0.07727997,0.0033824397,0.00006266504],"about_ca_topic_score_codex":0.00092193414,"about_ca_topic_score_gemma":0.0007473329,"teacher_disagreement_score":0.010003341,"about_ca_system_score_codex":0.00066467974,"about_ca_system_score_gemma":0.0019402954,"threshold_uncertainty_score":0.052903354},"labels":[],"label_agreement":null},{"id":"W2807223243","doi":"10.1002/sim.7689","title":"Regulatory assessment of drug dissolution profiles comparability via maximum deviation","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Comparability; Bioequivalence; Statistics; Standard deviation; Mathematics; Dissolution testing; Benchmark (surveying); Similarity (geometry); Computer science; Econometrics; Medicine; Drug; Pharmacology; Pharmacokinetics","score_opus":0.3097714492522006,"score_gpt":0.5680756412596896,"score_spread":0.25830419200748894,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2807223243","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07206694,0.0018659943,0.9157515,0.0005377293,0.00014792725,0.00069375103,0.0006584562,0.00066658075,0.0076110777],"genre_scores_gemma":[0.79245234,0.0006171424,0.20382556,0.00024278006,0.00012783548,0.0009182573,0.00091798155,0.0001715601,0.0007265663],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.88638765,0.07126547,0.00604688,0.011332096,0.023916971,0.0010508574],"domain_scores_gemma":[0.7207277,0.23515609,0.015198959,0.017185528,0.01120392,0.0005277941],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09060225,0.0013875543,0.0022800278,0.003822337,0.000765429,0.0035557495,0.00299294,0.0028083152,0.0025069294],"category_scores_gemma":[0.23530045,0.0005697284,0.003719667,0.002889687,0.003982081,0.003916953,0.0036831237,0.0028164096,0.00045327298],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0034613116,0.0007987717,0.07458629,0.0028012996,0.0030781871,0.00076675933,0.0018558346,0.2901181,0.015432845,0.13140076,0.003776174,0.4719237],"study_design_scores_gemma":[0.00031441925,0.0034298974,0.045744188,0.00041952237,0.00063830125,0.00092008384,0.0004715007,0.7972219,0.03353724,0.103933096,0.01298053,0.00038932322],"about_ca_topic_score_codex":0.0014674615,"about_ca_topic_score_gemma":0.0005414893,"teacher_disagreement_score":0.09060225,"about_ca_system_score_codex":0.0017263785,"about_ca_system_score_gemma":0.0020868499,"threshold_uncertainty_score":0.4791562},"labels":[],"label_agreement":null},{"id":"W2808396915","doi":"10.1002/sim.7841","title":"Tweedie family of generalized linear models with distribution‐free random effects for skewed longitudinal data","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of New Brunswick","funders":"Division of Mathematical Sciences; Natural Sciences and Engineering Research Council of Canada","keywords":"Random effects model; Skewness; Generalized linear mixed model; Covariance; Mathematics; Econometrics; Linear model; Mixed model; Statistics; Flexibility (engineering); Population; Computer science; Medicine","score_opus":0.13676609651227098,"score_gpt":0.4275039299316327,"score_spread":0.29073783341936177,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2808396915","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00035877712,0.00036286647,0.99829036,0.0002648261,0.00008790661,0.00011599523,0.00018081233,0.00013226242,0.00020612408],"genre_scores_gemma":[0.023298493,0.0029448404,0.963037,0.00061170786,0.0003687222,0.0035895468,0.0008667588,0.0002837972,0.0049991244],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97935355,0.01598406,0.00086812535,0.0017921008,0.0016686718,0.00033347888],"domain_scores_gemma":[0.9596441,0.03470002,0.0014734016,0.0025617841,0.0013243988,0.00029623427],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03395786,0.0026563662,0.0035997066,0.004015729,0.0010544083,0.0032669895,0.0072798464,0.0033644952,0.009507601],"category_scores_gemma":[0.08714507,0.0023067663,0.0047259596,0.0042938064,0.002749597,0.0050551393,0.0033847305,0.009154061,0.0032042498],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001436248,0.000103553495,0.0019482863,0.0008092418,0.0010022679,0.00072710303,0.00059280195,0.066208586,0.00070711126,0.68706423,0.009696996,0.23099625],"study_design_scores_gemma":[0.00014418695,0.00014637143,0.00095856504,0.00029920173,0.00025788028,0.00046505142,0.00008199175,0.38133034,0.0005043461,0.5840985,0.031577013,0.00013657141],"about_ca_topic_score_codex":0.0043415637,"about_ca_topic_score_gemma":0.004823979,"teacher_disagreement_score":0.03395786,"about_ca_system_score_codex":0.0016668076,"about_ca_system_score_gemma":0.0035663552,"threshold_uncertainty_score":0.17958844},"labels":[],"label_agreement":null},{"id":"W2808447433","doi":"10.1002/sim.7845","title":"Estimation for zero‐inflated beta‐binomial regression model with missing response data","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Windsor","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Overdispersion; Statistics; Missing data; Mathematics; Negative binomial distribution; Expectation–maximization algorithm; Count data; Zero-inflated model; Quasi-likelihood; Mean squared error; Regression analysis; Econometrics; Maximum likelihood; Poisson regression; Poisson distribution","score_opus":0.16683890694722295,"score_gpt":0.47400700736848506,"score_spread":0.3071681004212621,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2808447433","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0068718824,0.00021177852,0.99246687,0.00010911284,0.000011880326,0.00002497245,0.000054814653,0.00008554675,0.00016317941],"genre_scores_gemma":[0.30371454,0.001738676,0.6899432,0.00034274915,0.00014466193,0.0005860222,0.0009827184,0.00015338151,0.0023940634],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9873018,0.009617415,0.00040916816,0.0013288516,0.0010049111,0.00033791887],"domain_scores_gemma":[0.95092493,0.04217597,0.002819181,0.0024350695,0.0013875501,0.00025733974],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03082645,0.0010551277,0.0024919228,0.0014319848,0.00055037916,0.0014726593,0.0036334742,0.002267523,0.0017882055],"category_scores_gemma":[0.064352,0.0009712291,0.0018033131,0.0019363251,0.0016874397,0.0030448183,0.0018257804,0.0030606557,0.0006395287],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00047482894,0.00016933438,0.01648637,0.0010437901,0.00070133153,0.0010625257,0.00082541583,0.55818945,0.005317956,0.23384039,0.0028851514,0.17900345],"study_design_scores_gemma":[0.000042727712,0.000103041326,0.0018323244,0.00008038531,0.00008244288,0.0003610076,0.00006771549,0.8940758,0.0012355187,0.1005554,0.0015155016,0.000048126494],"about_ca_topic_score_codex":0.0021997772,"about_ca_topic_score_gemma":0.0014759303,"teacher_disagreement_score":0.03082645,"about_ca_system_score_codex":0.0007285209,"about_ca_system_score_gemma":0.0012359739,"threshold_uncertainty_score":0.16302782},"labels":[],"label_agreement":null},{"id":"W2808494377","doi":"10.1002/sim.7835","title":"A Bayesian finite mixture of bivariate regression model for causal mediation analyses","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Pregnancy and Medication Impact","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Hôpital du Sacré-Cœur de Montréal; Université de Montréal; Centre Hospitalier Universitaire Sainte-Justine; Université du Québec à Montréal","funders":"Natural Sciences and Engineering Research Council of Canada; Public Health Agency","keywords":"Categorical variable; Statistics; Mixture model; Bivariate analysis; Mediation; Econometrics; Bayesian probability; Regression analysis; Outcome (game theory); Confounding; Identifiability; Mathematics","score_opus":0.07185331397151307,"score_gpt":0.4371229684999162,"score_spread":0.36526965452840315,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2808494377","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005062186,0.000428283,0.99197245,0.0006372294,0.00004692936,0.000112152906,0.0003821734,0.0002573184,0.0011012525],"genre_scores_gemma":[0.31778243,0.0021224136,0.66441005,0.00086155743,0.00027853632,0.0023551246,0.0018985653,0.00026118968,0.010030157],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98841184,0.008541616,0.0003347694,0.0011924403,0.0009856564,0.0005337143],"domain_scores_gemma":[0.9787596,0.017731734,0.0009489094,0.0009297695,0.0013389562,0.00029092675],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.019889988,0.001500237,0.0032541675,0.0027971829,0.0011285492,0.0024755711,0.005873519,0.0027688597,0.0075764484],"category_scores_gemma":[0.044734146,0.0013835436,0.003577065,0.0034635768,0.0024315407,0.0027477448,0.002777633,0.0042237267,0.0013657146],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00033167243,0.00012867813,0.0055722655,0.00026019837,0.0005900234,0.00042319094,0.00075779227,0.3010512,0.0008470222,0.6318765,0.0036725418,0.054488897],"study_design_scores_gemma":[0.00013970378,0.000082962215,0.0013498183,0.00009086045,0.000267529,0.00017146062,0.00007921679,0.7884253,0.00021467764,0.2043193,0.0047760564,0.000083138126],"about_ca_topic_score_codex":0.045740608,"about_ca_topic_score_gemma":0.031295743,"teacher_disagreement_score":0.045740608,"about_ca_system_score_codex":0.0033980978,"about_ca_system_score_gemma":0.003967541,"threshold_uncertainty_score":0.10518962},"labels":[],"label_agreement":null},{"id":"W2883076904","doi":"10.1002/sim.7896","title":"Analyzing differences between microbiome communities using mixture distributions","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Gut microbiota and health","field":"Biochemistry, Genetics and Molecular Biology","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto; University Health Network","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Statistics; Mixture model; Poisson distribution; Mixture distribution; Mathematics; Count data; Truncation (statistics); Pairwise comparison; Sample size determination; Population; Computer science; Algorithm; Probability density function","score_opus":0.03552064600994028,"score_gpt":0.34900303886493295,"score_spread":0.3134823928549927,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2883076904","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.027087811,0.00019449701,0.97170705,0.00007087922,0.000018094992,0.000066192144,0.00013255399,0.0004495926,0.0002733345],"genre_scores_gemma":[0.35742778,0.00039982572,0.63914317,0.00015396775,0.00010609938,0.0004195372,0.0012451067,0.00031489876,0.00078962336],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99532,0.0020854871,0.00021251274,0.0012230051,0.000951143,0.00020784742],"domain_scores_gemma":[0.972013,0.022383122,0.0019333633,0.002289615,0.000986924,0.00039410318],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010381851,0.001108456,0.001674617,0.0051865038,0.0010473487,0.0022786919,0.00184063,0.0014768154,0.0018415047],"category_scores_gemma":[0.03904068,0.0007721743,0.0026663267,0.0022291723,0.001880439,0.0037871625,0.0028607312,0.00229496,0.0006192349],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015489805,0.00044351915,0.10870555,0.0006660996,0.00163889,0.00047552175,0.0014355995,0.347685,0.026381861,0.083440475,0.002095045,0.42548355],"study_design_scores_gemma":[0.000062821324,0.0001852775,0.011803244,0.00006003521,0.000118314405,0.0003447077,0.00017137443,0.90220284,0.004957683,0.07723739,0.0027372309,0.00011906441],"about_ca_topic_score_codex":0.0023316504,"about_ca_topic_score_gemma":0.0017770173,"teacher_disagreement_score":0.010381851,"about_ca_system_score_codex":0.00080107455,"about_ca_system_score_gemma":0.0009944158,"threshold_uncertainty_score":0.054905117},"labels":[],"label_agreement":null},{"id":"W2884919358","doi":"10.1002/sim.7908","title":"Modeling the random effects covariance matrix for longitudinal data with covariates measurement error","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Winnipeg; University of Manitoba","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Covariate; Random effects model; Statistics; Estimator; Covariance; Covariance matrix; Mathematics; Generalized linear mixed model; Computer science","score_opus":0.18409707406414916,"score_gpt":0.4458210375468594,"score_spread":0.26172396348271026,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2884919358","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0010554832,0.0002572651,0.998033,0.00018285659,0.000040417493,0.00004326292,0.00012084377,0.00010673603,0.00016011932],"genre_scores_gemma":[0.121261895,0.002431234,0.86854297,0.00074880675,0.0004048955,0.0014290478,0.0014325217,0.00025155093,0.003497046],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9827444,0.011112161,0.0007697947,0.0034398271,0.0015013634,0.0004325452],"domain_scores_gemma":[0.96832746,0.024381587,0.0029664105,0.0027877176,0.0013171267,0.00021971569],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.029407343,0.002262357,0.00261963,0.0032471642,0.0009875002,0.0021936835,0.0046348344,0.0034810365,0.0033032435],"category_scores_gemma":[0.06969568,0.0015327191,0.0036829659,0.0039707534,0.0025791887,0.003565259,0.002869245,0.0054233824,0.001148028],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001380021,0.00012862473,0.013002749,0.0007701673,0.0011604152,0.00068005634,0.0006124005,0.33895332,0.0017120583,0.52149475,0.005021317,0.11632622],"study_design_scores_gemma":[0.00006261845,0.000119067154,0.0027493744,0.00021655438,0.000293032,0.00028312072,0.00008580855,0.60426515,0.0006764927,0.3803007,0.010851387,0.000096772026],"about_ca_topic_score_codex":0.0102345655,"about_ca_topic_score_gemma":0.008857689,"teacher_disagreement_score":0.029407343,"about_ca_system_score_codex":0.0017075471,"about_ca_system_score_gemma":0.0032547012,"threshold_uncertainty_score":0.1555227},"labels":[],"label_agreement":null},{"id":"W2886727084","doi":"10.1002/sim.7854","title":"A validation sampling approach for consistent estimation of adverse drug reaction risk with misclassified right‐censored survival data","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Distribution Estimation and Applications","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Carleton University; Institute of Population and Public Health; McGill University Health Centre; University of Ottawa","funders":"Natural Sciences and Engineering Research Council of Canada; Mitacs; McGill University","keywords":"Estimator; Statistics; Univariate; Proportional hazards model; Unobservable; Event (particle physics); Computer science; Consistency (knowledge bases); Estimation; Econometrics; Sampling (signal processing); Survival analysis; Mathematics; Multivariate statistics; Artificial intelligence","score_opus":0.1482219059636884,"score_gpt":0.4136559518495407,"score_spread":0.2654340458858523,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2886727084","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00570234,0.00005524223,0.9938304,0.00006368397,0.00001463638,0.00007210905,0.000029647475,0.000066886205,0.00016517942],"genre_scores_gemma":[0.29648334,0.0002014291,0.70070827,0.00019435753,0.0000695405,0.000892779,0.00049546163,0.00010245987,0.0008523507],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.97524774,0.020710861,0.0006572132,0.0011692757,0.0018803042,0.0003347256],"domain_scores_gemma":[0.875967,0.10332211,0.005246109,0.009542273,0.005472499,0.00044997735],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.050072853,0.0008294142,0.001488201,0.0019301635,0.00083387445,0.0012891042,0.003240297,0.0018844025,0.0012826981],"category_scores_gemma":[0.14900158,0.0009060886,0.0017404443,0.0014403515,0.0021103586,0.0016028327,0.0023980332,0.0021205854,0.00025630146],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003859113,0.00021234585,0.02246585,0.00029011868,0.00055718963,0.0005349487,0.0007869167,0.5963609,0.0033839233,0.23886243,0.0018193634,0.1343402],"study_design_scores_gemma":[0.00005379909,0.0000836494,0.0013092499,0.000060388746,0.000039830335,0.000100830686,0.000028904735,0.9543383,0.0011957195,0.041940123,0.0008211052,0.000028075448],"about_ca_topic_score_codex":0.0032927166,"about_ca_topic_score_gemma":0.0027113806,"teacher_disagreement_score":0.94992715,"about_ca_system_score_codex":0.0012502951,"about_ca_system_score_gemma":0.001813457,"threshold_uncertainty_score":0.26481372},"labels":[],"label_agreement":null},{"id":"W2887614096","doi":"10.1002/sim.7693","title":"Comparing predictive abilities of longitudinal child growth models","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Child Nutrition and Water Access","field":"Nursing","cited_by":38,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Conselho Nacional de Desenvolvimento Científico e Tecnológico; Fogarty International Center; Fundação de Amparo à Pesquisa do Estado do Rio Grande do Sul; Department for International Development; Nutricia Research Foundation; British Heart Foundation; Wellcome Trust; Department for International Development, UK Government; World Health Organization; Ministério da Saúde; South African Medical Research Council; Department of Science and Technology, Ministry of Science and Technology, India; Wenner-Gren Foundation; Carolina Population Center, University of North Carolina at Chapel Hill; Andrew W. Mellon Foundation; Bill and Melinda Gates Foundation; National Institutes of Health; Indian Council of Medical Research; International Development Research Centre; United States Agency for International Development; National Science Foundation","keywords":"Estimation; Raw data; Computer science; Psychological intervention; Work (physics); Longitudinal data; Child development; Econometrics; Data science; Data mining; Psychology; Economics; Developmental psychology","score_opus":0.036968192901476814,"score_gpt":0.3174421561511075,"score_spread":0.2804739632496307,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2887614096","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.84799224,0.002095078,0.14126152,0.0019829785,0.0002666755,0.00019730833,0.001837496,0.0010522287,0.0033144776],"genre_scores_gemma":[0.9798902,0.0005130307,0.015833827,0.00012787491,0.000074667914,0.000112615046,0.0026564384,0.0001217075,0.0006695401],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98394483,0.013247092,0.0004953526,0.0012187394,0.00069405657,0.00039995616],"domain_scores_gemma":[0.6869467,0.2943093,0.0047966666,0.008371756,0.0045232,0.0010524013],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.075869076,0.0018654881,0.0011161434,0.0032574788,0.0009127903,0.002675808,0.0020823623,0.0013680805,0.0014968971],"category_scores_gemma":[0.118271664,0.0007241578,0.0026449715,0.0021272504,0.0012678414,0.0023503485,0.0020838322,0.002678065,0.0005837871],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014926986,0.0003878231,0.2840448,0.0002532297,0.0022271243,0.00017357792,0.0012561412,0.6100154,0.0003600721,0.008098769,0.0030929258,0.08859752],"study_design_scores_gemma":[0.00003816905,0.00033456428,0.021490451,0.00012664172,0.00019493162,0.00005487832,0.0003578086,0.9694397,0.00032632204,0.0067452006,0.000832122,0.00005925686],"about_ca_topic_score_codex":0.018641641,"about_ca_topic_score_gemma":0.010612783,"teacher_disagreement_score":0.075869076,"about_ca_system_score_codex":0.0012970911,"about_ca_system_score_gemma":0.001865149,"threshold_uncertainty_score":0.4012388},"labels":[],"label_agreement":null},{"id":"W2888844257","doi":"10.1002/sim.7942","title":"Estimation in generalized linear models under censored covariates with an application to MIREC data","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University; Health Canada","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Health Canada; Ministry of Natural Resources","keywords":"Covariate; Estimator; Generalized linear model; Statistics; Econometrics; Generalized estimating equation; Computer science; Linear model; Mathematics","score_opus":0.1248335094346998,"score_gpt":0.45313467404832947,"score_spread":0.3283011646136297,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2888844257","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010717333,0.00094896223,0.9869266,0.0006081414,0.0000327634,0.000058714515,0.0002310298,0.00021746273,0.0002589413],"genre_scores_gemma":[0.28099796,0.0029392547,0.7085554,0.00060154294,0.00040599483,0.0008374263,0.0019648473,0.0002481607,0.0034493848],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99127734,0.007007091,0.0002644132,0.00072031684,0.0005214416,0.00020927918],"domain_scores_gemma":[0.9376957,0.05581203,0.00284534,0.0020115539,0.0013355032,0.0002999395],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020960767,0.0010793698,0.0018345123,0.0017113895,0.0006693968,0.0014325824,0.0026263376,0.0019663817,0.0022531655],"category_scores_gemma":[0.06504246,0.0008271829,0.0019705317,0.0023129785,0.0015385147,0.0017185222,0.0024955452,0.0029493705,0.00045256427],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00032156284,0.00013675923,0.008628736,0.00068177306,0.0004977495,0.00076877204,0.00052396656,0.7056285,0.0013101951,0.19007473,0.002953942,0.088473305],"study_design_scores_gemma":[0.000054854496,0.00006673471,0.0014737898,0.000045090288,0.0000372957,0.00011109482,0.000051572966,0.9106771,0.00030459856,0.08428454,0.0028578977,0.000035465906],"about_ca_topic_score_codex":0.008777188,"about_ca_topic_score_gemma":0.0070534716,"teacher_disagreement_score":0.020960767,"about_ca_system_score_codex":0.0010556651,"about_ca_system_score_gemma":0.0015882454,"threshold_uncertainty_score":0.11085242},"labels":[],"label_agreement":null},{"id":"W2890626091","doi":"10.1002/sim.7963","title":"Modeling semicontinuous longitudinal data with order constraints","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Covariate; Inference; Computer science; Longitudinal data; Statistical inference; Econometrics; Statistical hypothesis testing; Joint (building); Machine learning; Statistics; Data mining; Artificial intelligence; Mathematics","score_opus":0.12914742725631273,"score_gpt":0.43149439201528045,"score_spread":0.3023469647589677,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2890626091","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.031910986,0.0006583119,0.9650033,0.00078271737,0.000042196076,0.00007458296,0.00046470587,0.0001729036,0.00089016097],"genre_scores_gemma":[0.62450415,0.001994464,0.36525849,0.0008880154,0.00025027373,0.0008776425,0.0017728587,0.00019592722,0.0042582015],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9882079,0.0075302264,0.00066920056,0.001828086,0.0011990177,0.00056555035],"domain_scores_gemma":[0.8763337,0.10751427,0.007343202,0.005992546,0.0020622278,0.0007540903],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.028524878,0.0012510928,0.0027913009,0.0017156181,0.000752939,0.0028409124,0.0036401593,0.0024491886,0.0032474818],"category_scores_gemma":[0.08301015,0.0014688289,0.0022621073,0.0027521031,0.003270574,0.003989852,0.002902214,0.004398377,0.00046199808],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00032224355,0.00012644572,0.015194575,0.0003560311,0.00032719766,0.0012777492,0.0009455065,0.49417093,0.0009452243,0.44101915,0.0019956667,0.043319337],"study_design_scores_gemma":[0.000048450347,0.00006449849,0.0018641618,0.00006709052,0.000048717182,0.00016749701,0.000054518416,0.6625278,0.00020127243,0.33379483,0.0011272191,0.00003400884],"about_ca_topic_score_codex":0.009968839,"about_ca_topic_score_gemma":0.0087991115,"teacher_disagreement_score":0.028524878,"about_ca_system_score_codex":0.0017676969,"about_ca_system_score_gemma":0.0029066245,"threshold_uncertainty_score":0.15085578},"labels":[],"label_agreement":null},{"id":"W2893312973","doi":"10.1002/sim.7984","title":"Judgment post‐stratification in finite mixture modeling: An example in estimating the prevalence of osteoporosis","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Estimator; Ranking (information retrieval); Statistics; Rank (graph theory); Small area estimation; Econometrics; Mathematics; Computer science; Expectation–maximization algorithm; Measure (data warehouse); Population; Data mining; Maximum likelihood; Artificial intelligence; Medicine","score_opus":0.05566959108421915,"score_gpt":0.3371850861906867,"score_spread":0.28151549510646756,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2893312973","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04836109,0.00023342599,0.9496799,0.0004178009,0.00003210047,0.00008953719,0.000041954434,0.0001421408,0.00100199],"genre_scores_gemma":[0.43131247,0.00022294204,0.56661135,0.00015062529,0.0000692201,0.00015813843,0.00014068504,0.00005314319,0.001281328],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98735285,0.010346218,0.00028488538,0.0006357049,0.0011735569,0.00020677994],"domain_scores_gemma":[0.95670223,0.038411155,0.0010327171,0.001955884,0.0016308604,0.00026715934],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.022892315,0.00074848003,0.0013798894,0.0014209842,0.0012399524,0.0011783302,0.00139212,0.0019880629,0.0012456344],"category_scores_gemma":[0.061714012,0.0004784988,0.0014907484,0.0024051883,0.0019366753,0.001386464,0.0020101967,0.0021624495,0.0002469231],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015855806,0.0004797553,0.04725633,0.0003869362,0.00038573673,0.0015598668,0.0035376372,0.41036338,0.0063000284,0.17455795,0.0022386045,0.35134825],"study_design_scores_gemma":[0.00006118036,0.00016588539,0.0051283147,0.000033875956,0.000059643102,0.0002245322,0.00015555664,0.91090363,0.0019679512,0.079822734,0.0014086354,0.00006813054],"about_ca_topic_score_codex":0.0073721665,"about_ca_topic_score_gemma":0.009950205,"teacher_disagreement_score":0.022892315,"about_ca_system_score_codex":0.0008963003,"about_ca_system_score_gemma":0.0010917068,"threshold_uncertainty_score":0.12106758},"labels":[],"label_agreement":null},{"id":"W2896981660","doi":"10.1002/sim.8008","title":"Propensity‐score matching with competing risks in survival analysis","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":147,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Work & Health; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Statistics; Covariate; Observational study; Matching (statistics); Hazard ratio; Confounding; Sample size determination; Medicine; Absolute risk reduction; Confidence interval; Econometrics; Mathematics","score_opus":0.23846640406203207,"score_gpt":0.45839961939779816,"score_spread":0.2199332153357661,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2896981660","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011680148,0.00048904406,0.9971131,0.00026509946,0.000058804308,0.0002051022,0.00010539911,0.00019118248,0.00040423457],"genre_scores_gemma":[0.07508327,0.0020113464,0.9173923,0.00029397226,0.00041798822,0.0022758234,0.000695556,0.00023157302,0.0015982051],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95360243,0.040152974,0.0012774811,0.0020497781,0.0025263403,0.00039084902],"domain_scores_gemma":[0.9225355,0.06704519,0.0037716373,0.004637277,0.0016384189,0.000372027],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.048631843,0.0014369081,0.002671584,0.0048871865,0.00085913634,0.0024228366,0.0032238693,0.0023027577,0.0067480505],"category_scores_gemma":[0.14023486,0.0013484393,0.0033089,0.0066310447,0.0023870934,0.0026912868,0.0038259246,0.0047140806,0.0011244258],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00030730575,0.00023046382,0.012941035,0.00095909403,0.0014810251,0.0004607292,0.0004223752,0.21208046,0.00051816023,0.52344567,0.011017365,0.23613638],"study_design_scores_gemma":[0.00015794863,0.000115455456,0.0020733087,0.00019227967,0.00013774094,0.00022547058,0.000048244572,0.51815975,0.0004674872,0.46951815,0.008840635,0.000063571184],"about_ca_topic_score_codex":0.004160546,"about_ca_topic_score_gemma":0.0025858989,"teacher_disagreement_score":0.95136815,"about_ca_system_score_codex":0.0018272123,"about_ca_system_score_gemma":0.004349089,"threshold_uncertainty_score":0.25719285},"labels":[],"label_agreement":null},{"id":"W2897505629","doi":"10.1002/sim.7990","title":"Methods to improve the estimation of time‐to‐event outcomes when data is de‐identified","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Privacy-Preserving Technologies in Data","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Ontario Institute for Cancer Research; Hamilton Health Sciences; University of Manitoba; Manitoba Health; Research Institute in Oncology and Hematology; Health Sciences Centre; CancerCare Manitoba; Cancer Care Ontario; McMaster University","funders":"","keywords":"Computer science; Identification (biology); Data set; Event (particle physics); Set (abstract data type); Estimation; Statistics; Patient confidentiality; Survival analysis; Data mining; Missing data; Confidentiality; Econometrics; Mathematics; Machine learning; Artificial intelligence","score_opus":0.0651600521362048,"score_gpt":0.43195683777357047,"score_spread":0.3667967856373657,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2897505629","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0037542647,0.0006045426,0.99456173,0.0002956576,0.0000803029,0.00012078738,0.00010786981,0.00015731138,0.00031759142],"genre_scores_gemma":[0.11664972,0.0013305675,0.87934047,0.00027064595,0.00024091019,0.00074565934,0.0004849313,0.000098906115,0.00083816866],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9760641,0.018196475,0.0011523602,0.0018107926,0.0025365616,0.00023966846],"domain_scores_gemma":[0.8062397,0.16085514,0.013582295,0.01245232,0.0063849213,0.00048569817],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.045218762,0.0013137626,0.001429177,0.0030839634,0.0004975278,0.0018277082,0.0024077399,0.0013438233,0.0018985386],"category_scores_gemma":[0.1822313,0.000911161,0.0016200807,0.0027505008,0.0010931772,0.0025697164,0.0019937968,0.0030337898,0.0006606055],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008272071,0.00038991187,0.027715044,0.0015415491,0.0020028027,0.00032486167,0.0011371984,0.1773981,0.0035555789,0.080766276,0.0044412646,0.69990015],"study_design_scores_gemma":[0.00040651122,0.00053070654,0.013348083,0.0005994211,0.0005711477,0.00080325425,0.0002015309,0.84261423,0.0075761215,0.11549845,0.017655155,0.0001953705],"about_ca_topic_score_codex":0.001965507,"about_ca_topic_score_gemma":0.0021252695,"teacher_disagreement_score":0.045218762,"about_ca_system_score_codex":0.0009681111,"about_ca_system_score_gemma":0.0028613624,"threshold_uncertainty_score":0.23914254},"labels":[],"label_agreement":null},{"id":"W2902445064","doi":"10.1002/sim.8051","title":"Bayesian adaptive group lasso with semiparametric hidden Markov models","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"National Institute of Mental Health; Canadian Institutes of Health Research; National Institutes of Health; Natural Sciences and Engineering Research Council of Canada; National Natural Science Foundation of China","keywords":"Covariate; Variable-order Bayesian network; Econometrics; Bayesian probability; Semiparametric regression; Prior probability; Mathematics; Lasso (programming language); Computer science; Nonparametric statistics; Bayesian inference; Artificial intelligence","score_opus":0.02237722273859513,"score_gpt":0.29250909172723244,"score_spread":0.2701318689886373,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2902445064","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0014705664,0.000099655954,0.9979426,0.00009237419,0.000012657865,0.000020880818,0.000023516171,0.00013053537,0.00020718275],"genre_scores_gemma":[0.17784828,0.00054223253,0.81647897,0.00045520902,0.00025852397,0.0007448381,0.00064798404,0.00034163558,0.002682416],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99409,0.004079492,0.00018637195,0.0005559668,0.00086895,0.00021911322],"domain_scores_gemma":[0.99365133,0.004831022,0.0005299718,0.00042906636,0.00042856907,0.00013012598],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007762008,0.0014511585,0.00277865,0.0011909214,0.0007702763,0.0013777497,0.0031695927,0.0017057405,0.001997937],"category_scores_gemma":[0.013079382,0.00093426084,0.0018264113,0.0015862739,0.0013993963,0.0016509687,0.0028572201,0.0040960545,0.00077927136],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017838163,0.00013542452,0.0012989595,0.00025746552,0.0003506998,0.00015823824,0.00026337212,0.76149654,0.0022786926,0.08334536,0.0042233723,0.14601351],"study_design_scores_gemma":[0.000016520022,0.000018727957,0.0000886076,0.000009682497,0.00001082735,0.000016007341,0.000007505546,0.97812825,0.00026966012,0.020799132,0.00062457606,0.000010529976],"about_ca_topic_score_codex":0.0026400264,"about_ca_topic_score_gemma":0.0029464415,"teacher_disagreement_score":0.007762008,"about_ca_system_score_codex":0.00077368674,"about_ca_system_score_gemma":0.0021396219,"threshold_uncertainty_score":0.041049957},"labels":[],"label_agreement":null},{"id":"W2902926083","doi":"10.1002/sim.8048","title":"Using a monotone single‐index model to stabilize the propensity score in missing data problems and causal inference","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"National Cancer Institute; Natural Sciences and Engineering Research Council of Canada; National Institutes of Health","keywords":"Causal inference; Propensity score matching; Inference; Missing data; Index (typography); Statistics; Monotone polygon; Computer science; Econometrics; Causal model; Mathematics; Artificial intelligence","score_opus":0.5427102071927957,"score_gpt":0.4837563441874663,"score_spread":0.0589538630053294,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2902926083","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0025479114,0.00010535132,0.9966383,0.00021452722,0.00002176869,0.000033293152,0.00003730986,0.00006376726,0.00033784422],"genre_scores_gemma":[0.16565685,0.00073111645,0.8289205,0.00044714788,0.0002978755,0.0007310827,0.0003387474,0.00014847364,0.0027282857],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9857664,0.011074599,0.00033119714,0.0012143791,0.0012876749,0.00032571156],"domain_scores_gemma":[0.9605853,0.030024635,0.0030921942,0.0040410277,0.001872944,0.00038388994],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.031485084,0.00094326196,0.0018400318,0.0021265978,0.0009759361,0.0014680002,0.0033198947,0.0020143364,0.0043834415],"category_scores_gemma":[0.06883758,0.00075230934,0.002028028,0.003149793,0.002916536,0.0037309527,0.0029013404,0.0039142994,0.0007475601],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017626764,0.00012651276,0.0041238843,0.00028863963,0.000230806,0.00026739392,0.00050265266,0.1553807,0.0012817123,0.71363187,0.002634611,0.121354975],"study_design_scores_gemma":[0.0000735361,0.00012941645,0.00077405147,0.00005985529,0.000058320555,0.00013060175,0.0000417205,0.59092784,0.0006492111,0.40356058,0.003553928,0.000040898576],"about_ca_topic_score_codex":0.003128469,"about_ca_topic_score_gemma":0.0028681688,"teacher_disagreement_score":0.031485084,"about_ca_system_score_codex":0.0011646898,"about_ca_system_score_gemma":0.0032153653,"threshold_uncertainty_score":0.166511},"labels":[],"label_agreement":null},{"id":"W2906741671","doi":"10.1002/sim.8075","title":"Should a propensity score model be super? The utility of ensemble procedures for causal adjustment","year":2018,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre; McGill University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Propensity score matching; Covariate; Logistic regression; Statistics; Inverse probability weighting; Overfitting; Weighting; Regression; Matching (statistics); Mean squared error; Econometrics; Mathematics; Confounding; Computer science; Artificial intelligence; Medicine","score_opus":0.352053424871457,"score_gpt":0.4614552599099146,"score_spread":0.10940183503845757,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2906741671","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0051162294,0.00027314946,0.9931681,0.00059897744,0.000044441153,0.000033841792,0.000052380212,0.0001954969,0.00051744416],"genre_scores_gemma":[0.35050544,0.0012307265,0.64299583,0.0014689396,0.00039490152,0.00049570075,0.00037089677,0.00034569568,0.0021917948],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9839675,0.0127116805,0.0003951665,0.0013390706,0.0013109659,0.00027569424],"domain_scores_gemma":[0.9247022,0.055892043,0.0028794096,0.013418931,0.002599351,0.00050810666],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.047776204,0.00091029494,0.002164606,0.0015221714,0.0007539026,0.0018754551,0.001934412,0.0014286328,0.004046321],"category_scores_gemma":[0.120453715,0.0008137058,0.0017521498,0.0020302855,0.0020600716,0.0056367507,0.002949504,0.0045901146,0.0007447908],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005211242,0.00012409544,0.012904895,0.0003340294,0.0010767753,0.00015448566,0.0005834086,0.26106444,0.0010277749,0.42792484,0.005799554,0.28848457],"study_design_scores_gemma":[0.000055903867,0.0001239183,0.0013521243,0.00008143565,0.000087762375,0.00008196032,0.000042304226,0.56137294,0.00084887125,0.43125752,0.0046603843,0.0000348157],"about_ca_topic_score_codex":0.002871396,"about_ca_topic_score_gemma":0.0021112766,"teacher_disagreement_score":0.9522238,"about_ca_system_score_codex":0.00086187077,"about_ca_system_score_gemma":0.0022627003,"threshold_uncertainty_score":0.25266773},"labels":[],"label_agreement":null},{"id":"W2908350920","doi":"10.1002/sim.8073","title":"Weighted causal inference methods with mismeasured covariates and misclassified outcomes","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Covariate; Causal inference; Inverse probability weighting; Logistic regression; Econometrics; Computer science; Statistics; Weighting; Inference; Observational error; Regression; Estimation; Mathematics; Estimator; Artificial intelligence; Medicine","score_opus":0.1192263844432153,"score_gpt":0.4720951276343318,"score_spread":0.3528687431911165,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2908350920","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011993265,0.00046992511,0.9974475,0.0002813318,0.00006072587,0.000060084134,0.00004252006,0.000050776696,0.00038790662],"genre_scores_gemma":[0.16262579,0.0029534907,0.8287,0.0006899707,0.00072413887,0.0011155636,0.00042019621,0.00014085505,0.0026299972],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97159,0.021570161,0.001057791,0.0026426434,0.0026797943,0.00045964736],"domain_scores_gemma":[0.87656814,0.104172446,0.0068775257,0.008952854,0.0030213185,0.00040767735],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.048348486,0.002005827,0.002649746,0.0038130903,0.0012052745,0.0024427476,0.0050764666,0.003162903,0.0052676415],"category_scores_gemma":[0.17079332,0.0010948929,0.0026196588,0.0048373025,0.0034554768,0.0054021887,0.0036420065,0.004754748,0.0007584113],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000911449,0.00009964533,0.0034060522,0.00070611195,0.00060943497,0.00023172874,0.00041400592,0.07715682,0.0004691393,0.7230208,0.0025988214,0.19119626],"study_design_scores_gemma":[0.00007538774,0.00007173503,0.000664822,0.00021380189,0.00017048226,0.00017288233,0.00006294011,0.25628948,0.0006884603,0.736787,0.004763966,0.00003903295],"about_ca_topic_score_codex":0.0029683923,"about_ca_topic_score_gemma":0.002270486,"teacher_disagreement_score":0.9516515,"about_ca_system_score_codex":0.0017794584,"about_ca_system_score_gemma":0.0032975243,"threshold_uncertainty_score":0.25569427},"labels":[],"label_agreement":null},{"id":"W2910522615","doi":"10.1002/sim.8503","title":"Propensity scores using missingness pattern information: a practical guide","year":2020,"lang":"en","type":"preprint","venue":"Statistics in Medicine","topic":"Healthcare Policy and Management","field":"Economics, Econometrics and Finance","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Health and Social Care Research and Development Division; Engineering and Physical Sciences Research Council; Economic and Social Research Council; Chief Scientist Office, Scottish Government Health and Social Care Directorate; Medical Research Council Canada; London School of Hygiene and Tropical Medicine; Medical Research Council; Public Health Agency; Department of Health and Social Care; Scottish Government; British Heart Foundation; Wellcome Trust","keywords":"Missing data; Propensity score matching; Confounding; Health records; Computer science; Statistics; Econometrics; Psychology; Medicine; Health care; Mathematics; Machine learning","score_opus":0.2291836310970289,"score_gpt":0.40090395501963755,"score_spread":0.17172032392260864,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2910522615","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00033870136,0.0017539834,0.9780743,0.007825652,0.00045460105,0.0011447959,0.0020917214,0.0030843697,0.0052318797],"genre_scores_gemma":[0.0025692694,0.0032147737,0.98636484,0.0012645851,0.0005336637,0.0017143227,0.0010280067,0.00056039006,0.00275016],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98518467,0.010000107,0.0018709278,0.00071531767,0.0021180077,0.00011102489],"domain_scores_gemma":[0.9256236,0.060769424,0.0028020947,0.003461046,0.0064917016,0.00085215643],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026853414,0.003361618,0.0030070078,0.010483846,0.0011534436,0.0053242375,0.0047031264,0.004253813,0.032036055],"category_scores_gemma":[0.11683829,0.003033568,0.0021794413,0.0074995533,0.0032640968,0.006416871,0.0042119883,0.01043712,0.017013676],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000096351214,0.00026998977,0.0018908449,0.0015718797,0.00024423792,0.00077965856,0.0011943785,0.012895656,0.00081505754,0.14471991,0.41307092,0.42245114],"study_design_scores_gemma":[0.00024835812,0.0001413381,0.001321936,0.0018749656,0.0000828593,0.0020147057,0.00068317965,0.041065007,0.00065262365,0.46966228,0.48200962,0.00024306681],"about_ca_topic_score_codex":0.0034798968,"about_ca_topic_score_gemma":0.004765759,"teacher_disagreement_score":0.032036055,"about_ca_system_score_codex":0.0013932536,"about_ca_system_score_gemma":0.004911482,"threshold_uncertainty_score":0.14201611},"labels":[],"label_agreement":null},{"id":"W2916449784","doi":"10.1002/sim.8119","title":"Optimising the two‐stage randomised trial design when some participants are indifferent in their treatment preferences","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Preference; Randomized controlled trial; Treatment and control groups; Treatment effect; Medicine; Clinical trial; Research design; Psychology; Statistics; Mathematics; Surgery; Internal medicine","score_opus":0.6713702372673928,"score_gpt":0.5579903408757302,"score_spread":0.11337989639166257,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2916449784","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008252745,0.0010403172,0.9726254,0.0015363438,0.00044072117,0.0145713715,0.0001334642,0.00020874993,0.0011909076],"genre_scores_gemma":[0.06687308,0.0005347267,0.89794177,0.0011674656,0.00018236467,0.03237144,0.00009874438,0.000048747846,0.00078170357],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.46848702,0.50103885,0.011462943,0.007488455,0.010040922,0.0014818462],"domain_scores_gemma":[0.59847015,0.34319302,0.01885546,0.029348992,0.008155435,0.0019769922],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.36937627,0.0030513129,0.006162073,0.0018880634,0.0008929354,0.0029172683,0.0036268304,0.006992755,0.0042820633],"category_scores_gemma":[0.41148534,0.0023670776,0.0058504962,0.002286752,0.00583076,0.005410722,0.0041854605,0.0074464274,0.0012567969],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.05567407,0.0019771883,0.007949741,0.01327291,0.006800582,0.00088292017,0.004282767,0.11166063,0.011277976,0.4047641,0.007010603,0.3744465],"study_design_scores_gemma":[0.035460096,0.02314561,0.004617323,0.0028635918,0.0039804764,0.00087740633,0.0002909618,0.26475298,0.008744112,0.62703204,0.027506141,0.00072938384],"about_ca_topic_score_codex":0.00044930988,"about_ca_topic_score_gemma":0.0006698849,"teacher_disagreement_score":0.6306237,"about_ca_system_score_codex":0.0023728125,"about_ca_system_score_gemma":0.007169557,"threshold_uncertainty_score":0.77767134},"labels":[],"label_agreement":null},{"id":"W2921476564","doi":"10.1002/sim.8132","title":"A test for the correct specification of marginal structural models","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université Laval; The Quebec Population Health Research Network","funders":"Natural Sciences and Engineering Research Council of Canada; Fondation du CHUM","keywords":"Marginal structural model; Inverse probability; Computer science; Estimator; Econometrics; Outcome (game theory); Confounding; Statistics; Causal model; Mathematics; Bayesian probability; Posterior probability","score_opus":0.14322652898192254,"score_gpt":0.43763620167038475,"score_spread":0.2944096726884622,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2921476564","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.18010063,0.00033629168,0.80738884,0.0026947795,0.00022877027,0.00060114526,0.002481255,0.0010573923,0.0051108385],"genre_scores_gemma":[0.8217149,0.00014111269,0.17383282,0.0004910215,0.00015124647,0.00095598283,0.0015942121,0.00013836141,0.0009803469],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.92477506,0.05655202,0.0031390542,0.008213588,0.0057261446,0.0015940618],"domain_scores_gemma":[0.32797372,0.6402159,0.010114361,0.014775012,0.0056568068,0.0012641312],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.062834494,0.0014556645,0.002962367,0.0029728496,0.0014925253,0.0023330397,0.003840146,0.0040395064,0.015070986],"category_scores_gemma":[0.44042715,0.0009139377,0.0031043016,0.0038852643,0.0047693495,0.0039891205,0.0025191416,0.0041654925,0.001125446],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0032049709,0.0009858547,0.2700348,0.0015665288,0.008088588,0.002406408,0.0033806989,0.11677309,0.004018512,0.30943337,0.012171131,0.26793608],"study_design_scores_gemma":[0.0011864555,0.0028649918,0.0626245,0.00040942512,0.0009816772,0.001309642,0.001655177,0.69370544,0.0041335807,0.21946992,0.011461753,0.00019747847],"about_ca_topic_score_codex":0.005548864,"about_ca_topic_score_gemma":0.0025979795,"teacher_disagreement_score":0.062834494,"about_ca_system_score_codex":0.0014535437,"about_ca_system_score_gemma":0.005931668,"threshold_uncertainty_score":0.33230448},"labels":[],"label_agreement":null},{"id":"W2921943956","doi":"10.1002/sim.8121","title":"Bayesian estimation of the average treatment effect on the treated using inverse weighting","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Fonds de Recherche du Québec - Santé; Natural Sciences and Engineering Research Council of Canada; Vanderbilt University","keywords":"Confounding; Statistics; Marginal structural model; Context (archaeology); Bayesian probability; Weighting; Posterior probability; Inverse probability; Econometrics; Estimation; Sample size determination; Average treatment effect; Inverse probability weighting; Marginal likelihood; Mathematics; Computer science; Medicine; Propensity score matching; Economics","score_opus":0.06999270852247445,"score_gpt":0.4013656679853423,"score_spread":0.3313729594628678,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2921943956","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0038244901,0.00019532652,0.99467975,0.0002769668,0.00002439753,0.00007064765,0.0000746405,0.000060647868,0.000793258],"genre_scores_gemma":[0.20311071,0.0010674213,0.7913427,0.00043720283,0.00018800954,0.0008630157,0.00044429998,0.00009951345,0.0024471306],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9843321,0.011073149,0.000560023,0.0014901771,0.0021484247,0.00039613058],"domain_scores_gemma":[0.95598865,0.036830492,0.0019949637,0.0031701587,0.0017392718,0.00027654954],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026091967,0.00086466793,0.0025507924,0.002635439,0.00059017073,0.0018117651,0.0032274183,0.0016791926,0.0052363873],"category_scores_gemma":[0.09408993,0.00077021646,0.0021110817,0.002282676,0.0017107739,0.0026344638,0.0022913958,0.0031930096,0.0006312846],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00022347775,0.00022714555,0.008641398,0.00043310496,0.0007454652,0.00020360594,0.0005648762,0.17123388,0.0018088565,0.5950545,0.0026700043,0.21819371],"study_design_scores_gemma":[0.00010936257,0.00013204683,0.0025409765,0.00015023607,0.00023679966,0.00014517426,0.000060437753,0.45328405,0.0012021598,0.536917,0.005170188,0.000051613024],"about_ca_topic_score_codex":0.004689835,"about_ca_topic_score_gemma":0.0036190057,"teacher_disagreement_score":0.026091967,"about_ca_system_score_codex":0.0012149289,"about_ca_system_score_gemma":0.00263149,"threshold_uncertainty_score":0.13798916},"labels":[],"label_agreement":null},{"id":"W2923835633","doi":"10.1002/sim.8120","title":"Emulating a trial of joint dynamic strategies: An application to monitoring and treatment of HIV‐positive individuals","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"HIV Research and Treatment","field":"Immunology and Microbiology","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SAIT Polytechnic","funders":"National Center for Advancing Translational Sciences; National Institute of Allergy and Infectious Diseases; National Heart, Lung, and Blood Institute; Medical Research Council; Center for AIDS Research, University of Washington; National Institutes of Health","keywords":"Regimen; Observational study; Human immunodeficiency virus (HIV); Medicine; Antiretroviral therapy; Leverage (statistics); Outcome (game theory); Computer science; Internal medicine; Viral load; Immunology; Mathematics; Artificial intelligence","score_opus":0.02339519020486872,"score_gpt":0.35897860717107233,"score_spread":0.3355834169662036,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2923835633","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2638395,0.0025564786,0.7127303,0.0071577705,0.00069160626,0.0046850787,0.0008138458,0.00065663725,0.006868802],"genre_scores_gemma":[0.6700553,0.00069105375,0.31729975,0.0022599283,0.00025562188,0.0070212176,0.00040690563,0.00009957036,0.0019106755],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8061272,0.18610628,0.0016595846,0.002799382,0.0025230048,0.00078451703],"domain_scores_gemma":[0.4761353,0.49804258,0.007856552,0.013709123,0.0027222182,0.0015342134],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.17722526,0.0017312166,0.004258134,0.0010835119,0.00072254165,0.0019807064,0.0033636973,0.0038361223,0.0043193772],"category_scores_gemma":[0.3292433,0.0009383077,0.0047420342,0.0016792679,0.0031084232,0.0033549406,0.0034938995,0.0057142754,0.00037017747],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.044351272,0.0045628655,0.019241203,0.0018177974,0.00756886,0.0005845711,0.0014637035,0.6246401,0.0019917341,0.10465709,0.0030004159,0.18612042],"study_design_scores_gemma":[0.012782671,0.01227947,0.003513991,0.00029508778,0.0015328676,0.00020974765,0.00022372112,0.85327685,0.0020930215,0.109454006,0.0041684993,0.00017010054],"about_ca_topic_score_codex":0.0033809366,"about_ca_topic_score_gemma":0.0017615922,"teacher_disagreement_score":0.17722526,"about_ca_system_score_codex":0.0025976656,"about_ca_system_score_gemma":0.004323372,"threshold_uncertainty_score":0.93726796},"labels":[],"label_agreement":null},{"id":"W2924458846","doi":"10.1002/sim.8142","title":"Randomised trials with provision for early stopping for benefit (or harm): The impact on the estimated treatment effect","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":42,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Mount Sinai Hospital; University of Ottawa; McMaster University; Ottawa Hospital; Impact","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Early stopping; Interim; Interim analysis; Type I and type II errors; Estimation; Harm; Sample size determination; Econometrics; Meta-analysis; Statistics; Stopping time; Treatment effect; Computer science; Clinical trial; Medicine; Mathematics; Psychology; Economics; Artificial intelligence","score_opus":0.5239753487763673,"score_gpt":0.6123429647629159,"score_spread":0.08836761598654863,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2924458846","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017233716,0.022164386,0.93330616,0.016190253,0.0014969682,0.0016845528,0.0005144006,0.00066679955,0.0067427787],"genre_scores_gemma":[0.37244388,0.008124603,0.6023066,0.0092161065,0.0012051369,0.0039895205,0.0003638312,0.00042558246,0.0019248872],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.6111485,0.34653357,0.017345285,0.006775934,0.01616812,0.0020287016],"domain_scores_gemma":[0.18458533,0.75460833,0.027431553,0.025030786,0.00733759,0.0010064106],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.30228665,0.0027999855,0.008580746,0.003795743,0.0013416616,0.008518602,0.0041689277,0.0125869755,0.005080618],"category_scores_gemma":[0.7201283,0.00224258,0.0087157795,0.0048780693,0.005800665,0.013260086,0.0050246245,0.013101613,0.0011852515],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0055188145,0.0002476519,0.009379184,0.010467206,0.0046300464,0.0009985189,0.0020565249,0.0940145,0.0011030663,0.5151613,0.009632502,0.34679073],"study_design_scores_gemma":[0.0030869215,0.0015588661,0.0039032823,0.004449188,0.0035770724,0.0012786213,0.00023841258,0.1622858,0.0019533765,0.80182713,0.015510819,0.00033038558],"about_ca_topic_score_codex":0.0012120093,"about_ca_topic_score_gemma":0.00086544396,"teacher_disagreement_score":0.6977134,"about_ca_system_score_codex":0.0034817941,"about_ca_system_score_gemma":0.0064533986,"threshold_uncertainty_score":0.8604048},"labels":[],"label_agreement":null},{"id":"W2938511274","doi":"10.1002/sim.8162","title":"One‐step validation method for surrogate endpoints using data from multiple randomized cancer clinical trials with failure‐time endpoints","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Secrétariat Général pour les Affaires Régionales, Etat en Région Aquitaine; Institute of Cancer Research; Institut National de la Santé et de la Recherche Médicale; Institut National Du Cancer; Association pour la Recherche sur le Cancer","keywords":"Surrogate endpoint; Context (archaeology); Estimator; Statistics; Clinical trial; Computer science; Random effects model; Medicine; Mathematics; Meta-analysis; Internal medicine","score_opus":0.5056720796718315,"score_gpt":0.5934294578687577,"score_spread":0.0877573781969262,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2938511274","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0059387824,0.0011525196,0.990803,0.0002754538,0.00012989981,0.00083002425,0.00030687798,0.00023785721,0.00032565248],"genre_scores_gemma":[0.22337835,0.0011804176,0.7641288,0.0007731357,0.00025985303,0.0068406817,0.001618303,0.00020913128,0.0016113193],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.89578956,0.08897864,0.004533345,0.005424519,0.0048188292,0.00045516802],"domain_scores_gemma":[0.6747041,0.28560704,0.010918437,0.021131996,0.0068663214,0.000772164],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.17823577,0.002323504,0.0040886346,0.0042011044,0.00097173254,0.0028051494,0.0039824555,0.00324016,0.004025907],"category_scores_gemma":[0.28028885,0.0011417712,0.0076058265,0.0031166999,0.0021790254,0.0026111258,0.0029421512,0.004477234,0.00075447035],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0041944957,0.00063716155,0.03826179,0.0077013657,0.020040365,0.0014839066,0.0012510712,0.26551977,0.0057924516,0.14684118,0.00701874,0.5012577],"study_design_scores_gemma":[0.0013594868,0.0018853649,0.008173769,0.0012697052,0.0046482985,0.0007871296,0.0001507431,0.8085291,0.0055680224,0.15392314,0.013451574,0.0002536843],"about_ca_topic_score_codex":0.001223034,"about_ca_topic_score_gemma":0.0009972331,"teacher_disagreement_score":0.82176423,"about_ca_system_score_codex":0.0011060054,"about_ca_system_score_gemma":0.003680472,"threshold_uncertainty_score":0.9426121},"labels":[],"label_agreement":null},{"id":"W2939964754","doi":"10.1002/sim.8155","title":"Score tests based on a finite mixture model of Markov processes under intermittent observation","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Systemic Lupus Erythematosus Research","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Confidence interval; Psoriatic arthritis; Mixture model; Permutation (music); Statistics; Estimator; Markov chain; Medicine; Arthritis; Mathematics; Computer science; Internal medicine","score_opus":0.059281321664442596,"score_gpt":0.34591616981542134,"score_spread":0.2866348481509787,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2939964754","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010011736,0.00011267946,0.9886929,0.00016263954,0.000038372225,0.0000678408,0.000085771244,0.00036746424,0.00046069248],"genre_scores_gemma":[0.4074833,0.0005336394,0.58520395,0.00023817357,0.00025162764,0.0008947865,0.0010026363,0.00024976878,0.004142228],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9917965,0.005610858,0.00029530158,0.0009328013,0.0009965915,0.0003679271],"domain_scores_gemma":[0.9089751,0.08270308,0.0026686606,0.0028370058,0.0020452356,0.00077086006],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014765923,0.0012980513,0.0025330996,0.0026668021,0.0010044049,0.0026086348,0.0034364169,0.0019514005,0.0040729763],"category_scores_gemma":[0.07651607,0.00080190407,0.0019839373,0.0023095885,0.0033080508,0.0035189528,0.002761408,0.0038923854,0.0007700712],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00048270088,0.00015782003,0.014960951,0.00021902629,0.0003433638,0.00051976496,0.00035882968,0.5067555,0.00229855,0.320819,0.0024803705,0.15060407],"study_design_scores_gemma":[0.00003079226,0.00010324523,0.0012261921,0.000022906774,0.000042813128,0.0001661198,0.000027007822,0.9226152,0.00045776562,0.07453425,0.00072130735,0.000052465093],"about_ca_topic_score_codex":0.005473951,"about_ca_topic_score_gemma":0.0038134435,"teacher_disagreement_score":0.014765923,"about_ca_system_score_codex":0.0016806936,"about_ca_system_score_gemma":0.0021850325,"threshold_uncertainty_score":0.07809055},"labels":[],"label_agreement":null},{"id":"W2940434429","doi":"10.1002/sim.8152","title":"Quantile regression and empirical likelihood for the analysis of longitudinal data with monotone missing responses due to dropout, with applications to quality of life measurements from clinical trials","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada; National Natural Science Foundation of China","keywords":"Missing data; Estimator; Quantile; Dropout (neural networks); Quantile regression; Statistics; Econometrics; Skewness; Regression analysis; Regression; Computer science; Mathematics; Machine learning","score_opus":0.6491396043007869,"score_gpt":0.6197232317086139,"score_spread":0.029416372592173,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2940434429","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0027504326,0.001883116,0.99435127,0.000459664,0.000043719097,0.000039776678,0.00009410027,0.00015161508,0.00022626393],"genre_scores_gemma":[0.26740092,0.010241234,0.7144491,0.00075166905,0.0009399072,0.0011667243,0.0011492353,0.00046559385,0.003435568],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98580813,0.011217737,0.00048705933,0.0011332365,0.001085069,0.00026877865],"domain_scores_gemma":[0.9242741,0.066801734,0.0035708128,0.0028523381,0.0021127393,0.00038834254],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.036756948,0.0013916707,0.0024581861,0.0031976555,0.00059994,0.001591645,0.003334114,0.0021165693,0.002502938],"category_scores_gemma":[0.11336849,0.0010544459,0.0020707706,0.003952262,0.0026912547,0.0029048768,0.002416799,0.0042197416,0.00072800706],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00034360847,0.00019795547,0.013390919,0.001340318,0.0009431313,0.000686572,0.0006455827,0.403335,0.0022180958,0.3321569,0.0043723746,0.24036962],"study_design_scores_gemma":[0.0000611391,0.00010630449,0.0023212226,0.00009619944,0.000071531496,0.00015276551,0.00007045065,0.81657124,0.00056195614,0.1760588,0.0038786402,0.000049613718],"about_ca_topic_score_codex":0.0034252815,"about_ca_topic_score_gemma":0.0016687313,"teacher_disagreement_score":0.036756948,"about_ca_system_score_codex":0.0012756949,"about_ca_system_score_gemma":0.0021954149,"threshold_uncertainty_score":0.19439167},"labels":[],"label_agreement":null},{"id":"W2946394381","doi":"10.1002/sim.8170","title":"Partitioning of time trends in prevalence and mortality of lung cancer","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Global Cancer Incidence and Screening","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Aging","funders":"National Institute on Aging; Shandong Academy of Sciences","keywords":"Lung cancer; Incidence (geometry); Medicine; Mortality rate; Demography; Epidemiology; Relative survival; Adenocarcinoma; Cancer; Disease; Stage (stratigraphy); Internal medicine; Cancer registry; Biology","score_opus":0.04763779418145317,"score_gpt":0.41391364758910515,"score_spread":0.366275853407652,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2946394381","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9712758,0.0013234441,0.019986257,0.00036331458,0.000034644054,0.00039324415,0.0037980496,0.0000605023,0.0027647358],"genre_scores_gemma":[0.99029106,0.0002823022,0.005201979,0.000050728522,0.000017812497,0.00013846079,0.0036689737,0.000017111957,0.00033154537],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9954259,0.0021786848,0.00035797802,0.0010482848,0.00068621774,0.00030295097],"domain_scores_gemma":[0.99448544,0.00216776,0.0015589722,0.0006273725,0.0010002747,0.0001602689],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005816236,0.00038297294,0.0003241505,0.0027690202,0.00029316818,0.0009084884,0.0004946907,0.00030115765,0.0010769586],"category_scores_gemma":[0.012937902,0.00020006229,0.0012277997,0.0021802597,0.0003386821,0.00055722136,0.00087568414,0.0003786982,0.00025669418],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00024099703,0.000024527553,0.9602616,0.00018889103,0.00060746784,0.00008167333,0.00079489243,0.0046915277,0.0015746438,0.0014021363,0.0005031981,0.029628381],"study_design_scores_gemma":[0.000005184571,0.00011913806,0.9910787,0.000034236225,0.00008652826,0.00015162451,0.00046151667,0.0052952445,0.00040770034,0.0010137915,0.0013314274,0.000014989068],"about_ca_topic_score_codex":0.009391805,"about_ca_topic_score_gemma":0.006922302,"teacher_disagreement_score":0.009391805,"about_ca_system_score_codex":0.0011019947,"about_ca_system_score_gemma":0.0008446605,"threshold_uncertainty_score":0.030759573},"labels":[],"label_agreement":null},{"id":"W2946435821","doi":"10.1002/sim.8203","title":"Adjusting for differential misclassification in matched case‐control studies utilizing health administrative data","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Manitoba; Nova Scotia Health Authority; Dalhousie University; Vancouver Coastal Health; University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada; National Multiple Sclerosis Society","keywords":"Bayesian probability; Observational study; Bayes' theorem; Computer science; Disease; Leverage (statistics); Econometrics; Differential (mechanical device); Data mining; Medicine; Statistics; Machine learning; Artificial intelligence; Mathematics","score_opus":0.3835311399523762,"score_gpt":0.535968901803228,"score_spread":0.15243776185085178,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2946435821","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06818583,0.0035387487,0.9235108,0.0018946478,0.00041535436,0.0005810394,0.00028329663,0.00025845767,0.0013318168],"genre_scores_gemma":[0.6930929,0.0013765673,0.3013871,0.0015890035,0.00025657463,0.0010438889,0.00047800737,0.00008960668,0.00068638835],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.81690437,0.14668736,0.009840374,0.013514549,0.011596563,0.0014566664],"domain_scores_gemma":[0.71755934,0.21284547,0.029659273,0.0339266,0.005376654,0.0006326169],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.19008614,0.0012524399,0.0019486888,0.0038860342,0.0014309998,0.002423414,0.0035649643,0.0021977718,0.0009820921],"category_scores_gemma":[0.42861068,0.0010784838,0.0032919091,0.0047238027,0.0028719476,0.002951716,0.0033491147,0.0020910664,0.0002109768],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001516542,0.00034732668,0.39749214,0.0023441704,0.016814178,0.001001334,0.00574596,0.03765221,0.0024111895,0.1924227,0.0036084366,0.3386438],"study_design_scores_gemma":[0.00058367965,0.0010742865,0.18145615,0.00137509,0.006421447,0.0011823103,0.0010653207,0.16328652,0.006310513,0.61949766,0.017457293,0.00028967683],"about_ca_topic_score_codex":0.0072417436,"about_ca_topic_score_gemma":0.006132742,"teacher_disagreement_score":0.8099139,"about_ca_system_score_codex":0.001708,"about_ca_system_score_gemma":0.0027307787,"threshold_uncertainty_score":0.998768},"labels":[],"label_agreement":null},{"id":"W2946565097","doi":"10.1002/sim.8194","title":"To add or not to add a new treatment arm to a multiarm study: A decision‐theoretic framework","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada","keywords":"Computer science; Medicine","score_opus":0.4236848403530717,"score_gpt":0.6166050279663471,"score_spread":0.1929201876132754,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2946565097","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009748085,0.003899236,0.9710824,0.008711103,0.00027108373,0.0011994381,0.00019696594,0.00007256913,0.0048192623],"genre_scores_gemma":[0.30426,0.0037994848,0.68129563,0.0034564428,0.0010044492,0.003735556,0.00020634195,0.00009121385,0.0021509097],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8886953,0.098128155,0.0028809502,0.0042656474,0.004759797,0.001270149],"domain_scores_gemma":[0.7852915,0.19739087,0.006988791,0.0042351857,0.0030823243,0.0030113286],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.11831375,0.0025350403,0.0044350303,0.0032233787,0.0009587923,0.0053472235,0.004171201,0.0049344907,0.004497319],"category_scores_gemma":[0.13484858,0.0013613792,0.003001592,0.0021094438,0.008317143,0.006839151,0.00349761,0.0061303093,0.00064452447],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010069511,0.00034501398,0.0015388061,0.001542397,0.0004979635,0.00022573085,0.00037224713,0.09688696,0.0009865802,0.82490027,0.0017438647,0.06995314],"study_design_scores_gemma":[0.00060502865,0.0012337104,0.0004908618,0.0004108505,0.0002410448,0.00014038073,0.00007167909,0.15886678,0.00071437715,0.83117664,0.005960761,0.000087853296],"about_ca_topic_score_codex":0.0007068979,"about_ca_topic_score_gemma":0.0006241421,"teacher_disagreement_score":0.8816863,"about_ca_system_score_codex":0.003680414,"about_ca_system_score_gemma":0.0068472526,"threshold_uncertainty_score":0.62571037},"labels":[],"label_agreement":null},{"id":"W2948643372","doi":"10.1002/sim.8213","title":"Analysis of clustered failure time data with cure fraction using copula","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"National Center for Advancing Translational Sciences; Assam Science Technology and Environment Council","keywords":"Akaike information criterion; Jackknife resampling; Copula (linguistics); Estimator; Statistics; Logistic regression; Computer science; Econometrics; Mathematics","score_opus":0.031513655357949436,"score_gpt":0.3467253180189006,"score_spread":0.3152116626609512,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2948643372","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.058327414,0.00069868943,0.93969256,0.00017419743,0.0000309368,0.000084135856,0.0002726506,0.0002832937,0.00043617375],"genre_scores_gemma":[0.7967022,0.0014924116,0.19656247,0.00016496168,0.00017958821,0.0003948317,0.0019053298,0.00023912656,0.0023590806],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99671197,0.00174439,0.0001291079,0.00073155755,0.00048192823,0.00020103931],"domain_scores_gemma":[0.9805202,0.014122209,0.0023221262,0.0014861451,0.0012262656,0.00032314405],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008749362,0.0015121404,0.002803756,0.003840741,0.0005940557,0.0014848278,0.0023156847,0.0016713159,0.0016007951],"category_scores_gemma":[0.024637103,0.00087483606,0.0025847638,0.0034202652,0.0011694155,0.0018133293,0.0016556673,0.0023043007,0.00049324165],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00020729301,0.00017591707,0.022886917,0.00039983878,0.0008719471,0.00079930894,0.00056798535,0.84034157,0.0035469662,0.061454628,0.0018775925,0.06687005],"study_design_scores_gemma":[0.0000074631566,0.00003508959,0.0025568567,0.000019418492,0.000051424227,0.000092521426,0.000041334995,0.98325646,0.00031260037,0.01307134,0.0005285145,0.00002695317],"about_ca_topic_score_codex":0.005050379,"about_ca_topic_score_gemma":0.0029552393,"teacher_disagreement_score":0.008749362,"about_ca_system_score_codex":0.00095698755,"about_ca_system_score_gemma":0.0011644029,"threshold_uncertainty_score":0.046271622},"labels":[],"label_agreement":null},{"id":"W2950576350","doi":"10.1002/sim.8180","title":"Four papers on child growth modelling","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Child Nutrition and Water Access","field":"Nursing","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Child development; Resource (disambiguation); Quarter (Canadian coin); Child health; Task (project management); Work (physics); Global health; Psychology; Medicine; Pediatrics; Environmental health; Developmental psychology; Computer science; Public health; Geography; Engineering; Nursing","score_opus":0.020261413899094544,"score_gpt":0.2850264011441614,"score_spread":0.26476498724506686,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2950576350","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007148045,0.18296707,0.39083025,0.060687028,0.019220464,0.00034071426,0.009275002,0.0029507459,0.3265806],"genre_scores_gemma":[0.11750691,0.2720643,0.29862508,0.012695895,0.009443035,0.000982035,0.028797483,0.005154039,0.25473133],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9964431,0.0011626761,0.00028098823,0.0006275279,0.001327176,0.00015846157],"domain_scores_gemma":[0.9903212,0.0063152565,0.000316861,0.0010484358,0.0016784022,0.00031983442],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005545311,0.0018992932,0.0011223572,0.0043290206,0.0014463912,0.005444483,0.0036590467,0.0033026058,0.03636557],"category_scores_gemma":[0.024996758,0.0008930235,0.002842645,0.01033095,0.0013589491,0.00405117,0.003809236,0.003932087,0.010597318],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009726614,0.00008748957,0.0031427776,0.0013809382,0.00020362606,0.00037590667,0.0006531039,0.024525203,0.00022479951,0.21003518,0.2917826,0.46749127],"study_design_scores_gemma":[0.00001662331,0.000020434234,0.0015302541,0.0013698847,0.00005715269,0.000259421,0.00023202466,0.010785257,0.0003400062,0.089082144,0.8962514,0.000055402896],"about_ca_topic_score_codex":0.016248334,"about_ca_topic_score_gemma":0.012537834,"teacher_disagreement_score":0.03636557,"about_ca_system_score_codex":0.0044797254,"about_ca_system_score_gemma":0.0032606202,"threshold_uncertainty_score":0.12165493},"labels":[],"label_agreement":null},{"id":"W2954707123","doi":"10.1002/sim.8281","title":"The Integrated Calibration Index (ICI) and related metrics for quantifying the calibration of logistic regression models","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":313,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Work & Health; Institute for Clinical Evaluative Sciences; Institute of Health Services and Policy Research; Sunnybrook Hospital; University of Toronto","funders":"National Institute of Neurological Disorders and Stroke; National Institutes of Health; Ontario Ministry of Health and Long-Term Care; Canadian Institutes of Health Research; Patient-Centered Outcomes Research Institute; Heart and Stroke Foundation of Canada","keywords":"Calibration; Statistics; Percentile; Logistic regression; Mathematics; Range (aeronautics); Smoothing; Regression; Computer science","score_opus":0.2038005309243087,"score_gpt":0.4336053912476601,"score_spread":0.22980486032335143,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2954707123","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03875963,0.0039488627,0.94743985,0.0007626055,0.00021099097,0.00026317794,0.0015087038,0.0012172613,0.00588888],"genre_scores_gemma":[0.58087504,0.002938396,0.40589643,0.0010096269,0.00058705185,0.0012818638,0.0043946835,0.0008760622,0.0021408745],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9738236,0.012113246,0.0023536566,0.0034718704,0.007594933,0.0006427259],"domain_scores_gemma":[0.781864,0.16632119,0.02338548,0.016278783,0.010940295,0.0012102724],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.044023037,0.0026421011,0.0022619092,0.012069485,0.0010930123,0.0033764825,0.002287825,0.0033141468,0.0023120178],"category_scores_gemma":[0.21744926,0.0007763252,0.00215769,0.00910558,0.0033340994,0.005532961,0.0038746912,0.0049568703,0.0009342631],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00048049178,0.0002723754,0.14429815,0.0009911953,0.0017843065,0.00034025667,0.00083201064,0.48820096,0.003586958,0.07502525,0.012650249,0.27153778],"study_design_scores_gemma":[0.000077186545,0.0006526493,0.07085264,0.000582044,0.00030676887,0.0012023369,0.00040086746,0.7384056,0.0064728633,0.1664782,0.014051221,0.0005176348],"about_ca_topic_score_codex":0.0027265188,"about_ca_topic_score_gemma":0.0022290025,"teacher_disagreement_score":0.044023037,"about_ca_system_score_codex":0.001955717,"about_ca_system_score_gemma":0.0020999738,"threshold_uncertainty_score":0.23281884},"labels":[],"label_agreement":null},{"id":"W2955494793","doi":"10.1002/sim.8283","title":"Dependence modeling for multi‐type recurrent events via copulas","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Actua; University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Copula (linguistics); Inference; Multivariate statistics; Econometrics; Marginal model; Computer science; Poisson distribution; Marginal likelihood; Marginal distribution; Statistics; Mathematics; Artificial intelligence; Machine learning; Regression analysis; Random variable; Bayesian probability","score_opus":0.22345526336841892,"score_gpt":0.48588751595579366,"score_spread":0.26243225258737474,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2955494793","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0104922475,0.0002765314,0.9880884,0.00016256925,0.000024112933,0.000033737295,0.00010865095,0.00011503901,0.00069871644],"genre_scores_gemma":[0.73964375,0.0022798134,0.24480796,0.00045156435,0.000310045,0.0005645085,0.0010995006,0.0004056329,0.0104372995],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9948965,0.0026882584,0.0002488205,0.0010213595,0.0007074996,0.00043758657],"domain_scores_gemma":[0.9719821,0.02093725,0.0031063566,0.00222801,0.0012181883,0.00052820117],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014214406,0.0016304541,0.002264446,0.00229021,0.000759714,0.002871097,0.004169282,0.0021861182,0.0041893236],"category_scores_gemma":[0.037546158,0.0015044507,0.0031466598,0.002032153,0.002929969,0.0046230583,0.0033677744,0.0043762904,0.0008051697],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009727981,0.000110059875,0.007730406,0.00018975897,0.00043286977,0.00093398854,0.00062437245,0.36338162,0.0017554096,0.5980116,0.0016796465,0.02505287],"study_design_scores_gemma":[0.000009814226,0.0000255675,0.0009128976,0.000024706093,0.000050930572,0.00011789956,0.00004548681,0.8632837,0.00025141786,0.13438122,0.0008657871,0.000030579493],"about_ca_topic_score_codex":0.004623394,"about_ca_topic_score_gemma":0.0038095198,"teacher_disagreement_score":0.014214406,"about_ca_system_score_codex":0.0015776737,"about_ca_system_score_gemma":0.001157848,"threshold_uncertainty_score":0.075173914},"labels":[],"label_agreement":null},{"id":"W2956065293","doi":"10.1002/sim.8497","title":"Estimation of treatment effects following a sequential trial of multiple treatments","year":2020,"lang":"en","type":"preprint","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council Canada; Department of Health and Social Care; Medical Research Council; National Institute for Health and Care Research","keywords":"Frequentist inference; Interim; Interim analysis; Confidence interval; Statistics; Replicate; Econometrics; Computer science; Nominal level; Bayesian probability; Mathematics; Randomized controlled trial; Medicine; Bayesian inference","score_opus":0.4719794924011268,"score_gpt":0.5884031063488133,"score_spread":0.11642361394768652,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2956065293","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01696607,0.0009724599,0.9769941,0.00060516596,0.00035601156,0.0025237168,0.00017379333,0.00025208312,0.0011566881],"genre_scores_gemma":[0.2802555,0.0017266384,0.7004708,0.0006081278,0.00043271645,0.011859608,0.00048442854,0.000096735224,0.0040654424],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.89792573,0.08556129,0.003356666,0.005818952,0.006142815,0.0011945915],"domain_scores_gemma":[0.7157639,0.25738636,0.012227964,0.009845845,0.003614948,0.001160994],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.12349098,0.0021188357,0.006497925,0.0024583251,0.0006842473,0.0029106182,0.0032694503,0.0038960716,0.0089782905],"category_scores_gemma":[0.23309992,0.00208615,0.004200552,0.0025453265,0.003702242,0.0045595923,0.002919308,0.004606746,0.0009448315],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009719646,0.0010182672,0.007961831,0.004507169,0.0031014802,0.00055558985,0.0011720508,0.2192448,0.0023353426,0.39850092,0.0036934298,0.34818944],"study_design_scores_gemma":[0.0029402943,0.0057906806,0.0025671716,0.0006745881,0.0009927787,0.00024593464,0.0001264022,0.5592469,0.0034814714,0.41850188,0.0052798414,0.00015192323],"about_ca_topic_score_codex":0.0010894763,"about_ca_topic_score_gemma":0.0009667248,"teacher_disagreement_score":0.876509,"about_ca_system_score_codex":0.0022380722,"about_ca_system_score_gemma":0.0049070283,"threshold_uncertainty_score":0.65309054},"labels":[],"label_agreement":null},{"id":"W2956147361","doi":"10.1002/sim.8741","title":"Formulating causal questions and principled statistical answers","year":2020,"lang":"en","type":"preprint","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Natural Sciences and Engineering Research Council of Canada; Lorentz Center; Medical Research Council; Vetenskapsrådet","keywords":"Causal inference; Covariate; Outcome (game theory); Computer science; Observational study; Set (abstract data type); Confounding; Variable (mathematics); Code (set theory); Statistical inference; Inference; Propensity score matching; Econometrics; Data science; Machine learning; Statistics; Artificial intelligence; Mathematics","score_opus":0.15779582121857488,"score_gpt":0.47360673047319685,"score_spread":0.31581090925462196,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2956147361","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00044627694,0.0061883004,0.97662145,0.012356399,0.00058881677,0.00025941947,0.0003845792,0.00021712547,0.00293774],"genre_scores_gemma":[0.038891524,0.016165033,0.9273694,0.0074301204,0.0032728543,0.003974911,0.00088024786,0.00033718706,0.001678754],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9023748,0.07867232,0.0051823216,0.00503629,0.008045228,0.00068896136],"domain_scores_gemma":[0.62990654,0.3495545,0.0061766873,0.008421992,0.005406196,0.0005341098],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.11979465,0.0026848326,0.004178431,0.005885442,0.001640779,0.0076009077,0.006059299,0.00774062,0.014402317],"category_scores_gemma":[0.2952356,0.0023243465,0.005118921,0.004722359,0.011939158,0.009630354,0.0057886597,0.013450193,0.0024526336],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000030455141,0.000030241548,0.00030781992,0.001942093,0.00027865786,0.00009609452,0.00038669442,0.0070711467,0.00006946,0.9345165,0.0058895126,0.04938129],"study_design_scores_gemma":[0.000024326133,0.0000086711125,0.000047641748,0.00045484776,0.000025167788,0.000027746366,0.000035340418,0.005008957,0.00006985941,0.9882237,0.0060600378,0.000013543975],"about_ca_topic_score_codex":0.0025959327,"about_ca_topic_score_gemma":0.0019088048,"teacher_disagreement_score":0.88020533,"about_ca_system_score_codex":0.0041132625,"about_ca_system_score_gemma":0.007788996,"threshold_uncertainty_score":0.63354224},"labels":[],"label_agreement":null},{"id":"W2961162542","doi":"10.1002/sim.8316","title":"Bayesian consensus‐based sample size criteria for binomial proportions","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"","keywords":"Prior probability; Frequentist inference; Sample size determination; Bayesian probability; Statistics; Econometrics; Credible interval; Sample (material); Mathematics; Point estimation; Bayes' theorem; Computer science; Bayesian inference","score_opus":0.14867597247685566,"score_gpt":0.5096998485410166,"score_spread":0.3610238760641609,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2961162542","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0027388628,0.00064422784,0.99019486,0.00067809905,0.00014710907,0.00082802656,0.00017406655,0.00020885516,0.004385878],"genre_scores_gemma":[0.08269298,0.0005907317,0.9069695,0.0009392407,0.0003239064,0.0062389034,0.0004314012,0.0002701744,0.0015430811],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8813911,0.08357892,0.006062664,0.0074232984,0.020179532,0.0013645237],"domain_scores_gemma":[0.6058661,0.34820652,0.009350507,0.013694951,0.021018075,0.0018638907],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.13537778,0.0019105133,0.0042254897,0.005732763,0.0018206318,0.0037934338,0.005676676,0.00488353,0.009482085],"category_scores_gemma":[0.44442594,0.0011322016,0.0028188133,0.0030120842,0.005406072,0.006454331,0.005125268,0.006381859,0.0020561246],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010787896,0.00018247265,0.004375207,0.0017906321,0.00042314373,0.0002260496,0.0012080078,0.07158306,0.002233431,0.68881905,0.010630622,0.21744958],"study_design_scores_gemma":[0.00051361625,0.00046171164,0.0018205475,0.00094778417,0.00017660914,0.00032914698,0.00021994358,0.19203305,0.0035631473,0.7817197,0.018066555,0.0001481944],"about_ca_topic_score_codex":0.0011204894,"about_ca_topic_score_gemma":0.0009622111,"teacher_disagreement_score":0.13537778,"about_ca_system_score_codex":0.0029667036,"about_ca_system_score_gemma":0.003913818,"threshold_uncertainty_score":0.71595466},"labels":[],"label_agreement":null},{"id":"W2963325162","doi":"10.1002/sim.8327","title":"Information content of stepped‐wedge designs when treatment effect heterogeneity and/or implementation periods are present","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"National Health and Medical Research Council; National Medical Research Council; Medical Research Council","keywords":"Computer science; Sequence (biology); Cluster (spacecraft); Wedge (geometry); Econometrics; Treatment effect; Statistics; Data mining; Mathematics; Medicine; Biology","score_opus":0.510180498051622,"score_gpt":0.5711178178960157,"score_spread":0.060937319844393656,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2963325162","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07267818,0.004172735,0.90664035,0.0020438877,0.00029418623,0.003339893,0.0026431289,0.00046857254,0.0077191885],"genre_scores_gemma":[0.56451255,0.0024813039,0.4139568,0.0016336206,0.0003877065,0.011025149,0.0027871828,0.0002351274,0.0029805987],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8647347,0.10837894,0.007869549,0.008984711,0.008931538,0.0011005001],"domain_scores_gemma":[0.39577398,0.5421388,0.02441094,0.031799454,0.0048842393,0.0009925996],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.113087386,0.0011637121,0.0035313112,0.0031738288,0.0007930861,0.0040045627,0.0027315333,0.0032221796,0.0069809076],"category_scores_gemma":[0.42184496,0.0013719545,0.0023474244,0.0039083264,0.0042169956,0.007335979,0.0030227453,0.003148226,0.0007834545],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.014061986,0.00083333015,0.018334003,0.010028811,0.00417637,0.00092201,0.0030878785,0.07899179,0.006058569,0.40336627,0.008246866,0.4518921],"study_design_scores_gemma":[0.0016993085,0.0028126382,0.010553279,0.002086234,0.0021500743,0.00072134566,0.000331396,0.20186989,0.007824214,0.75611496,0.013607303,0.00022933031],"about_ca_topic_score_codex":0.00039211242,"about_ca_topic_score_gemma":0.000343586,"teacher_disagreement_score":0.8869126,"about_ca_system_score_codex":0.001960522,"about_ca_system_score_gemma":0.0027064765,"threshold_uncertainty_score":0.5980704},"labels":[],"label_agreement":null},{"id":"W2963467474","doi":"10.1002/sim.8299","title":"Reconciling randomized trial evidence on proximal versus distal outcomes, with application to trials of influenza vaccination for healthcare workers","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Influenza Virus Research Studies","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; Vancouver Coastal Health","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Outcome (game theory); Medicine; Randomized controlled trial; Context (archaeology); Psychological intervention; Intervention (counseling); Health care; Clinical trial; Intensive care medicine; Surgery; Internal medicine","score_opus":0.20998245380367037,"score_gpt":0.5147954252479449,"score_spread":0.3048129714442745,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2963467474","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010930126,0.012772598,0.95845217,0.011341543,0.00096810877,0.0020685464,0.00043545914,0.00033680865,0.0026947171],"genre_scores_gemma":[0.3562079,0.0044903965,0.6218931,0.008160371,0.0011353712,0.0066273194,0.0003740436,0.0002451359,0.00086639135],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.19939046,0.750542,0.022866387,0.009622184,0.016775941,0.00080303475],"domain_scores_gemma":[0.1134731,0.81578636,0.032477792,0.0320436,0.005494859,0.0007242918],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.56741935,0.003872892,0.008519431,0.010187943,0.0017899113,0.010367883,0.0053845476,0.008440253,0.0051688463],"category_scores_gemma":[0.8054254,0.002453937,0.013048617,0.010370956,0.012010671,0.008241405,0.009086231,0.008842828,0.00052318966],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.006083173,0.0003052687,0.016843723,0.024352303,0.054964244,0.0011487391,0.0032614472,0.15287247,0.001520899,0.4955448,0.0048285793,0.23827429],"study_design_scores_gemma":[0.0022767186,0.001655362,0.00458046,0.004969435,0.00878298,0.0003882971,0.000365457,0.09997967,0.0013218045,0.86401284,0.011420347,0.00024664632],"about_ca_topic_score_codex":0.0020877526,"about_ca_topic_score_gemma":0.0019298943,"teacher_disagreement_score":0.43258065,"about_ca_system_score_codex":0.005948661,"about_ca_system_score_gemma":0.008874928,"threshold_uncertainty_score":0.533449},"labels":[],"label_agreement":null},{"id":"W2963571436","doi":"10.1002/sim.8318","title":"A new perspective on loss to follow‐up in failure time and life history studies","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Birth, Development, and Health","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Perspective (graphical); Time perspective; Computer science; Econometrics; Psychology; Mathematics; Artificial intelligence","score_opus":0.034291194211084945,"score_gpt":0.35064325690880327,"score_spread":0.3163520626977183,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2963571436","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009887273,0.01660503,0.940668,0.024460545,0.001257996,0.00010018428,0.00036245282,0.00008742671,0.006571199],"genre_scores_gemma":[0.5506039,0.029904809,0.38721436,0.010112027,0.009902625,0.0013138636,0.00057289546,0.00019383316,0.010181738],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98295045,0.012635507,0.000730623,0.0015304318,0.0016326107,0.00052041025],"domain_scores_gemma":[0.8984975,0.08863285,0.004552599,0.0047286297,0.0021827365,0.0014057088],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03705262,0.0014603566,0.002205519,0.004628969,0.0015850383,0.005209022,0.004731631,0.0041045034,0.0038231106],"category_scores_gemma":[0.06450082,0.0009394125,0.0029305345,0.0035922045,0.007824909,0.008035829,0.0041542943,0.00875318,0.0003574622],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000022741047,0.00005309752,0.0034778877,0.00018241337,0.00011823744,0.00027210763,0.00089527667,0.012178764,0.000107861546,0.965666,0.001533342,0.015492239],"study_design_scores_gemma":[0.000021669635,0.00014060074,0.0013118929,0.0002595957,0.00006863241,0.00036403557,0.000306222,0.035441324,0.000082402665,0.94895476,0.01299774,0.000051250878],"about_ca_topic_score_codex":0.0049809692,"about_ca_topic_score_gemma":0.0034019414,"teacher_disagreement_score":0.03705262,"about_ca_system_score_codex":0.003588344,"about_ca_system_score_gemma":0.003517644,"threshold_uncertainty_score":0.19595534},"labels":[],"label_agreement":null},{"id":"W2963777543","doi":"10.1002/sim.8297","title":"Stein‐type shrinkage estimators in gamma regression model with application to prostate cancer data","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Winnipeg; University of Manitoba","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Estimator; Statistics; Regression analysis; Mathematics; Regression; Linear regression; Proportional hazards model; Computer science; Econometrics; Applied mathematics","score_opus":0.025846596910134332,"score_gpt":0.3604758722269298,"score_spread":0.3346292753167954,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2963777543","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00867554,0.00036862286,0.99031454,0.00014450493,0.00001662316,0.000025177278,0.000026852547,0.000088303685,0.00033988696],"genre_scores_gemma":[0.33211672,0.0025148296,0.6603477,0.00022610444,0.00017708678,0.00039023705,0.00039265296,0.00018287112,0.0036518201],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9965856,0.0023450288,0.00013776861,0.00029742278,0.00051174144,0.00012225848],"domain_scores_gemma":[0.98127365,0.0151583245,0.0009816436,0.001069657,0.0013649135,0.00015187712],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.015100296,0.0009304592,0.0011247564,0.0016460185,0.00046096346,0.00077838043,0.0014298959,0.0014124278,0.001113298],"category_scores_gemma":[0.048195098,0.0005246699,0.0011635593,0.0019134999,0.0014345906,0.0018765846,0.0017981216,0.0018868705,0.00034995817],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015624671,0.00008285066,0.00729348,0.00039004112,0.0002270486,0.00031669938,0.00041333205,0.57558846,0.004190427,0.19601089,0.002559763,0.2127707],"study_design_scores_gemma":[0.00002286415,0.00005344281,0.0010288422,0.000031738917,0.000035443165,0.00010968441,0.00003391186,0.95190215,0.0013624857,0.04399666,0.0013892609,0.000033473778],"about_ca_topic_score_codex":0.0025940298,"about_ca_topic_score_gemma":0.0027793664,"teacher_disagreement_score":0.015100296,"about_ca_system_score_codex":0.00065017055,"about_ca_system_score_gemma":0.0013095252,"threshold_uncertainty_score":0.07985896},"labels":[],"label_agreement":null},{"id":"W2964523523","doi":"10.1002/sim.8286","title":"Measuring variability between clusters by subgroup: An extension of the median odds ratio","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Mount Sinai Hospital; Health Sciences Centre; University of Toronto; Sunnybrook Health Science Centre; Institute for Work & Health","funders":"Canadian Institutes of Health Research; Physicians' Services Incorporated Foundation; Heart and Stroke Foundation of Canada","keywords":"Extension (predicate logic); Odds; Odds ratio; Subgroup analysis; Statistics; Demography; Mathematics; Computer science; Confidence interval; Logistic regression; Sociology","score_opus":0.07676482540831311,"score_gpt":0.37053227655468124,"score_spread":0.29376745114636815,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2964523523","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029718624,0.0016466223,0.9647191,0.00092734623,0.00014929751,0.00027076833,0.0005276571,0.00022619551,0.0018143533],"genre_scores_gemma":[0.4356899,0.0013391711,0.5593248,0.0006366812,0.0004925866,0.0009892841,0.00060301163,0.00021520007,0.0007093814],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.92853004,0.052333195,0.003483947,0.008834415,0.0061050346,0.00071334315],"domain_scores_gemma":[0.7446463,0.21398756,0.015517217,0.019256102,0.0054192822,0.0011736052],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07108311,0.0012102972,0.0028954425,0.0057267677,0.0008494893,0.00366553,0.0035784931,0.001817372,0.002484937],"category_scores_gemma":[0.27709615,0.00074851926,0.0042869723,0.0060563623,0.0041579613,0.0069550457,0.004512013,0.0036944782,0.0003325564],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009870197,0.00031715765,0.23324826,0.0017986267,0.0061358237,0.0009926042,0.004251531,0.047502086,0.001969227,0.27400628,0.005170455,0.42362085],"study_design_scores_gemma":[0.00019407873,0.0013437116,0.07663964,0.000690343,0.001503241,0.0026016403,0.0014727373,0.24688397,0.0019258283,0.65134656,0.015009026,0.00038929872],"about_ca_topic_score_codex":0.0026451112,"about_ca_topic_score_gemma":0.0010976132,"teacher_disagreement_score":0.07108311,"about_ca_system_score_codex":0.0012383644,"about_ca_system_score_gemma":0.0015410417,"threshold_uncertainty_score":0.37592793},"labels":[],"label_agreement":null},{"id":"W2965457979","doi":"10.1002/sim.8306","title":"Untangling serially dependent underreported count data for gender‐based violence","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Intimate Partner and Family Violence","field":"Social Sciences","cited_by":50,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Ministerio de Economía y Competitividad; World Health Organization","keywords":"Count data; Statistics; Computer science; Econometrics; Mathematics","score_opus":0.09538025833044142,"score_gpt":0.4139783812437226,"score_spread":0.31859812291328116,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2965457979","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.57342577,0.0011587262,0.4163072,0.0017408531,0.00020846551,0.00021736023,0.0030887462,0.00036447085,0.0034884065],"genre_scores_gemma":[0.97960126,0.00042975947,0.015467653,0.00009183073,0.00009002287,0.00011773949,0.0018177214,0.000047838286,0.002336086],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9918207,0.00526655,0.0005075995,0.0013335876,0.0006734706,0.00039811843],"domain_scores_gemma":[0.92454207,0.051236898,0.014391707,0.008002182,0.0015090415,0.00031799334],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016041297,0.00057063764,0.00077062723,0.001871562,0.00041769946,0.00234339,0.0022579278,0.0009805163,0.0024454028],"category_scores_gemma":[0.055091992,0.00062558544,0.001315498,0.002595328,0.0012713718,0.0017176461,0.0015958878,0.0016740446,0.00049304473],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00022285902,0.00021684219,0.6957475,0.00029314889,0.0009762418,0.0005909195,0.0023893658,0.12759294,0.00086009694,0.08608713,0.002063671,0.08295919],"study_design_scores_gemma":[0.00002600483,0.00025498544,0.19157,0.000254406,0.00027888667,0.0003678367,0.0010301406,0.7417328,0.001461328,0.055813275,0.0071215974,0.000088787485],"about_ca_topic_score_codex":0.009711071,"about_ca_topic_score_gemma":0.008352223,"teacher_disagreement_score":0.016041297,"about_ca_system_score_codex":0.0013364741,"about_ca_system_score_gemma":0.0013903752,"threshold_uncertainty_score":0.08483547},"labels":[],"label_agreement":null},{"id":"W2966416868","doi":"10.1002/sim.8690","title":"Measurement error and precision medicine: Error‐prone tailoring covariates in dynamic treatment regimes","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Covariate; Precision medicine; Observational error; Context (archaeology); Accuracy and precision; Simple (philosophy); Error detection and correction","score_opus":0.1879669184940262,"score_gpt":0.4346654424349814,"score_spread":0.2466985239409552,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2966416868","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008443927,0.0015887807,0.9823685,0.004696418,0.00024119095,0.00016078827,0.00021381665,0.00014569033,0.002140826],"genre_scores_gemma":[0.515312,0.0027042069,0.4743597,0.0029466473,0.00078919297,0.00096522307,0.00032618138,0.00013306993,0.0024637647],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9109691,0.071642295,0.0026566552,0.007813376,0.005842624,0.0010760182],"domain_scores_gemma":[0.6473408,0.30119005,0.01698687,0.028790107,0.004791242,0.0009010306],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.082757264,0.0016545006,0.0031230266,0.0021906768,0.0015236384,0.0050592106,0.00395673,0.0040985085,0.00428763],"category_scores_gemma":[0.33802173,0.001149743,0.002661753,0.0033004438,0.0082056755,0.00733857,0.004927273,0.007608011,0.000536374],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00038592453,0.00013675721,0.010040212,0.0005469879,0.00085960835,0.00021286837,0.00091853726,0.088147044,0.0006411463,0.77843046,0.0023502896,0.11733011],"study_design_scores_gemma":[0.00015780683,0.00020940788,0.0019407333,0.00027887398,0.00020667155,0.00011779878,0.00008499987,0.109034665,0.0008229683,0.883035,0.004033303,0.00007784108],"about_ca_topic_score_codex":0.0046277298,"about_ca_topic_score_gemma":0.0033790653,"teacher_disagreement_score":0.91724277,"about_ca_system_score_codex":0.0031043403,"about_ca_system_score_gemma":0.0056739366,"threshold_uncertainty_score":0.4376675},"labels":[],"label_agreement":null},{"id":"W2967839411","doi":"10.1002/sim.8941","title":"Bayesian design and analysis of external pilot trials for complex interventions","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Programme Grants for Applied Research; Medical Research Council Canada; Department of Health and Social Care; Medical Research Council; National Institute for Health and Care Research","keywords":"Bayesian probability; Monte Carlo method; Piecewise; Markov chain Monte Carlo; Psychological intervention; Sample size determination; Design of experiments; Sensitivity (control systems); Research design","score_opus":0.45041322823576735,"score_gpt":0.5294635819590232,"score_spread":0.07905035372325586,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2967839411","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0078057405,0.00043245615,0.9880376,0.00043022764,0.00006060084,0.002113179,0.00010908369,0.00016972583,0.00084133644],"genre_scores_gemma":[0.16429354,0.0007698927,0.8192567,0.00047558857,0.00009867015,0.013644734,0.00024834598,0.000089074085,0.001123424],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8733771,0.1165696,0.002151318,0.0031924644,0.0038105554,0.00089893665],"domain_scores_gemma":[0.82986546,0.14666985,0.00991096,0.008084574,0.004150953,0.0013181773],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11567706,0.0021148087,0.003852372,0.0022367907,0.00083889544,0.0027141243,0.0030074122,0.0031397422,0.0047897655],"category_scores_gemma":[0.20391087,0.001767952,0.0028708186,0.0013345216,0.0037637667,0.0026835485,0.0037457033,0.0045742593,0.00063005363],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.007499036,0.00068438856,0.00559392,0.002397531,0.0019038501,0.00043517118,0.00096046535,0.33222923,0.0036945427,0.4517711,0.003504957,0.18932576],"study_design_scores_gemma":[0.0021796736,0.002215554,0.0020132402,0.0005042818,0.0006182595,0.00012860888,0.00006810632,0.70067644,0.0022791014,0.28387374,0.0053122914,0.00013073054],"about_ca_topic_score_codex":0.0012882787,"about_ca_topic_score_gemma":0.0013665643,"teacher_disagreement_score":0.11567706,"about_ca_system_score_codex":0.002538253,"about_ca_system_score_gemma":0.005413189,"threshold_uncertainty_score":0.6117661},"labels":[],"label_agreement":null},{"id":"W2969784647","doi":"10.1002/sim.8986","title":"Incorporating partial adherence into the principal stratification analysis framework","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; Providence Health Care; University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Inference; Estimator; Stratification (seeds); Principal (computer security); Computer science; Principal component analysis; Population; Clinical trial; Monte Carlo method; Statistics; Econometrics; Mathematics; Artificial intelligence; Medicine","score_opus":0.11815568465399728,"score_gpt":0.45816061350439785,"score_spread":0.34000492885040057,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2969784647","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0037216705,0.00034235383,0.9939161,0.0006459937,0.000086990105,0.00043306497,0.00008874263,0.00013790998,0.000627089],"genre_scores_gemma":[0.20622401,0.00067003537,0.7876383,0.0008932011,0.000286557,0.0028405695,0.0002762739,0.00010100316,0.0010700498],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.84817255,0.13712822,0.0032666593,0.005151569,0.005142147,0.001138847],"domain_scores_gemma":[0.8199343,0.15054555,0.0075014485,0.015256568,0.0056655286,0.0010966058],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1299705,0.0018852224,0.003425467,0.002831848,0.0016877186,0.0026865925,0.0026791554,0.0025433637,0.004159725],"category_scores_gemma":[0.22506355,0.0016273941,0.004515413,0.002597668,0.0040318286,0.0039377725,0.0047461027,0.0059289723,0.0007361686],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013334085,0.00031055827,0.019918123,0.001436887,0.0028117804,0.00059843087,0.0030323863,0.10762351,0.0012670208,0.6169737,0.0051172525,0.23957705],"study_design_scores_gemma":[0.0005092153,0.00048268534,0.0026738984,0.0002772245,0.0006433031,0.00019032812,0.00013007638,0.31163368,0.000655093,0.675956,0.006733848,0.000114638184],"about_ca_topic_score_codex":0.0048910608,"about_ca_topic_score_gemma":0.00535341,"teacher_disagreement_score":0.8700295,"about_ca_system_score_codex":0.0019344949,"about_ca_system_score_gemma":0.007641186,"threshold_uncertainty_score":0.68735796},"labels":[],"label_agreement":null},{"id":"W2969865418","doi":"10.1002/sim.8345","title":"On estimands arising from misspecified semiparametric rate‐based analysis of recurrent episodic conditions","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; GlaxoSmithKline","keywords":"Estimator; Econometrics; Semiparametric model; Statistics; Asymptotic distribution; Mathematics; Medicine","score_opus":0.09883824884580657,"score_gpt":0.43573433639621634,"score_spread":0.3368960875504098,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2969865418","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.028828546,0.00236808,0.96402895,0.0017298167,0.00009542415,0.00041929926,0.00035767155,0.0003544254,0.0018178511],"genre_scores_gemma":[0.6612028,0.004494474,0.3215766,0.0020390379,0.0004789387,0.0027171466,0.0013955502,0.0004234109,0.0056721647],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.93237257,0.05793007,0.00217861,0.0036954111,0.0028645443,0.0009588801],"domain_scores_gemma":[0.33968458,0.63427705,0.011890637,0.010651019,0.0027386239,0.0007580569],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.17181228,0.0020832631,0.0062143947,0.0028945864,0.0006328256,0.0039916583,0.0031029726,0.0033043786,0.0047215815],"category_scores_gemma":[0.48222435,0.0013788943,0.003726197,0.0024570627,0.005684885,0.0064621884,0.004361748,0.006123887,0.0005622703],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018429202,0.00026447914,0.034454565,0.0024589533,0.003060541,0.0025018256,0.0015761808,0.250522,0.0019903919,0.5728149,0.0037412678,0.12477205],"study_design_scores_gemma":[0.0003329043,0.0005082043,0.005537732,0.0005747974,0.0007236036,0.00080270995,0.00018768678,0.5590923,0.0016115189,0.42776328,0.0027523194,0.00011292249],"about_ca_topic_score_codex":0.0025673027,"about_ca_topic_score_gemma":0.0013153589,"teacher_disagreement_score":0.17181228,"about_ca_system_score_codex":0.0017583099,"about_ca_system_score_gemma":0.0025466755,"threshold_uncertainty_score":0.90864104},"labels":[],"label_agreement":null},{"id":"W2970007864","doi":"10.1002/sim.8344","title":"Overdispersion models for correlated multinomial data: Applications to blinding assessment","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Work & Health; Public Health Ontario; University of Toronto","funders":"National Center for Advancing Translational Sciences","keywords":"Overdispersion; Statistics; Estimator; Generalized estimating equation; Multinomial distribution; Intraclass correlation; Mathematics; Gee; Econometrics; Estimating equations; Negative binomial distribution; Quasi-likelihood; Poisson distribution","score_opus":0.15022813602425147,"score_gpt":0.48922819164064546,"score_spread":0.339000055616394,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2970007864","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0057084747,0.000662816,0.99214,0.0005120473,0.00007169966,0.00022518083,0.00009181071,0.00013167228,0.0004563771],"genre_scores_gemma":[0.2767572,0.0021971052,0.7130499,0.0013978883,0.0004205744,0.002299663,0.00042296486,0.00027813492,0.003176552],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.850364,0.12546031,0.0045128516,0.010793533,0.0076916236,0.0011776104],"domain_scores_gemma":[0.4756946,0.46102938,0.026501156,0.028774925,0.0069622835,0.0010376868],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.18340898,0.0021693478,0.0045217215,0.0035252406,0.0019821227,0.003565952,0.005347723,0.0044773,0.0045666536],"category_scores_gemma":[0.40271902,0.0017913431,0.0050949887,0.004925459,0.009416494,0.0062309047,0.0056231576,0.0059507303,0.00065391866],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004956335,0.0001490177,0.016772116,0.0013010644,0.0022817324,0.0008764765,0.002556316,0.14790703,0.0010530103,0.6962341,0.0030842912,0.12728932],"study_design_scores_gemma":[0.00022797621,0.00020972772,0.0030658718,0.00037953793,0.00044204903,0.00040053643,0.00019076576,0.2953544,0.0009481768,0.69394094,0.0046673613,0.00017265424],"about_ca_topic_score_codex":0.0054745083,"about_ca_topic_score_gemma":0.004000998,"teacher_disagreement_score":0.18340898,"about_ca_system_score_codex":0.0029882013,"about_ca_system_score_gemma":0.004144773,"threshold_uncertainty_score":0.96997094},"labels":[],"label_agreement":null},{"id":"W2970526340","doi":"10.1002/sim.8354","title":"A more intuitive and modern way to compute a small‐sample confidence interval for the mean of a Poisson distribution","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"","keywords":"Poisson distribution; Link (geometry); Confidence interval; Computer science; TRACE (psycholinguistics); Computation; Sample (material); Sample size determination; Statistics; Interval (graph theory); Distribution (mathematics); Mathematics; Algorithm; Mathematical analysis","score_opus":0.0681805430217062,"score_gpt":0.40129177527727117,"score_spread":0.33311123225556494,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2970526340","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00063284585,0.0015179076,0.982656,0.0040745083,0.0020787346,0.000053390417,0.00032997082,0.001002567,0.0076541137],"genre_scores_gemma":[0.044512525,0.002820748,0.92797387,0.00624267,0.0034773438,0.0005173983,0.0005841312,0.0021117902,0.011759519],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98664993,0.007419617,0.00108468,0.0016094976,0.0029617124,0.00027459313],"domain_scores_gemma":[0.93446666,0.050294887,0.0034969284,0.006284676,0.0049631414,0.0004936623],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017234376,0.0020862848,0.0014239027,0.0053022723,0.0015574085,0.0063498216,0.0033747116,0.0037208106,0.043331847],"category_scores_gemma":[0.16854797,0.0009570167,0.002573819,0.004904428,0.0048287245,0.009850107,0.0037620969,0.011588052,0.01848997],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000110008245,0.00008900145,0.00079552655,0.00055672857,0.00014339776,0.0001919059,0.0007766788,0.006009776,0.0007536007,0.8274989,0.055149544,0.107925035],"study_design_scores_gemma":[0.00007472416,0.00014848703,0.0006406261,0.00079681457,0.00008652371,0.00081952417,0.00025338546,0.01895144,0.002267596,0.7963402,0.17941058,0.00021019678],"about_ca_topic_score_codex":0.0019452575,"about_ca_topic_score_gemma":0.0015559375,"teacher_disagreement_score":0.043331847,"about_ca_system_score_codex":0.0016408805,"about_ca_system_score_gemma":0.0015044858,"threshold_uncertainty_score":0.14495939},"labels":[],"label_agreement":null},{"id":"W2971381351","doi":"10.1002/sim.8320","title":"Discussion of Schuemie et al: “A plea to stop using the case‐control design in retrospective database studies”","year":2019,"lang":"en","type":"letter","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":20,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Jewish General Hospital","funders":"","keywords":"Medicine; Pharmacoepidemiology; Epidemiology; Family medicine; Plea; Biostatistics; Pharmacoeconomics; Library science; Medical prescription; Internal medicine; Political science","score_opus":0.5841074927165224,"score_gpt":0.605698572106588,"score_spread":0.02159107939006555,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2971381351","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000117365285,0.0005017545,0.00032856944,0.9849615,0.013630553,0.000015778103,0.00006179292,0.000019963241,0.0003626244],"genre_scores_gemma":[0.0007434423,0.0002000861,0.0005462079,0.97773165,0.020309964,0.000052660416,0.00001165821,0.0000143597545,0.00039005704],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9097417,0.046217654,0.017438393,0.008392953,0.015099926,0.003109453],"domain_scores_gemma":[0.66179097,0.276204,0.015495068,0.0064751804,0.031693067,0.008341745],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.098382205,0.0013582626,0.0030264952,0.0025272376,0.006027517,0.008987318,0.006356857,0.08829217,0.0061275586],"category_scores_gemma":[0.35124502,0.002405147,0.0031359477,0.003330921,0.009411047,0.008865181,0.003555003,0.06534479,0.006666904],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00011666332,0.00003164973,0.000995032,0.0001807858,0.000068779,0.0007010362,0.00029849782,0.000096873395,0.00011093526,0.005477269,0.9838769,0.008045595],"study_design_scores_gemma":[0.0005670289,0.000144115,0.0025262125,0.0019939751,0.00023880035,0.0025927997,0.0009345961,0.0012160346,0.00075362436,0.036880244,0.9518338,0.00031874087],"about_ca_topic_score_codex":0.0067990385,"about_ca_topic_score_gemma":0.012875942,"teacher_disagreement_score":0.90161777,"about_ca_system_score_codex":0.0076933824,"about_ca_system_score_gemma":0.009745179,"threshold_uncertainty_score":0.52030104},"labels":[],"label_agreement":null},{"id":"W2973185263","doi":"10.1002/sim.8809","title":"SuRF: A new method for sparse variable selection, with application in microbiome data analysis","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Gut microbiota and health","field":"Biochemistry, Genetics and Molecular Biology","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Feature selection; Lasso (programming language); Ranking (information retrieval); Variable (mathematics); Selection (genetic algorithm); Microbiome; A priori and a posteriori; Interpretability","score_opus":0.028516412973631257,"score_gpt":0.3642394012441812,"score_spread":0.3357229882705499,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2973185263","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00088845,0.0001984141,0.99761665,0.00008297637,0.000043931912,0.000034298104,0.00015454038,0.00086888607,0.00011187979],"genre_scores_gemma":[0.02557253,0.00049848214,0.9691863,0.00029712296,0.00025234275,0.00051705574,0.0015173617,0.00082792336,0.0013309754],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9952095,0.002723138,0.00023551268,0.0006692319,0.0010076755,0.00015493082],"domain_scores_gemma":[0.99229264,0.0055013103,0.00042665636,0.00072757184,0.0008877923,0.00016404506],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00780257,0.0019064894,0.0022279923,0.0028673767,0.000929431,0.0012202936,0.0020536496,0.0015672954,0.0049297614],"category_scores_gemma":[0.018789977,0.0009166477,0.0027839786,0.0030452514,0.0011565004,0.0015730298,0.0023093082,0.0031317258,0.002095488],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00046378892,0.00016120638,0.005191082,0.00077725406,0.0011227417,0.0003602059,0.0002773696,0.1734377,0.011967183,0.028048271,0.026268978,0.75192416],"study_design_scores_gemma":[0.0001139356,0.00016033927,0.0014398478,0.0000803477,0.00010906798,0.00029540475,0.000046237594,0.9417927,0.004037027,0.033578414,0.018265542,0.00008124351],"about_ca_topic_score_codex":0.0033389519,"about_ca_topic_score_gemma":0.0041986336,"teacher_disagreement_score":0.00780257,"about_ca_system_score_codex":0.00044883363,"about_ca_system_score_gemma":0.0018489794,"threshold_uncertainty_score":0.041264474},"labels":[],"label_agreement":null},{"id":"W2982693598","doi":"10.1002/sim.8399","title":"A review of the use of time‐varying covariates in the Fine‐Gray subdistribution hazard competing risk regression model","year":2019,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":151,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; International Council for the Exploration of the Sea; Ontario Ministry of Health and Long-Term Care; Heart and Stroke Foundation of Canada","keywords":"Covariate; Proportional hazards model; Statistics; Gray (unit); Regression analysis; Econometrics; Regression; Mathematics; Medicine","score_opus":0.27567504430334744,"score_gpt":0.4708426292700219,"score_spread":0.19516758496667447,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2982693598","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00024153531,0.9921955,0.005271917,0.0010572269,0.00032525323,0.000028576147,0.00018925681,0.000019348132,0.00067133433],"genre_scores_gemma":[0.004464666,0.9879238,0.0052533746,0.0010232565,0.00064918044,0.00009590162,0.00024697135,0.000028066708,0.000314852],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99164283,0.004507151,0.0015352162,0.00074809045,0.0014418075,0.00012482416],"domain_scores_gemma":[0.9571891,0.038166717,0.002017598,0.0004377612,0.0020244499,0.00016434278],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.011992748,0.0013424314,0.0029322947,0.004783782,0.00033182988,0.0021992004,0.0019620154,0.0021085748,0.0049877213],"category_scores_gemma":[0.043264315,0.000820516,0.004896664,0.007907396,0.00084867695,0.0023252002,0.0008726228,0.0025158033,0.001279458],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003074879,0.00010303342,0.003479604,0.14083208,0.0028089131,0.0004380919,0.00044061968,0.005956118,0.00032902593,0.026051326,0.04148019,0.77777356],"study_design_scores_gemma":[0.0001480986,0.00045432645,0.008825173,0.13185246,0.007154505,0.0028229095,0.00030728837,0.0075707794,0.0006302115,0.03172141,0.80822325,0.00028968282],"about_ca_topic_score_codex":0.0062979185,"about_ca_topic_score_gemma":0.0057230433,"teacher_disagreement_score":0.98800725,"about_ca_system_score_codex":0.0019049123,"about_ca_system_score_gemma":0.0048431153,"threshold_uncertainty_score":0.06342447},"labels":[],"label_agreement":null},{"id":"W2986248248","doi":"10.1002/sim.8852","title":"Model diagnostics for censored regression via randomized survival probabilities","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University; University of Saskatchewan","funders":"Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada; Canada Foundation for Innovation","keywords":"Residual; Quantile; Statistical hypothesis testing; Regression; Regression analysis; Nonlinear regression; Nonlinear system; Studentized residual; Accelerated failure time model","score_opus":0.13827593050603806,"score_gpt":0.4197208800041434,"score_spread":0.2814449494981054,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2986248248","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.027474163,0.00022594686,0.9708185,0.00029003882,0.00003211991,0.00008367955,0.00016652716,0.00045512713,0.0004539108],"genre_scores_gemma":[0.7075624,0.00034135542,0.28920063,0.0003123479,0.00010766269,0.00051302364,0.0007663333,0.00015998182,0.0010362745],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9852181,0.010827584,0.00046290806,0.001986937,0.001181735,0.00032273843],"domain_scores_gemma":[0.8755313,0.107213035,0.0075114784,0.006812456,0.0023328033,0.0005989209],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.021894231,0.0010233554,0.0014586255,0.0025713756,0.00051286834,0.0013417503,0.0024183793,0.0016489171,0.0024116165],"category_scores_gemma":[0.13389416,0.0004976165,0.0015453729,0.001548183,0.0037473997,0.0032033478,0.002034182,0.0027355289,0.00031411948],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005261205,0.00015129604,0.03210098,0.00036492423,0.0004152486,0.000461408,0.00030904863,0.5366053,0.0016753296,0.32040027,0.0021631534,0.10482692],"study_design_scores_gemma":[0.0000729986,0.00013853071,0.0017805517,0.00003682992,0.00003581053,0.00008696372,0.00003921224,0.88802,0.00079483277,0.10825541,0.0007087721,0.000030042967],"about_ca_topic_score_codex":0.0035258452,"about_ca_topic_score_gemma":0.0020935384,"teacher_disagreement_score":0.021894231,"about_ca_system_score_codex":0.0015002898,"about_ca_system_score_gemma":0.0020763772,"threshold_uncertainty_score":0.115789175},"labels":[],"label_agreement":null},{"id":"W2990278103","doi":"10.1002/sim.8403","title":"Joint modeling of binary response and survival for clustered data in clinical trials","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Population Health Research Institute; Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada; Compute Canada","keywords":"Jackknife resampling; Covariate; Inference; Resampling; Sample size determination; Outcome (game theory); Statistics; Statistical inference; Computer science; Multivariate statistics; Statistical model; Survival analysis; Random effects model; Econometrics; Mathematics; Artificial intelligence; Medicine; Estimator; Meta-analysis; Internal medicine","score_opus":0.5370167228892887,"score_gpt":0.5753942456398025,"score_spread":0.03837752275051376,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2990278103","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0070418348,0.00075611396,0.99004245,0.0008508744,0.00008382354,0.00044142784,0.00022679286,0.00015611465,0.00040046612],"genre_scores_gemma":[0.32113895,0.0016605856,0.66610765,0.0013562668,0.00037613817,0.005645552,0.0010912312,0.000165734,0.0024579319],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8785556,0.10611914,0.0030436555,0.0071240775,0.004242615,0.00091494335],"domain_scores_gemma":[0.7702827,0.20042993,0.015698463,0.009323486,0.0033345679,0.00093087245],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.12879993,0.0014598832,0.004132553,0.0029526805,0.0009331066,0.0027845467,0.005501695,0.0041925376,0.0031214694],"category_scores_gemma":[0.22767122,0.0014012834,0.0039164377,0.003950762,0.0036698251,0.0033960296,0.002812895,0.005330079,0.00074095774],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00120468,0.00030949674,0.019811621,0.002132837,0.0023025745,0.00080518593,0.0013536569,0.44241253,0.0011497142,0.42084762,0.004524798,0.10314525],"study_design_scores_gemma":[0.00034787762,0.00035893824,0.0032202827,0.0002776437,0.0004486771,0.00015303104,0.000082292616,0.72936165,0.0005151824,0.26231638,0.00284073,0.00007729482],"about_ca_topic_score_codex":0.0042694644,"about_ca_topic_score_gemma":0.0043638716,"teacher_disagreement_score":0.12879993,"about_ca_system_score_codex":0.0028111327,"about_ca_system_score_gemma":0.0038195946,"threshold_uncertainty_score":0.68116724},"labels":[],"label_agreement":null},{"id":"W2991920218","doi":"10.1002/sim.8435","title":"Assessing the prior event rate ratio method via probabilistic bias analysis on a Bayesian network","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ontario Drug Policy Research Network; University of Toronto; University of Manitoba; Manitoba Health; Sanofi (Canada); University of Guelph","funders":"","keywords":"Confounding; Observational study; Bayesian probability; Econometrics; Robustness (evolution); Statistics; Computer science; Population; Probabilistic logic; Medicine; Mathematics; Environmental health","score_opus":0.07427757909138852,"score_gpt":0.44618420140823795,"score_spread":0.37190662231684946,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2991920218","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0044648913,0.00029629204,0.99360484,0.00038432254,0.000025519848,0.00028949275,0.00015202563,0.00014429355,0.0006385051],"genre_scores_gemma":[0.20973104,0.0009865108,0.7849427,0.00042928176,0.00020164308,0.0019702907,0.0005305418,0.00012596606,0.0010820463],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9367964,0.053003613,0.0017845707,0.0034633412,0.0044061746,0.0005459521],"domain_scores_gemma":[0.67337465,0.30207017,0.010777045,0.006733049,0.0061416733,0.000903416],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09197383,0.0019421912,0.0028743423,0.0048409007,0.0010389071,0.003034869,0.0042136186,0.002842353,0.0048770905],"category_scores_gemma":[0.26085702,0.0014856268,0.003681451,0.002683339,0.0027017747,0.0038619363,0.0037932934,0.0046466757,0.0005875676],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005700093,0.00015444764,0.020234268,0.0009981947,0.001688926,0.0003735666,0.0005084165,0.63636434,0.00090543507,0.1865128,0.002989314,0.14870039],"study_design_scores_gemma":[0.00011646959,0.00012017626,0.0012848763,0.0001911026,0.00026476034,0.00014867041,0.000038659913,0.8669629,0.0004313259,0.12828301,0.0021132997,0.000044842607],"about_ca_topic_score_codex":0.0095567,"about_ca_topic_score_gemma":0.0046149706,"teacher_disagreement_score":0.09197383,"about_ca_system_score_codex":0.0026412068,"about_ca_system_score_gemma":0.0042448873,"threshold_uncertainty_score":0.4864099},"labels":[],"label_agreement":null},{"id":"W2992396062","doi":"10.1002/sim.8413","title":"Association measures for clustered competing risks","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; McGill University","funders":"Centre de Recherches Mathématiques","keywords":"Estimator; Multivariate statistics; Statistics; Proportional hazards model; Copula (linguistics); Expectation–maximization algorithm; Econometrics; Hazard ratio; Marginal model; Bone marrow transplantation; Sample size determination; Odds ratio; Maximization; Mathematics; Computer science; Confidence interval; Medicine; Regression analysis; Maximum likelihood; Internal medicine; Transplantation; Mathematical optimization","score_opus":0.16984392854236996,"score_gpt":0.45805928438306515,"score_spread":0.2882153558406952,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2992396062","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015592882,0.0009354016,0.9813323,0.000319667,0.000075475626,0.00015390627,0.00079366745,0.000197557,0.00059914304],"genre_scores_gemma":[0.40993553,0.0012394035,0.5798292,0.00048570862,0.0005386453,0.002272136,0.003961266,0.00017810431,0.001560019],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9807643,0.011408218,0.0014506544,0.003577655,0.0023945959,0.00040450043],"domain_scores_gemma":[0.88916314,0.07901964,0.014787513,0.011022277,0.0048105326,0.0011968944],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.033371236,0.0013884373,0.0022394645,0.006638779,0.00087184523,0.0029990617,0.005329103,0.0021542483,0.004143856],"category_scores_gemma":[0.096393555,0.000595629,0.0026159922,0.006625381,0.0022329702,0.004228129,0.0036333534,0.0037219631,0.0007248337],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00034645476,0.00028390274,0.0461958,0.000986451,0.0015910928,0.00052497705,0.00073463586,0.21014388,0.0011289496,0.54413486,0.006488462,0.18744053],"study_design_scores_gemma":[0.00006276664,0.0003395701,0.009479934,0.00019858658,0.00025162648,0.000518785,0.0001982617,0.5411159,0.0004153069,0.44075367,0.0065618795,0.00010372231],"about_ca_topic_score_codex":0.00082498114,"about_ca_topic_score_gemma":0.00070903735,"teacher_disagreement_score":0.033371236,"about_ca_system_score_codex":0.001468084,"about_ca_system_score_gemma":0.0018092558,"threshold_uncertainty_score":0.17648607},"labels":[],"label_agreement":null},{"id":"W2992636254","doi":"10.1002/sim.8414","title":"Optimizing interim analysis timing for Bayesian adaptive commensurate designs","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Xenon Pharmaceuticals (Canada)","funders":"Sanofi","keywords":"Interim; Bayesian probability; Interim analysis; Computer science; Adaptive design; Econometrics; Statistics; Artificial intelligence; Mathematics; Medicine; Clinical trial; Internal medicine","score_opus":0.2877791635705335,"score_gpt":0.511439466930051,"score_spread":0.22366030335951753,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2992636254","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0071202773,0.0002906604,0.99036634,0.00025853372,0.000052673746,0.0006994635,0.00006239077,0.00022353222,0.00092609477],"genre_scores_gemma":[0.16520758,0.00033122397,0.82887787,0.00037973025,0.00007493908,0.0037479673,0.00017472397,0.00014034352,0.0010655884],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.95928943,0.03473224,0.0013303959,0.0019489459,0.0022579469,0.0004410124],"domain_scores_gemma":[0.9175348,0.06864009,0.004986262,0.005132742,0.0027022874,0.0010037496],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.059426658,0.0016631879,0.0022573296,0.001437695,0.0006296439,0.0018904543,0.002467409,0.0021514834,0.005359083],"category_scores_gemma":[0.14370102,0.0013362933,0.0015820151,0.0011318258,0.0017488935,0.0027207343,0.0025758336,0.0037020044,0.0011581635],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004431246,0.00071391976,0.0071628243,0.0009167575,0.00061061967,0.00023574421,0.00055719184,0.3401378,0.0055885655,0.2721179,0.0029203217,0.36460713],"study_design_scores_gemma":[0.0018916321,0.0034603556,0.0025111379,0.00037471286,0.0003366344,0.00023799781,0.0000749733,0.6797807,0.006694012,0.29451457,0.009947698,0.0001754854],"about_ca_topic_score_codex":0.00049715687,"about_ca_topic_score_gemma":0.00080450287,"teacher_disagreement_score":0.94057333,"about_ca_system_score_codex":0.0015538201,"about_ca_system_score_gemma":0.003654566,"threshold_uncertainty_score":0.31428194},"labels":[],"label_agreement":null},{"id":"W2993745637","doi":"10.1002/sim.8419","title":"Causal inference with noisy data: Bias analysis and estimation approaches to simultaneously addressing missingness and misclassification in binary outcomes","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs","keywords":"Missing data; Estimator; Inference; Causal inference; Computer science; Statistics; Estimation; Statistical inference; Econometrics; Binary number; Artificial intelligence; Mathematics; Machine learning","score_opus":0.43835106484182346,"score_gpt":0.46189622136305486,"score_spread":0.023545156521231403,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2993745637","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0012739946,0.00095451693,0.99681085,0.00050433836,0.00006314434,0.000031249205,0.00003670854,0.000036585418,0.00028853962],"genre_scores_gemma":[0.18567896,0.00651273,0.8024429,0.0008970616,0.0012584535,0.0007274041,0.0002704956,0.00013024657,0.0020817367],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9662651,0.026191326,0.0011306141,0.0025430862,0.0033962366,0.00047372733],"domain_scores_gemma":[0.8807174,0.1002733,0.008213503,0.007154925,0.003160829,0.00047991908],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.051228125,0.0019946585,0.003254667,0.0053228233,0.001191044,0.0027187879,0.0046728896,0.003744787,0.0028795882],"category_scores_gemma":[0.16266638,0.0010198039,0.0034096243,0.0064836326,0.0042413655,0.004346053,0.00378166,0.004418565,0.00042008056],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000081554,0.00010486281,0.0060183206,0.0009875956,0.0009017912,0.0002796083,0.0006212875,0.11541111,0.0008675067,0.69984734,0.0017053492,0.17317377],"study_design_scores_gemma":[0.000046234385,0.00008125182,0.0012032894,0.00026827745,0.00024768463,0.00018811645,0.00009295768,0.26411974,0.00093982695,0.7284229,0.004318308,0.00007145429],"about_ca_topic_score_codex":0.002652686,"about_ca_topic_score_gemma":0.0017849351,"teacher_disagreement_score":0.9487719,"about_ca_system_score_codex":0.0020805162,"about_ca_system_score_gemma":0.003482274,"threshold_uncertainty_score":0.27092344},"labels":[],"label_agreement":null},{"id":"W2994663156","doi":"10.1002/sim.8416","title":"A flexible copula‐based approach for the analysis of secondary phenotypes in ascertained samples","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic Associations and Epidemiology","field":"Biochemistry, Genetics and Molecular Biology","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Jewish General Hospital; Université du Québec à Montréal; McGill University Health Centre; Lunenfeld-Tanenbaum Research Institute","funders":"Medical Research Council; Natural Sciences and Engineering Research Council of Canada; Wellcome Trust","keywords":"Copula (linguistics); Computer science; Statistics; Phenotype; Econometrics; Mathematics; Biology; Genetics","score_opus":0.024022248024884275,"score_gpt":0.3229461029495333,"score_spread":0.298923854924649,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2994663156","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0012938848,0.00012625681,0.9980083,0.00006171743,0.000016708891,0.00006924006,0.00009252487,0.00014828629,0.00018310056],"genre_scores_gemma":[0.08644049,0.00064536487,0.90666395,0.0003128825,0.00016200558,0.0008969845,0.0012631414,0.00042113487,0.003194069],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9926138,0.0043693394,0.0003258419,0.001495627,0.0009085384,0.0002868024],"domain_scores_gemma":[0.9843544,0.0100201,0.0012745404,0.002303678,0.0015791251,0.00046815717],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014385634,0.0021446398,0.0027599116,0.003962701,0.0011783566,0.002195627,0.005343038,0.0016097664,0.005367014],"category_scores_gemma":[0.040681545,0.001637552,0.0041430965,0.003933546,0.0018155203,0.0021763472,0.003471014,0.0036494043,0.0021080591],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00031907324,0.00028210273,0.02261674,0.000594128,0.0019218703,0.0022682317,0.0009909185,0.48203105,0.006010804,0.20275316,0.009996075,0.27021587],"study_design_scores_gemma":[0.00004240373,0.00006393199,0.0025237205,0.0000417074,0.00011328213,0.0003961166,0.00005170778,0.9512332,0.00062386034,0.041467715,0.00338644,0.000055871966],"about_ca_topic_score_codex":0.008483009,"about_ca_topic_score_gemma":0.008645825,"teacher_disagreement_score":0.014385634,"about_ca_system_score_codex":0.0011609149,"about_ca_system_score_gemma":0.002610026,"threshold_uncertainty_score":0.07607943},"labels":[],"label_agreement":null},{"id":"W2995768992","doi":"10.1002/sim.8410","title":"Safety surveillance and the estimation of risk in select populations: Flexible methods to control for confounding while targeting marginal comparisons via standardization","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Institutes of Health; U.S. Food and Drug Administration; Hamilton Health Sciences Foundation; National Center for Advancing Translational Sciences; U.S. Department of Health and Human Services","keywords":"Covariate; Causal inference; Confounding; Computer science; Inference; Statistics; Context (archaeology); Econometrics; Machine learning; Mathematics; Artificial intelligence","score_opus":0.07507788337893141,"score_gpt":0.472539222786533,"score_spread":0.3974613394076016,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2995768992","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0068595516,0.00015846627,0.992161,0.00028481253,0.000022486016,0.0000855553,0.00004355454,0.00009117751,0.00029342808],"genre_scores_gemma":[0.35497716,0.0005560233,0.6416765,0.00046486603,0.00021333902,0.0008715341,0.00026494547,0.00010558187,0.00086998206],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9613798,0.03141581,0.0010745815,0.0032528283,0.0023873851,0.00048969284],"domain_scores_gemma":[0.922535,0.057929765,0.0056978236,0.011473907,0.0018926795,0.0004708516],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04870896,0.001189314,0.0019532803,0.0023527169,0.000864553,0.0019308957,0.0025988235,0.0014248393,0.0017230716],"category_scores_gemma":[0.1294689,0.0007232426,0.0028629696,0.0023589593,0.003955954,0.0026322736,0.0043850695,0.0033111037,0.00022411681],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003163221,0.00019816821,0.038751323,0.00037243427,0.001451501,0.0005526995,0.0014928731,0.20910494,0.0018341405,0.37082398,0.002070936,0.37303063],"study_design_scores_gemma":[0.00015241264,0.00025850837,0.008508856,0.00012336198,0.00026669248,0.00021173745,0.00016146571,0.5191088,0.001802864,0.46491665,0.0044030137,0.00008575126],"about_ca_topic_score_codex":0.0037753074,"about_ca_topic_score_gemma":0.002356123,"teacher_disagreement_score":0.04870896,"about_ca_system_score_codex":0.0009879689,"about_ca_system_score_gemma":0.0028839386,"threshold_uncertainty_score":0.25760067},"labels":[],"label_agreement":null},{"id":"W2995814717","doi":"10.1002/sim.9356","title":"Sample size estimation using a latent variable model for mixed outcome co‐primary, multiple primary and composite endpoints","year":2022,"lang":"en","type":"preprint","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"NIHR Cambridge Biomedical Research Centre; Medical Research Council; National Institute for Health and Care Research; Medical Research Council Canada; Cancer Research UK","keywords":"Sample size determination; Clinical endpoint; Outcome (game theory); Statistics; Sample (material); Latent variable; Econometrics; Computer science; Mathematics; Clinical trial; Medicine; Physics; Internal medicine","score_opus":0.1837940103163691,"score_gpt":0.43787408768378433,"score_spread":0.2540800773674152,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2995814717","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0016808911,0.00009873234,0.99745494,0.00021536415,0.000026021899,0.00013676786,0.000049048682,0.000082839106,0.00025545238],"genre_scores_gemma":[0.094169624,0.00038608006,0.90057814,0.00030004897,0.00012427608,0.0029169435,0.0003123715,0.00012368798,0.0010887162],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95928985,0.034565765,0.00092664314,0.002403018,0.0023555197,0.00045920745],"domain_scores_gemma":[0.8122177,0.17237069,0.0045875506,0.0075864736,0.002756372,0.00048128175],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06461814,0.0014626252,0.0026773354,0.0024255419,0.0008163401,0.0025834732,0.0034899786,0.0031269572,0.004267059],"category_scores_gemma":[0.20549874,0.0010209695,0.0024147744,0.0022498842,0.0038838214,0.0040406156,0.003093844,0.0049729575,0.0007844547],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007934478,0.00029306128,0.006890784,0.0010571805,0.0006784635,0.000375757,0.0010373988,0.18672763,0.0029374138,0.62555146,0.0055815834,0.16807583],"study_design_scores_gemma":[0.0002563967,0.00027220868,0.0011824336,0.00015614122,0.00010435635,0.00012880983,0.000099798766,0.54942936,0.0013589702,0.44418997,0.002770079,0.000051408362],"about_ca_topic_score_codex":0.0017294119,"about_ca_topic_score_gemma":0.0015144222,"teacher_disagreement_score":0.9353819,"about_ca_system_score_codex":0.0022326687,"about_ca_system_score_gemma":0.0025959292,"threshold_uncertainty_score":0.34173745},"labels":[],"label_agreement":null},{"id":"W2996227145","doi":"10.1002/sim.8411","title":"Multistate analysis from cross‐sectional and auxiliary samples","year":2019,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Cross-sectional study; Statistics; Computer science; Mathematics","score_opus":0.10578371827638448,"score_gpt":0.46670829117387863,"score_spread":0.36092457289749413,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2996227145","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.068641745,0.00035742507,0.9287801,0.00028916748,0.00007262327,0.00017583878,0.00046468768,0.00021612477,0.0010023016],"genre_scores_gemma":[0.67825377,0.0007907983,0.3123833,0.00031245523,0.00018280408,0.0010982901,0.0031055189,0.00015461755,0.0037184858],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99059945,0.0071472437,0.00032899776,0.001131717,0.0005992528,0.00019336372],"domain_scores_gemma":[0.88800204,0.08486386,0.0048451414,0.019079773,0.0025583687,0.00065080565],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.034901306,0.0007729745,0.0014305925,0.0014170604,0.0007451216,0.0021161505,0.0025031797,0.0013397958,0.003991368],"category_scores_gemma":[0.104080215,0.00073922594,0.0020859356,0.0015548615,0.0020265356,0.002735836,0.0031829071,0.0031574038,0.00056803913],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000975307,0.00051227055,0.10530693,0.0007134117,0.0013831033,0.0008244003,0.0018509954,0.21012312,0.0042867577,0.52844816,0.0043999185,0.14117561],"study_design_scores_gemma":[0.00006055096,0.00029266058,0.018030284,0.00011552766,0.00022810021,0.00024576008,0.00019999921,0.7479986,0.0014574932,0.22641975,0.004892078,0.000059247544],"about_ca_topic_score_codex":0.0020423622,"about_ca_topic_score_gemma":0.0023722686,"teacher_disagreement_score":0.034901306,"about_ca_system_score_codex":0.0006600265,"about_ca_system_score_gemma":0.0011484371,"threshold_uncertainty_score":0.18457794},"labels":[],"label_agreement":null},{"id":"W2998463403","doi":"10.1002/sim.8481","title":"A threshold‐free summary index for quantifying the capacity of covariates to yield efficient treatment rules","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Canadian Institutes of Health Research","keywords":"Covariate; Context (archaeology); Econometrics; Estimator; Index (typography); Computer science; Metric (unit); Statistics; Parametric statistics; Mathematics; Economics","score_opus":0.28486632970212217,"score_gpt":0.4353815638659449,"score_spread":0.15051523416382273,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2998463403","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03196518,0.0024683094,0.9530537,0.0010070466,0.00018932867,0.00049250136,0.0052849427,0.0010170577,0.0045219855],"genre_scores_gemma":[0.46931192,0.0014840679,0.5160487,0.00083976233,0.00046744052,0.0018259805,0.0077088973,0.0005080241,0.0018052193],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97840524,0.011998924,0.0029264598,0.0021618516,0.0040827575,0.00042464607],"domain_scores_gemma":[0.82739985,0.13692027,0.013482535,0.0136793805,0.0071430537,0.0013749443],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.035937913,0.0011747758,0.002782916,0.0059949,0.00046916254,0.0033217822,0.001814619,0.0021439102,0.0055614663],"category_scores_gemma":[0.18011642,0.0005215791,0.002178188,0.0046069496,0.0013127194,0.0039167637,0.0018258481,0.002669204,0.0011013687],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016682375,0.00030868474,0.056661207,0.0029560812,0.003312315,0.0002737624,0.00050679676,0.27131698,0.0049361354,0.18972886,0.023725027,0.44460595],"study_design_scores_gemma":[0.00033636345,0.0020857635,0.034908753,0.0007646946,0.0012073677,0.00080777856,0.00015735642,0.64857626,0.0071077636,0.28013656,0.023560533,0.00035084705],"about_ca_topic_score_codex":0.00076624384,"about_ca_topic_score_gemma":0.00071593927,"teacher_disagreement_score":0.035937913,"about_ca_system_score_codex":0.0016031945,"about_ca_system_score_gemma":0.0017933719,"threshold_uncertainty_score":0.19006014},"labels":[],"label_agreement":null},{"id":"W3003667353","doi":"10.1002/sim.8468","title":"A fair comparison of tree‐based and parametric methods in multiple imputation by chained equations","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":102,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Norman Cousins Center for Psychoneuroimmunology; National Institute on Aging; National Institute of Biomedical Imaging and Bioengineering; Canadian Institutes of Health Research; Genentech; National Institutes of Health; H. Lundbeck A/S; Servier; National Institute of Environmental Health Sciences; Northern California Institute for Research and Education; Alzheimer's Disease Neuroimaging Initiative; GE Healthcare; Alzheimer's Association; Fujirebio US; Pfizer; BioClinica; Biogen; Novartis Pharmaceuticals Corporation; U.S. Department of Defense; Eli Lilly and Company; Bristol-Myers Squibb; Roche; Merck; Alzheimer's Drug Discovery Foundation; Takeda Pharmaceutical Company; AbbVie; Foundation for the National Institutes of Health","keywords":"Imputation (statistics); Parametric statistics; Computer science; Nonparametric statistics; Inference; Missing data; Statistics; Econometrics; Data mining; Mathematics; Machine learning; Artificial intelligence","score_opus":0.172531786709154,"score_gpt":0.4936504161418888,"score_spread":0.32111862943273484,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3003667353","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008792108,0.002814976,0.98497945,0.0008993193,0.00022679327,0.00018615162,0.0003181238,0.00026501546,0.0015181045],"genre_scores_gemma":[0.22451434,0.0029856602,0.7660345,0.00096447393,0.0003830031,0.0013067544,0.0014699383,0.0004446188,0.001896738],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8841032,0.09839463,0.003044368,0.0042037107,0.009333769,0.0009202937],"domain_scores_gemma":[0.62584394,0.3244499,0.0076512275,0.029048214,0.011991602,0.0010150406],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.12852415,0.0011722852,0.0025894335,0.0024966756,0.0012512052,0.0032859144,0.0034578685,0.002768196,0.004561874],"category_scores_gemma":[0.35220265,0.0008893479,0.0025766238,0.0030283302,0.0032992517,0.006616035,0.004790709,0.0037752108,0.0009589983],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002369834,0.00023748216,0.014880997,0.0015077221,0.0022687595,0.00019898999,0.001056365,0.2382236,0.000966165,0.38761473,0.012436934,0.33823845],"study_design_scores_gemma":[0.00048497543,0.0006678093,0.0042704935,0.0009288207,0.0004645045,0.00030909668,0.0002252754,0.61648786,0.0020012842,0.35678336,0.017201062,0.00017545794],"about_ca_topic_score_codex":0.0035227959,"about_ca_topic_score_gemma":0.002768308,"teacher_disagreement_score":0.8714758,"about_ca_system_score_codex":0.0019483004,"about_ca_system_score_gemma":0.0038672136,"threshold_uncertainty_score":0.67970884},"labels":[],"label_agreement":null},{"id":"W3003781265","doi":"10.1002/sim.8487","title":"Correction to Walter SD, Turner RM, Macaskill P. Optimising the two‐stage randomised trial design when some participants are indifferent in their treatment preferences (2019). Statistics in Medicine 38, 2317‐2331","year":2020,"lang":"en","type":"erratum","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Hamilton Health Sciences; Impact","funders":"","keywords":"Citation; Order (exchange); Medicine; Library science; Psychology; Computer science","score_opus":0.5087324334082165,"score_gpt":0.5251660379332699,"score_spread":0.01643360452505338,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3003781265","genre_codex":"editorial","genre_gemma":"other","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000029227533,0.0018478616,0.0011787856,0.051329542,0.9383645,0.000104184925,0.0044709966,0.00056260824,0.0021122503],"genre_scores_gemma":[0.005011037,0.013920201,0.017430395,0.19621359,0.48865828,0.003114186,0.010085147,0.0051351786,0.26043198],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9690745,0.008981785,0.0071035265,0.0028280853,0.010457237,0.0015549064],"domain_scores_gemma":[0.76365346,0.09013035,0.011151164,0.0114757065,0.11802309,0.005566261],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.033289406,0.0051742587,0.005720087,0.012385101,0.005246324,0.008758836,0.008057082,0.012415813,0.12294217],"category_scores_gemma":[0.3191622,0.0037184106,0.0048541566,0.00955887,0.0052771396,0.004516807,0.0037450115,0.025023367,0.09829635],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000024029761,0.0000034386742,0.000016547021,0.00021092666,0.0000090988715,0.00003720805,0.000015193007,0.000023938566,0.00000945525,0.00044729677,0.9963092,0.0028935438],"study_design_scores_gemma":[0.00015379564,0.000018801602,0.00045604663,0.0013685954,0.00006982485,0.00021021011,0.000047656617,0.0002082175,0.00009852867,0.0024845183,0.99482214,0.00006158673],"about_ca_topic_score_codex":0.068922125,"about_ca_topic_score_gemma":0.06734822,"teacher_disagreement_score":0.96671057,"about_ca_system_score_codex":0.014335729,"about_ca_system_score_gemma":0.020275783,"threshold_uncertainty_score":0.41128254},"labels":[],"label_agreement":null},{"id":"W3004002562","doi":"10.1002/sim.8465","title":"Bayesian hierarchical meta‐analytic methods for modeling surrogate relationships that vary across treatment classes using aggregate data","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council Canada; Medical Research Council; National Institute for Health and Care Research","keywords":"Surrogate endpoint; Bayesian probability; Computer science; Surrogate model; Surrogate data; Bayesian inference; Meta-analysis; Class (philosophy); Econometrics; Statistics; Data mining; Artificial intelligence; Machine learning; Mathematics; Medicine; Internal medicine","score_opus":0.8988886626769544,"score_gpt":0.6656536011715628,"score_spread":0.23323506150539164,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3004002562","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0017167646,0.0036823077,0.99281913,0.00042751097,0.00007367528,0.00044745844,0.00033630093,0.00018785868,0.00030889004],"genre_scores_gemma":[0.108259395,0.0051384727,0.87911725,0.00071573583,0.00027221302,0.004690874,0.0009084278,0.00016803808,0.0007296085],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9025554,0.087770276,0.0025119025,0.002732517,0.004118512,0.0003114225],"domain_scores_gemma":[0.84111404,0.13916215,0.0073645827,0.009811128,0.002187443,0.00036059108],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.12602347,0.0027772554,0.0045514167,0.007158913,0.000868779,0.003394089,0.0050002392,0.0028682507,0.0030636236],"category_scores_gemma":[0.18144353,0.0020226499,0.010003993,0.005915876,0.0016692589,0.0034092858,0.0027886438,0.005663833,0.0005313222],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001347638,0.00025940986,0.009461269,0.006884365,0.037876714,0.00035485983,0.0007176499,0.46098918,0.001339922,0.23178747,0.0060377913,0.24294372],"study_design_scores_gemma":[0.00067769043,0.0005079633,0.0021585291,0.0011815605,0.008094993,0.00017040085,0.00008486644,0.6110743,0.0010122111,0.36471796,0.010159173,0.00016037382],"about_ca_topic_score_codex":0.004866332,"about_ca_topic_score_gemma":0.006041428,"teacher_disagreement_score":0.8739765,"about_ca_system_score_codex":0.0022601376,"about_ca_system_score_gemma":0.0034145797,"threshold_uncertainty_score":0.66648376},"labels":[],"label_agreement":null},{"id":"W3004845713","doi":"10.1002/sim.8479","title":"Ordinal outcomes: A cumulative probability model with the log link and an assumption of proportionality","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Ordinal regression; Logit; Mathematics; Link (geometry); Ordered logit; Odds; Statistics; Logistic regression; Ordinal data; Log-linear model; Econometrics; Uniqueness; Linear model; Combinatorics","score_opus":0.1775003639773146,"score_gpt":0.44978809162532396,"score_spread":0.27228772764800935,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3004845713","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.019171402,0.0022350915,0.9546183,0.00644634,0.00029471883,0.00018169971,0.0010261761,0.0002476757,0.015778651],"genre_scores_gemma":[0.6672083,0.005579244,0.27621368,0.0034994131,0.0024194024,0.002018898,0.0017507282,0.00041808345,0.04089222],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9895476,0.0063200225,0.00030931013,0.0016382366,0.0014634754,0.0007213402],"domain_scores_gemma":[0.94258314,0.04620459,0.004368607,0.0039027887,0.0019366515,0.001004246],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.019354831,0.0013217485,0.0021835952,0.0029202271,0.0013184563,0.0053155925,0.004374247,0.0031955529,0.017208405],"category_scores_gemma":[0.07413495,0.0008362046,0.0024488461,0.0042685284,0.004797079,0.010475737,0.0035902779,0.0065141344,0.0026785554],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000073698786,0.00007815564,0.0038591311,0.00018484211,0.00009862241,0.00027887762,0.00044524542,0.022370301,0.00014829989,0.94475317,0.002930135,0.02477952],"study_design_scores_gemma":[0.000047641333,0.00010944926,0.0015470558,0.00013471776,0.00008527828,0.00040826912,0.00012063567,0.108262785,0.000109755114,0.87923926,0.009885971,0.00004924151],"about_ca_topic_score_codex":0.005717457,"about_ca_topic_score_gemma":0.0027025884,"teacher_disagreement_score":0.019354831,"about_ca_system_score_codex":0.0023446109,"about_ca_system_score_gemma":0.0021466836,"threshold_uncertainty_score":0.102359354},"labels":[],"label_agreement":null},{"id":"W3005876327","doi":"10.1002/sim.8484","title":"Comparing a multivariate response Bayesian random effects logistic regression model with a latent variable item response theory model for provider profiling on multiple binary indicators simultaneously","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sunnybrook Hospital; Institute of Health Services and Policy Research; University Health Network; University of Toronto; Institute for Work & Health; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Economic and Social Research Council; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Logistic regression; Statistics; Multivariate statistics; Bayesian probability; Medicine; Multivariate analysis; Econometrics; Mathematics","score_opus":0.08760402899998009,"score_gpt":0.3791329511887469,"score_spread":0.2915289221887668,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3005876327","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1379022,0.00044466977,0.85483015,0.0027146847,0.0001781374,0.0008107678,0.00078100944,0.0005420093,0.0017965026],"genre_scores_gemma":[0.641859,0.0005318937,0.34614852,0.0011471298,0.0002976338,0.0032069448,0.0022354748,0.00024792977,0.0043254737],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9026798,0.08584392,0.0014117322,0.0055808425,0.0027796535,0.0017041184],"domain_scores_gemma":[0.8079376,0.1723882,0.00846473,0.0047305226,0.0054557994,0.0010231133],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.085547715,0.002433722,0.0034589285,0.003931355,0.0010910601,0.0036832548,0.006556328,0.003094847,0.005568454],"category_scores_gemma":[0.15881415,0.0017823091,0.004829217,0.004149464,0.0023755806,0.0044311327,0.003695396,0.0057490445,0.0014328511],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024974768,0.0011071193,0.059275754,0.00046232482,0.002953263,0.00049030344,0.0018906994,0.7694158,0.0005026353,0.07747348,0.003002611,0.08092848],"study_design_scores_gemma":[0.00029755034,0.0003352953,0.0048686382,0.00007492369,0.00015674463,0.000046426732,0.00018491255,0.970311,0.00010353274,0.02272364,0.00080774643,0.000089683635],"about_ca_topic_score_codex":0.031272087,"about_ca_topic_score_gemma":0.0135876965,"teacher_disagreement_score":0.085547715,"about_ca_system_score_codex":0.004153105,"about_ca_system_score_gemma":0.0038538347,"threshold_uncertainty_score":0.45242494},"labels":[],"label_agreement":null},{"id":"W3006396558","doi":"10.1002/sim.8495","title":"A hierarchical testing approach for detecting safety signals in clinical trials","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"National Cancer Institute; National Institute of General Medical Sciences; Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Clinical trial; Data mining; Safety monitoring; Curse of dimensionality; Class (philosophy); Artificial intelligence; Medicine; Bioinformatics","score_opus":0.8299342855077583,"score_gpt":0.649968645019091,"score_spread":0.17996564048866726,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3006396558","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0056141065,0.00024117628,0.99225336,0.00043554383,0.000045794302,0.00029530126,0.00020075821,0.0002762962,0.00063774484],"genre_scores_gemma":[0.34270075,0.0003274491,0.65156174,0.00078677514,0.00034901482,0.0021943573,0.00068127963,0.00012363323,0.0012750207],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9145066,0.069721,0.0023284156,0.0055856453,0.00675907,0.0010992769],"domain_scores_gemma":[0.71021247,0.25585744,0.010495117,0.015231747,0.006365923,0.0018372479],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07009605,0.0018126519,0.0027139839,0.004316766,0.0013213217,0.0018552912,0.0040364787,0.002086674,0.0045084613],"category_scores_gemma":[0.17231254,0.0009033388,0.003364968,0.0036888812,0.0045057745,0.0025699048,0.0029202201,0.0038115578,0.00072244724],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013079548,0.00041566757,0.030220738,0.0013110288,0.0028776955,0.0010444112,0.0012986626,0.2517477,0.004456624,0.32297006,0.007178493,0.37517092],"study_design_scores_gemma":[0.00030918597,0.0008968038,0.0050448077,0.00009852582,0.00035376544,0.00021737877,0.0000705336,0.7115688,0.0012795314,0.27777985,0.0023161573,0.00006471927],"about_ca_topic_score_codex":0.0052510845,"about_ca_topic_score_gemma":0.0046957037,"teacher_disagreement_score":0.9299039,"about_ca_system_score_codex":0.0023918482,"about_ca_system_score_gemma":0.00411437,"threshold_uncertainty_score":0.3707078},"labels":[],"label_agreement":null},{"id":"W3008905790","doi":"10.1002/sim.8502","title":"Variance estimation when using propensity‐score matching with replacement with survival or time‐to‐event outcomes","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":86,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Sunnybrook Hospital; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Statistics; Event data; Matching (statistics); Variance (accounting); Estimation; Event (particle physics); Survival analysis; Econometrics; Computer science; Mathematics; Covariate","score_opus":0.1810926378835545,"score_gpt":0.4193115110956037,"score_spread":0.2382188732120492,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3008905790","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0018678621,0.00018374616,0.99687517,0.00016609774,0.000052914605,0.00013856562,0.00008040792,0.000139564,0.0004956486],"genre_scores_gemma":[0.13688096,0.0007917577,0.85758156,0.00046051995,0.0002553126,0.0015496506,0.0007228439,0.00018927407,0.0015680646],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96054405,0.031701457,0.0013979598,0.0025849927,0.0032568465,0.00051465846],"domain_scores_gemma":[0.93285334,0.051921483,0.0043057324,0.008588676,0.0021088545,0.00022185635],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04785462,0.0011279677,0.0018629399,0.0030295944,0.00072280125,0.0018869791,0.00263779,0.0019474298,0.003658494],"category_scores_gemma":[0.18165366,0.000826866,0.0026055821,0.0040811272,0.001770384,0.0028191227,0.0027345133,0.0025834758,0.0007202231],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00028613047,0.00023555305,0.024559198,0.000700607,0.0016658185,0.00036993765,0.0006562025,0.14850783,0.0009291477,0.51428354,0.008739842,0.29906622],"study_design_scores_gemma":[0.00017736718,0.00021749457,0.005203915,0.00023485259,0.00025203524,0.00029916517,0.00013999222,0.43006966,0.0016347741,0.5460188,0.015660204,0.000091891736],"about_ca_topic_score_codex":0.003867557,"about_ca_topic_score_gemma":0.0021978375,"teacher_disagreement_score":0.04785462,"about_ca_system_score_codex":0.0011004442,"about_ca_system_score_gemma":0.0024568674,"threshold_uncertainty_score":0},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W3008923208","doi":"10.1002/sim.8507","title":"A bootstrap semiparametric homogeneity test for the distributions of multigroup proportional data, with applications to analysis of quality of life outcomes in clinical trials","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University; Canadian Cancer Society","funders":"Fundamental Research Funds for the Central Universities; Natural Sciences and Engineering Research Council of Canada; National Natural Science Foundation of China","keywords":"Statistics; Mathematics; Homogeneity (statistics); Nonparametric statistics; Type I and type II errors; Statistical hypothesis testing; Parametric statistics; Likelihood-ratio test; Wald test; Test statistic; Null hypothesis; Empirical likelihood; Econometrics; Confidence interval","score_opus":0.8162034951521221,"score_gpt":0.6841353634128662,"score_spread":0.1320681317392559,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3008923208","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015708152,0.0007382371,0.98072386,0.00041077562,0.00009837559,0.00054516067,0.00023006875,0.00036914935,0.0011761885],"genre_scores_gemma":[0.52683485,0.00090213097,0.46370253,0.00083696743,0.00062727137,0.0043802396,0.0013189133,0.00027917651,0.0011179912],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9366553,0.050562922,0.0018904656,0.0036564828,0.006661335,0.0005735624],"domain_scores_gemma":[0.7470211,0.22399315,0.009963153,0.011542051,0.0060982755,0.0013823527],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06070236,0.0010418822,0.0028630623,0.0047271755,0.0010714377,0.0020525272,0.0031894087,0.0022537133,0.004583705],"category_scores_gemma":[0.2585535,0.0006104832,0.0027721385,0.0032648721,0.0038310054,0.0033658303,0.004057674,0.0034159527,0.0009054336],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0023583416,0.0005827279,0.042955503,0.0018262282,0.002906722,0.0014333745,0.0014407559,0.069833614,0.006528799,0.25157356,0.008311038,0.6102492],"study_design_scores_gemma":[0.00081110134,0.003139216,0.03064742,0.0004978568,0.0007245104,0.001931882,0.00056595873,0.60452014,0.0049143923,0.34108353,0.010870554,0.0002934123],"about_ca_topic_score_codex":0.0006872686,"about_ca_topic_score_gemma":0.00036238687,"teacher_disagreement_score":0.06070236,"about_ca_system_score_codex":0.0010260317,"about_ca_system_score_gemma":0.0022447265,"threshold_uncertainty_score":0.32102865},"labels":[],"label_agreement":null},{"id":"W3009924189","doi":"10.1002/sim.8517","title":"Mitigating bias from intermittent measurement of time‐dependent covariates in failure time analysis","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Estimator; Covariate; Proportional hazards model; Statistics; Regression analysis; Computer science; Regression; Linear regression; Econometrics; Mathematics","score_opus":0.12201525913244021,"score_gpt":0.36799358052698866,"score_spread":0.24597832139454845,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3009924189","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.054036032,0.00089814386,0.9435769,0.00058108906,0.00009417984,0.00006348955,0.00009852959,0.00025052164,0.00040120044],"genre_scores_gemma":[0.7776868,0.00077532104,0.21887249,0.0004478068,0.00030560314,0.0003556384,0.00031027914,0.00018199605,0.0010641157],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9530435,0.0389347,0.0015559052,0.0028546317,0.0029443644,0.0006669378],"domain_scores_gemma":[0.5487257,0.3952781,0.021844534,0.028435426,0.0046519097,0.0010643037],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.10945189,0.00072085485,0.001838429,0.0016624832,0.0007609436,0.0020474566,0.0027149364,0.00208483,0.000924287],"category_scores_gemma":[0.35081294,0.0008767525,0.0013622437,0.0025469444,0.00322295,0.002552716,0.0034664385,0.0030788716,0.00023631824],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014358672,0.00026628844,0.21872579,0.00093358837,0.0017934891,0.0012272905,0.0043515004,0.29326966,0.0036238043,0.18814346,0.0035413972,0.28268787],"study_design_scores_gemma":[0.00017991228,0.000604251,0.032000672,0.0002770613,0.00039774328,0.00058952445,0.0003243272,0.75667256,0.0033598163,0.20148239,0.0039619305,0.00014977485],"about_ca_topic_score_codex":0.0039907577,"about_ca_topic_score_gemma":0.0033070021,"teacher_disagreement_score":0.10945189,"about_ca_system_score_codex":0.00107673,"about_ca_system_score_gemma":0.002022937,"threshold_uncertainty_score":0.57884383},"labels":[],"label_agreement":null},{"id":"W3010442290","doi":"10.1002/sim.8519","title":"Sample size and power calculations for open cohort longitudinal cluster randomized trials","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":44,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; National Health and Medical Research Council; Medical Research Council Canada","keywords":"Sample size determination; Statistics; Cohort; Sample (material); Sampling (signal processing); Cluster sampling; Cluster (spacecraft); Statistical power; Autocorrelation; Econometrics; Mathematics; Psychology; Demography; Computer science; Population","score_opus":0.6232407975323997,"score_gpt":0.6131850654261651,"score_spread":0.01005573210623456,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3010442290","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.003833467,0.0056617325,0.92968446,0.002953477,0.0020987205,0.04142924,0.0023187567,0.0009563344,0.011063795],"genre_scores_gemma":[0.042272415,0.0016741053,0.8138602,0.002176141,0.00051777845,0.13618155,0.00074351154,0.00053930347,0.002035054],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7709667,0.18201263,0.018985718,0.005825594,0.02076852,0.0014408029],"domain_scores_gemma":[0.5694803,0.38131765,0.015331439,0.01737181,0.015430862,0.0010679369],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.20158377,0.0025204674,0.0044768513,0.0057106433,0.0012489924,0.0033541466,0.0052571055,0.0053598583,0.02033441],"category_scores_gemma":[0.5822688,0.0015703988,0.005473439,0.0042007156,0.0030842125,0.004852386,0.0031806629,0.008086449,0.0043098703],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.007745462,0.00062730676,0.0051561957,0.019678794,0.003379938,0.00065296615,0.0031406106,0.023457201,0.001675835,0.31458154,0.07680328,0.5431009],"study_design_scores_gemma":[0.009428175,0.005529869,0.007243645,0.017736243,0.003877917,0.0016472243,0.00073146116,0.09797319,0.011681813,0.5857025,0.25800538,0.0004427208],"about_ca_topic_score_codex":0.00092286046,"about_ca_topic_score_gemma":0.00085957645,"teacher_disagreement_score":0.79841626,"about_ca_system_score_codex":0.0027329181,"about_ca_system_score_gemma":0.0038736002,"threshold_uncertainty_score":0.98458934},"labels":[],"label_agreement":null},{"id":"W3014896206","doi":"10.1002/sim.8532","title":"STRATOS guidance document on measurement error and misclassification of variables in observational epidemiology: Part 1—Basic theory and simple methods of adjustment","year":2020,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Nutritional Studies and Diet","field":"Medicine","cited_by":176,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; University of British Columbia","funders":"National Institutes of Health; National Cancer Institute; Medical Research Council; National Institute of Allergy and Infectious Diseases; Natural Sciences and Engineering Research Council of Canada; Patient-Centered Outcomes Research Institute","keywords":"Covariate; Observational error; Statistics; Computer science; Observational study; Errors-in-variables models; Econometrics; Type I and type II errors; Regression; Regression analysis; Sample size determination; Extrapolation; Calibration; Linear regression; Data mining; Mathematics","score_opus":0.3319225737265194,"score_gpt":0.49109950696514776,"score_spread":0.15917693323862836,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3014896206","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0032087213,0.05811043,0.3147441,0.19573422,0.03118744,0.014094706,0.14020368,0.008965378,0.23375137],"genre_scores_gemma":[0.013349351,0.080415614,0.56425303,0.07881773,0.0110114,0.021001404,0.086515926,0.0040613003,0.14057429],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.95975256,0.016853482,0.0045640767,0.0016622151,0.015976045,0.0011915487],"domain_scores_gemma":[0.8965233,0.061853874,0.005866391,0.006502762,0.02746163,0.0017919736],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.041072957,0.0037374722,0.0033955793,0.007812687,0.0017194272,0.005003806,0.008268571,0.015936673,0.032947253],"category_scores_gemma":[0.12413955,0.002741273,0.003650377,0.006735467,0.0031696907,0.004075933,0.005021681,0.011115181,0.029031474],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008232251,0.00011338545,0.0010497958,0.0018514823,0.00005609439,0.00014475717,0.00032100704,0.0030414995,0.0003396651,0.039273776,0.88259405,0.07113214],"study_design_scores_gemma":[0.00013869385,0.00013141167,0.0032512576,0.0045072204,0.00007636485,0.00028586137,0.000151024,0.0017455885,0.00049182586,0.032034893,0.9570813,0.00010439162],"about_ca_topic_score_codex":0.059592687,"about_ca_topic_score_gemma":0.049514093,"teacher_disagreement_score":0.95892704,"about_ca_system_score_codex":0.005187134,"about_ca_system_score_gemma":0.02717891,"threshold_uncertainty_score":0.21721709},"labels":[],"label_agreement":null},{"id":"W3014964148","doi":"10.1002/sim.8531","title":"STRATOS guidance document on measurement error and misclassification of variables in observational epidemiology: Part 2—More complex methods of adjustment and advanced topics","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":126,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; University of British Columbia","funders":"National Institutes of Health; National Cancer Institute; Medical Research Council; National Institute of Allergy and Infectious Diseases; Natural Sciences and Engineering Research Council of Canada; Patient-Centered Outcomes Research Institute","keywords":"Categorical variable; Covariate; Observational error; Computer science; Statistics; Imputation (statistics); Bayesian probability; Errors-in-variables models; Data mining; Bayes' theorem; Missing data; Standard error; Econometrics; Mathematics","score_opus":0.4407366589677345,"score_gpt":0.5053344638457099,"score_spread":0.0645978048779754,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3014964148","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0022542158,0.07428978,0.4628504,0.18517363,0.037224423,0.004842915,0.05860157,0.0072121383,0.16755089],"genre_scores_gemma":[0.010752948,0.0876628,0.652219,0.07025423,0.014564453,0.007183733,0.039025437,0.005332122,0.11300527],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9688326,0.014464184,0.0029490262,0.0014651293,0.011442187,0.0008469036],"domain_scores_gemma":[0.91655034,0.05060347,0.0040500863,0.0074723046,0.019882921,0.0014407736],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.034636084,0.0026271923,0.0028987993,0.006447534,0.0015467509,0.004830299,0.005774295,0.009141687,0.046246335],"category_scores_gemma":[0.09947564,0.0016554579,0.0034114106,0.00685479,0.0027789965,0.0039660656,0.004099357,0.009065861,0.028180419],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000046952588,0.00006307361,0.0006322987,0.0014178671,0.00004945773,0.00009703802,0.00027200326,0.0023466032,0.00034222042,0.066271715,0.8309164,0.097544424],"study_design_scores_gemma":[0.000042887834,0.00004872251,0.0012152109,0.002413619,0.000035134428,0.00018939067,0.000077295634,0.001231326,0.0003842606,0.04383164,0.95047176,0.000058812846],"about_ca_topic_score_codex":0.037,"about_ca_topic_score_gemma":0.03298335,"teacher_disagreement_score":0.9653639,"about_ca_system_score_codex":0.0042822342,"about_ca_system_score_gemma":0.016216498,"threshold_uncertainty_score":0.18317527},"labels":[],"label_agreement":null},{"id":"W3015122744","doi":"10.1002/sim.8595","title":"Graphical approaches for the control of generalized error rates","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada","keywords":"False discovery rate; Multiple comparisons problem; Graphical model; Context (archaeology); Word error rate; Control (management); Statistical hypothesis testing","score_opus":0.6677080684007569,"score_gpt":0.5681343876762122,"score_spread":0.09957368072454476,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3015122744","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00030008203,0.0003360807,0.9982242,0.0002840136,0.00008127501,0.00009403148,0.000047259637,0.0001907934,0.00044218748],"genre_scores_gemma":[0.055989627,0.0017522825,0.9353787,0.0009852336,0.00066195533,0.0031993953,0.00021366151,0.00044816907,0.0013710213],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7967098,0.17073055,0.0058993213,0.011695918,0.013279266,0.0016852161],"domain_scores_gemma":[0.49122587,0.4513897,0.018799685,0.025283558,0.012207051,0.0010941095],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.16187073,0.0055506094,0.004936186,0.00816677,0.0016045728,0.006288022,0.008702096,0.0068897554,0.0101646865],"category_scores_gemma":[0.41466942,0.002210646,0.0045737447,0.0069188145,0.011667391,0.007477795,0.008911767,0.013300019,0.002302108],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019289582,0.0000733289,0.0013089011,0.0006737842,0.0006214236,0.00025360996,0.0007840307,0.055880733,0.000833975,0.87306416,0.0027448556,0.06356826],"study_design_scores_gemma":[0.00019211632,0.0001853671,0.0006261682,0.00035314675,0.00023547673,0.00014966629,0.000077766636,0.21866627,0.0011162027,0.76946455,0.008807402,0.00012581525],"about_ca_topic_score_codex":0.0047020777,"about_ca_topic_score_gemma":0.0028965604,"teacher_disagreement_score":0.8381293,"about_ca_system_score_codex":0.0055488404,"about_ca_system_score_gemma":0.0055554174,"threshold_uncertainty_score":0.8560645},"labels":[],"label_agreement":null},{"id":"W3019567365","doi":"10.1002/sim.8541","title":"Doubly robust estimation and causal inference for recurrent event data","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Jewish General Hospital","funders":"Canadian Institutes of Health Research","keywords":"Estimator; Semiparametric regression; Causal inference; Computer science; Statistics; Semiparametric model; Inference; Event (particle physics); Econometrics; Estimating equations; Mathematics; Artificial intelligence","score_opus":0.32461444492443875,"score_gpt":0.5000145513770762,"score_spread":0.17540010645263743,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3019567365","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0055185095,0.0003593078,0.9930733,0.00017977378,0.000026856056,0.000034055596,0.00025234764,0.00016133915,0.00039457987],"genre_scores_gemma":[0.44559282,0.0014461385,0.54724854,0.00035166257,0.00025921047,0.00054547505,0.0019374941,0.00015054809,0.002468114],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9895377,0.0073270244,0.0005015093,0.0013461123,0.0010856807,0.00020196923],"domain_scores_gemma":[0.9445382,0.04245719,0.004831881,0.0062550805,0.0016678917,0.00024970795],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.021237196,0.0007002605,0.0017644025,0.0026544502,0.00045925667,0.0014763345,0.0030860456,0.0015169046,0.003246229],"category_scores_gemma":[0.096683435,0.0005269766,0.0016841159,0.002407922,0.0013344809,0.0019852426,0.0017529624,0.0022175235,0.0005486113],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00025646598,0.00019563649,0.022549061,0.00054055615,0.0011260669,0.00037597236,0.00044897507,0.18346137,0.001732697,0.51652527,0.0036919971,0.269096],"study_design_scores_gemma":[0.00006617786,0.00012905312,0.004972631,0.00011334408,0.00022286412,0.00024045337,0.00006995918,0.6511586,0.001770808,0.3353276,0.005859147,0.00006945706],"about_ca_topic_score_codex":0.002758307,"about_ca_topic_score_gemma":0.0018829824,"teacher_disagreement_score":0.021237196,"about_ca_system_score_codex":0.0007351401,"about_ca_system_score_gemma":0.0010632293,"threshold_uncertainty_score":0.11231434},"labels":[],"label_agreement":null},{"id":"W3019617362","doi":"10.1002/sim.8508","title":"A multiparameter regression model for interval‐censored survival data","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Irish Research Council","keywords":"Proportional hazards model; Statistics; Interval (graph theory); Weibull distribution; Survival analysis; Computer science; Regression; Parametric statistics; Accelerated failure time model; Confidence interval; Regression analysis; Parametric model; Longitudinal data; Econometrics; Mathematics; Data mining","score_opus":0.40073178722091246,"score_gpt":0.49977133220344255,"score_spread":0.09903954498253009,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3019617362","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0042723003,0.00048132893,0.99357796,0.000580003,0.000035770063,0.000058161844,0.00026928616,0.00015701796,0.0005680946],"genre_scores_gemma":[0.4347225,0.0039128102,0.5457262,0.0010139286,0.00039685398,0.001679997,0.0021269086,0.00035755048,0.010063184],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99389887,0.0039989897,0.000276618,0.000761381,0.00077229494,0.00029179532],"domain_scores_gemma":[0.9849893,0.010935038,0.0017683948,0.0014093596,0.00066121126,0.00023679099],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014513662,0.001276364,0.001691324,0.0016496904,0.00037085472,0.0017015315,0.0036201235,0.0020445026,0.0038997096],"category_scores_gemma":[0.03248673,0.00064967346,0.002433913,0.0024889933,0.0011983818,0.0023719335,0.0015911857,0.0040073935,0.0011507806],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001276032,0.00007593783,0.0033302293,0.0003446417,0.00023185312,0.00043388837,0.00032134532,0.64097315,0.0018375642,0.27297145,0.0037406369,0.075611636],"study_design_scores_gemma":[0.00004985662,0.00015013886,0.000977833,0.00008971872,0.00007141066,0.0002899945,0.000030519815,0.8744789,0.00037120606,0.11826313,0.0051598353,0.000067518624],"about_ca_topic_score_codex":0.0027132623,"about_ca_topic_score_gemma":0.0016736822,"teacher_disagreement_score":0.014513662,"about_ca_system_score_codex":0.0010419025,"about_ca_system_score_gemma":0.0014109167,"threshold_uncertainty_score":0.07675648},"labels":[],"label_agreement":null},{"id":"W3021389985","doi":"10.1002/sim.8560","title":"A tractable method to account for high‐dimensional nonignorable missing data in intensive longitudinal data","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"National Cancer Institute","keywords":"Missing data; Covariate; Computer science; Curse of dimensionality; Multinomial distribution; Outcome (game theory); Econometrics; Data mining; Statistics; Mathematics; Machine learning","score_opus":0.2896584364384498,"score_gpt":0.49417319198457804,"score_spread":0.20451475554612825,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3021389985","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0006790777,0.000056446974,0.998814,0.000117793,0.000014224172,0.0000365319,0.000036421556,0.000072238174,0.00017332062],"genre_scores_gemma":[0.0572943,0.00046490025,0.93792015,0.00036645454,0.00020192467,0.00085511664,0.00027569523,0.00020277395,0.002418668],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99505544,0.0031022436,0.00025484493,0.0005042788,0.0009248609,0.00015830851],"domain_scores_gemma":[0.97244656,0.022500986,0.0012781061,0.0020699436,0.0013140263,0.00039043205],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.018664353,0.0012640205,0.0019524689,0.002401417,0.0009802646,0.0014667497,0.0033942726,0.0018456578,0.0058637927],"category_scores_gemma":[0.06359266,0.001154318,0.0023014655,0.0023202754,0.0017090419,0.0027033852,0.0033906903,0.0036173519,0.0009859562],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00010669386,0.00014505119,0.005459231,0.0005165899,0.00047678532,0.0006165167,0.0006184344,0.4220672,0.0029212087,0.3454357,0.0071484526,0.2144881],"study_design_scores_gemma":[0.000044152683,0.00004115074,0.0006539859,0.00007434955,0.00006475094,0.00027559526,0.000041814397,0.83903986,0.0006755718,0.15400003,0.005052742,0.000036052043],"about_ca_topic_score_codex":0.00610754,"about_ca_topic_score_gemma":0.009347152,"teacher_disagreement_score":0.018664353,"about_ca_system_score_codex":0.0015436729,"about_ca_system_score_gemma":0.005455103,"threshold_uncertainty_score":0.098707676},"labels":[],"label_agreement":null},{"id":"W3022690991","doi":"10.1002/sim.8552","title":"Statistical properties of minimal sufficient balance and minimization as methods for controlling baseline covariate imbalance at the design stage of sequential clinical trials","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"National Center for Advancing Translational Sciences; National Institute of Arthritis and Musculoskeletal and Skin Diseases; National Institute of Neurological Disorders and Stroke; National Institute of Diabetes and Digestive and Kidney Diseases; National Institute of General Medical Sciences","keywords":"Covariate; Randomness; Minification; Baseline (sea); Computer science; Statistics; Type I and type II errors; Balance (ability); Mathematical optimization; Mathematics; Econometrics; Medicine","score_opus":0.7238147778658908,"score_gpt":0.6262184680458172,"score_spread":0.09759630982007361,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3022690991","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005257929,0.0019754672,0.988596,0.0011481257,0.00017717356,0.0007879669,0.00023188963,0.00022462527,0.0016007846],"genre_scores_gemma":[0.16225874,0.0023669002,0.8158245,0.00178192,0.00087153143,0.014633003,0.00049447484,0.0005353915,0.0012335062],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.6671675,0.29947186,0.00812889,0.0074761934,0.016524678,0.0012309478],"domain_scores_gemma":[0.3586441,0.58193976,0.025380658,0.02296343,0.009780174,0.001291916],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.24225706,0.0023524393,0.006972153,0.006072558,0.0012871829,0.003508496,0.0030904822,0.0040405984,0.004839093],"category_scores_gemma":[0.5237642,0.0016249331,0.004546312,0.0051215175,0.009652372,0.006594437,0.004670763,0.007081664,0.0009452398],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0052005886,0.00028670268,0.00660844,0.0043338714,0.0028035291,0.00036664555,0.0017347839,0.05162836,0.002879971,0.7248849,0.0071474,0.19212474],"study_design_scores_gemma":[0.0013923841,0.0021501665,0.0028227447,0.0012458148,0.0006262502,0.00039784968,0.00015099686,0.17039414,0.0031745825,0.8057046,0.011755398,0.00018513457],"about_ca_topic_score_codex":0.0006483825,"about_ca_topic_score_gemma":0.0003481861,"teacher_disagreement_score":0.75774294,"about_ca_system_score_codex":0.0030663218,"about_ca_system_score_gemma":0.006293834,"threshold_uncertainty_score":0.93443197},"labels":[],"label_agreement":null},{"id":"W3023339592","doi":"10.1002/sim.8557","title":"Point and interval estimation in two‐stage adaptive designs with time to event data and biomarker‐driven subpopulation selection","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Cancer Institute; Medical Research Council; National Institutes of Health; Medical Research Council Canada","keywords":"Estimator; Stage (stratigraphy); Event (particle physics); Population; Statistics; Categorical variable; Interval (graph theory); Hazard ratio; Computer science; Confidence interval; Estimation; Selection (genetic algorithm); Interval estimation; Proportional hazards model; Mathematics; Medicine; Artificial intelligence; Biology","score_opus":0.5202491863090454,"score_gpt":0.5605198589785412,"score_spread":0.0402706726694958,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3023339592","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0055702026,0.00030106428,0.99352187,0.00010033481,0.000039172806,0.00018820565,0.00004233511,0.00007350326,0.00016334806],"genre_scores_gemma":[0.22482888,0.00087706884,0.76852834,0.00041909824,0.00025086277,0.0029972848,0.00052988355,0.00009957404,0.0014689785],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9314802,0.060102373,0.0015020027,0.003507578,0.0026677197,0.0007400657],"domain_scores_gemma":[0.7205163,0.2554936,0.008068937,0.010374473,0.0046251416,0.00092143996],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09617643,0.0019576503,0.00304188,0.0022169666,0.00060242706,0.0018063963,0.0038348837,0.003654698,0.0030894347],"category_scores_gemma":[0.19051227,0.0015255931,0.003230351,0.002107356,0.0033552991,0.0032841286,0.0030298664,0.0046541337,0.0004681768],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0030488586,0.00055313675,0.019920945,0.0012723972,0.0021857205,0.0005990756,0.0014630244,0.44068637,0.003785595,0.28408742,0.0020339366,0.24036346],"study_design_scores_gemma":[0.000626795,0.0010067258,0.00254175,0.00014847191,0.00027918725,0.00013696577,0.00007370754,0.85022026,0.0016424964,0.14128488,0.0019288633,0.0001099527],"about_ca_topic_score_codex":0.0012598808,"about_ca_topic_score_gemma":0.00080602866,"teacher_disagreement_score":0.09617643,"about_ca_system_score_codex":0.00090431963,"about_ca_system_score_gemma":0.0018924861,"threshold_uncertainty_score":0.50863564},"labels":[],"label_agreement":null},{"id":"W3025024871","doi":"10.1002/sim.8555","title":"One‐stage individual participant data meta‐analysis models for continuous and binary outcomes: Comparison of treatment coding options and estimation methods","year":2020,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council Canada; Medical Research Council; National Institute for Health and Care Research","keywords":"Restricted maximum likelihood; Statistics; Random effects model; Meta-analysis; Confidence interval; Mathematics; Estimation; Mixed model; Maximum likelihood; Variance (accounting); Covariate; Standard error; Binary number; Medicine","score_opus":0.9611799200684782,"score_gpt":0.7025338956772175,"score_spread":0.2586460243912607,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3025024871","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005279276,0.048479203,0.92951965,0.002199588,0.0007634626,0.008666579,0.0026743775,0.00089613197,0.001521724],"genre_scores_gemma":[0.087068446,0.0184118,0.8457195,0.0018678504,0.00024511595,0.043022495,0.0019330225,0.00045829118,0.001273544],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.63612753,0.3362302,0.0127914995,0.00694718,0.0071959053,0.0007077478],"domain_scores_gemma":[0.63537025,0.3306889,0.010882176,0.016267555,0.0062353155,0.00055583636],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2467861,0.00441155,0.01088817,0.006482336,0.00090774597,0.0066286293,0.006762646,0.005256345,0.009731521],"category_scores_gemma":[0.42616966,0.0032911394,0.038572673,0.0063422774,0.0019250253,0.0074348836,0.0041705673,0.009654119,0.0015346206],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.015411821,0.0006658616,0.0070925937,0.13683788,0.243324,0.00034062212,0.0026273525,0.08769717,0.0012538977,0.10181772,0.011119045,0.3918121],"study_design_scores_gemma":[0.019589035,0.0047377693,0.005663226,0.04765752,0.26720136,0.0007630164,0.00059806916,0.2379872,0.0041900002,0.36717224,0.043136965,0.0013036112],"about_ca_topic_score_codex":0.003264129,"about_ca_topic_score_gemma":0.0037233727,"teacher_disagreement_score":0.7532139,"about_ca_system_score_codex":0.0030224177,"about_ca_system_score_gemma":0.0056741275,"threshold_uncertainty_score":0.92884684},"labels":[],"label_agreement":null},{"id":"W3028357212","doi":"10.1002/sim.8868","title":"The optimal design of clinical trials with potential biomarker effects: A novel computational approach","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Saskatchewan; University of Victoria","funders":"Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs; Western Canada Research Grid; Compute Canada","keywords":"Computer science; Scalability; Smoothing; Curse of dimensionality; Personalized medicine; Clinical trial; Software; Data mining; Machine learning; Bioinformatics","score_opus":0.6846230401092185,"score_gpt":0.6258049677645551,"score_spread":0.05881807234466341,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3028357212","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0037811606,0.00062735076,0.9926668,0.0008650966,0.00008790555,0.00016781125,0.00013007033,0.0002189611,0.0014549867],"genre_scores_gemma":[0.10186774,0.0007936087,0.89205766,0.0008357665,0.0002632521,0.0015181379,0.0004595561,0.00024424677,0.0019600072],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.995188,0.003182682,0.00021871958,0.00049318443,0.00067698944,0.00024036314],"domain_scores_gemma":[0.9672125,0.029488305,0.0010335494,0.0007421088,0.0009444384,0.0005791241],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009266485,0.0019096206,0.003983685,0.002923657,0.0007985735,0.0025953983,0.0025741446,0.0023094951,0.006069054],"category_scores_gemma":[0.028708087,0.002024596,0.0032871375,0.002793238,0.0022568968,0.0018612285,0.0027567649,0.0034147205,0.0005777348],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001463509,0.000064746724,0.0007011702,0.0002272268,0.00020987084,0.00007835445,0.000057407895,0.9434711,0.00025889036,0.027773658,0.0016258816,0.025385294],"study_design_scores_gemma":[0.00010515776,0.000036489666,0.000076662305,0.000031034724,0.000058669884,0.000019881818,0.000009030301,0.9626015,0.0001089053,0.03563892,0.0013038367,0.000009954641],"about_ca_topic_score_codex":0.0061132936,"about_ca_topic_score_gemma":0.00662145,"teacher_disagreement_score":0.009266485,"about_ca_system_score_codex":0.0019243784,"about_ca_system_score_gemma":0.007739719,"threshold_uncertainty_score":0.049006402},"labels":[],"label_agreement":null},{"id":"W3032046824","doi":"10.1002/sim.8757","title":"Analysis of time‐to‐event for observational studies: Guidance to the use of intensity models","year":2020,"lang":"en","type":"preprint","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; University of Waterloo","funders":"Canadian Institutes of Health Research; National Cancer Institute; National Institutes of Health; Javna Agencija za Raziskovalno Dejavnost RS; Natural Sciences and Engineering Research Council of Canada; McGill University","keywords":"Censoring (clinical trials); Covariate; Event (particle physics); Hazard; Computer science; Proportional hazards model; Observational study; Econometrics; Regression analysis; Goodness of fit; Software; Statistics; Data mining; Machine learning; Mathematics","score_opus":0.6177402007733431,"score_gpt":0.5192164900705176,"score_spread":0.0985237107028255,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3032046824","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00024377814,0.00091261754,0.9930454,0.002050551,0.00026858004,0.00032274635,0.0014107821,0.0009225922,0.0008230071],"genre_scores_gemma":[0.0053435354,0.0016001129,0.98567134,0.0011202502,0.00048020165,0.0023091745,0.0012544555,0.0012484441,0.00097253226],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8686574,0.11031056,0.00857197,0.003827861,0.008093101,0.00053909933],"domain_scores_gemma":[0.44741157,0.50222063,0.0113606835,0.02646254,0.010934618,0.0016100019],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.13830979,0.002772323,0.0039276783,0.007300742,0.0012324424,0.0052921693,0.006726359,0.0042555346,0.036025126],"category_scores_gemma":[0.4717564,0.002609852,0.0057056206,0.007245979,0.0034037058,0.005789439,0.0059114387,0.011031285,0.0154573545],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004787412,0.00023911973,0.0058322838,0.0060037607,0.001131507,0.0012964357,0.002073588,0.027849352,0.002065697,0.49203852,0.16034605,0.3006449],"study_design_scores_gemma":[0.0002434453,0.00017442692,0.0014239427,0.00220813,0.00019097199,0.00068275013,0.00019219797,0.049893223,0.0015284172,0.766178,0.17710179,0.00018270983],"about_ca_topic_score_codex":0.0039449385,"about_ca_topic_score_gemma":0.004188687,"teacher_disagreement_score":0.8616902,"about_ca_system_score_codex":0.0018355359,"about_ca_system_score_gemma":0.0074940673,"threshold_uncertainty_score":0.7314608},"labels":[],"label_agreement":null},{"id":"W3034503810","doi":"10.1002/sim.8570","title":"Graphical calibration curves and the integrated calibration index (ICI) for survival models","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":214,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Sunnybrook Hospital; University of Toronto","funders":"National Center for Advancing Translational Sciences; Canadian Institutes of Health Research; National Institutes of Health; Ontario Ministry of Health and Long-Term Care; Georgia Clinical and Translational Science Alliance; Heart and Stroke Foundation of Canada; Institute for Clinical Evaluative Sciences; Patient-Centered Outcomes Research Institute; Vanderbilt Institute for Clinical and Translational Research; Vanderbilt University","keywords":"Calibration; Context (archaeology); Statistics; Proportional hazards model; Percentile; Regression analysis; Mathematics; Survival analysis; Regression; Computer science","score_opus":0.32714729481711025,"score_gpt":0.4229108498472749,"score_spread":0.09576355503016465,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3034503810","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014209677,0.0014778593,0.97691786,0.00074258656,0.000095390336,0.00019192846,0.00073955505,0.0012203293,0.0044049365],"genre_scores_gemma":[0.5083797,0.0022018896,0.48014408,0.0006662584,0.0003850683,0.0015177832,0.0032618747,0.0015146858,0.0019285704],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.95744133,0.029576488,0.0018491294,0.0032168955,0.0071200537,0.0007960971],"domain_scores_gemma":[0.68040365,0.25501037,0.0265016,0.023053043,0.013707163,0.0013241451],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.060516156,0.0021421064,0.0015826124,0.011027962,0.0010295439,0.0042027347,0.0030379363,0.003744332,0.0053733224],"category_scores_gemma":[0.33016378,0.0009500146,0.002436946,0.009534793,0.004331253,0.0067864344,0.004423625,0.0057667117,0.0011859908],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000244765,0.00012561689,0.023084318,0.00056326773,0.00040462904,0.0002132832,0.0007282507,0.6102901,0.00087693427,0.21856639,0.0081557175,0.13674669],"study_design_scores_gemma":[0.000051258576,0.00022438758,0.009718898,0.0003875072,0.00009001567,0.00041720486,0.00019619123,0.7303763,0.001541395,0.24698018,0.009764533,0.00025209624],"about_ca_topic_score_codex":0.0029306782,"about_ca_topic_score_gemma":0.0012886918,"teacher_disagreement_score":0.060516156,"about_ca_system_score_codex":0.0035160666,"about_ca_system_score_gemma":0.0026502172,"threshold_uncertainty_score":0.32004386},"labels":[],"label_agreement":null},{"id":"W3038020972","doi":"10.1002/sim.8657","title":"Constructing inverse probability weights for institutional comparisons in healthcare","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Princess Margaret Cancer Centre; University Health Network; Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Health care; Inverse; Inverse probability; Computer science; Statistics; Econometrics; Mathematics; Bayesian probability; Posterior probability; Political science","score_opus":0.4797407362015364,"score_gpt":0.46809264349482754,"score_spread":0.011648092706708857,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3038020972","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01358039,0.00014843665,0.9829768,0.0002499348,0.00011818273,0.0003649212,0.00018546479,0.00024290025,0.002132899],"genre_scores_gemma":[0.15876634,0.00015981881,0.8377595,0.00017115961,0.000110149536,0.0016586133,0.00034683218,0.00022090992,0.00080668397],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8919434,0.08671098,0.0046150316,0.0076134345,0.008031399,0.0010857392],"domain_scores_gemma":[0.7162766,0.23320343,0.013503327,0.0268755,0.008902803,0.0012383945],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.110144064,0.0017360855,0.0019945807,0.010289413,0.0018635553,0.0050742924,0.0032485072,0.0025325431,0.0075527756],"category_scores_gemma":[0.432826,0.0012831693,0.0030466383,0.011245446,0.0063606375,0.009507458,0.008054517,0.004658191,0.0008069247],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00037989055,0.00020771843,0.021363322,0.0006106602,0.00080203736,0.00022120867,0.0029715472,0.048609056,0.001005913,0.6015901,0.004359076,0.3178794],"study_design_scores_gemma":[0.00009335562,0.00024072907,0.0059270966,0.00023762518,0.00015250318,0.0001845118,0.0005819048,0.13337074,0.0012245815,0.8460459,0.011815041,0.00012593703],"about_ca_topic_score_codex":0.0042693797,"about_ca_topic_score_gemma":0.0030495971,"teacher_disagreement_score":0.110144064,"about_ca_system_score_codex":0.003433415,"about_ca_system_score_gemma":0.002468004,"threshold_uncertainty_score":0.58250445},"labels":[],"label_agreement":null},{"id":"W3043139790","doi":"10.1002/sim.8584","title":"A general method for elicitation, imputation, and sensitivity analysis for incomplete repeated binary data","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"NIHR School for Primary Care Research; Medical Research Council Canada; Medical Research Council; National Institute for Health and Care Research","keywords":"Pooling; Missing data; Imputation (statistics); Expert opinion; Expert elicitation; Statistics; Computer science; Econometrics; Medicine; Artificial intelligence; Mathematics","score_opus":0.16647558434536588,"score_gpt":0.4701183694083897,"score_spread":0.3036427850630238,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3043139790","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00016441381,0.00005503445,0.9983222,0.00010245982,0.000023140125,0.000625804,0.00016803209,0.00022107083,0.00031793694],"genre_scores_gemma":[0.009281385,0.00019082295,0.9824572,0.00019012853,0.000060425216,0.0069128578,0.00027798626,0.00018405633,0.00044501474],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8196725,0.15964685,0.005511345,0.004502681,0.009835472,0.0008311364],"domain_scores_gemma":[0.66240865,0.2942905,0.009775685,0.024804233,0.008117535,0.0006034103],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.13609527,0.0033646487,0.0044271736,0.008036437,0.0017540633,0.0036578388,0.0047253068,0.0035821656,0.015952181],"category_scores_gemma":[0.33274195,0.003126297,0.008016099,0.006983652,0.0034857607,0.0040248935,0.007394305,0.008117261,0.0029867569],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004831387,0.00028260704,0.003362135,0.0038347566,0.0028300164,0.0006203878,0.0018859418,0.13729607,0.0027345163,0.52627164,0.01787045,0.30252832],"study_design_scores_gemma":[0.0003697047,0.00030707027,0.001098001,0.001001936,0.0005254502,0.00059670897,0.00018095439,0.33710673,0.002902771,0.61623275,0.039359033,0.0003188683],"about_ca_topic_score_codex":0.0026954776,"about_ca_topic_score_gemma":0.0025793812,"teacher_disagreement_score":0.13609527,"about_ca_system_score_codex":0.0025303192,"about_ca_system_score_gemma":0.006834042,"threshold_uncertainty_score":0.7197492},"labels":[],"label_agreement":null},{"id":"W3044494018","doi":"10.1002/sim.8592","title":"Design and sample size considerations for valuation studies of multi‐attribute utility instruments","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research","keywords":"Statistics; Valuation (finance); Mean squared error; Econometrics; Gaussian; Correlation; Mathematics; Sample size determination; Regression analysis; Value (mathematics); Economics","score_opus":0.7317871135772142,"score_gpt":0.5185467643539607,"score_spread":0.2132403492232534,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3044494018","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0694278,0.0028066714,0.7246683,0.00974417,0.0038033845,0.17223217,0.0020774035,0.00045020456,0.014789858],"genre_scores_gemma":[0.14172569,0.0005415567,0.5185741,0.00264743,0.00044232764,0.33368132,0.0004050313,0.000079496975,0.0019029974],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.63181657,0.33042064,0.012967,0.005487945,0.017754378,0.0015535199],"domain_scores_gemma":[0.64798194,0.29246038,0.012307636,0.027920628,0.017033989,0.0022953318],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2880413,0.001481307,0.002612398,0.0015531229,0.0017087164,0.0025278183,0.0031720945,0.0038535786,0.009472125],"category_scores_gemma":[0.4583406,0.0015529956,0.0031876643,0.0021298109,0.0022770022,0.0025345495,0.0028371087,0.0033479978,0.0016843212],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.08897051,0.006042918,0.03967064,0.008970817,0.0047898563,0.0022267564,0.0075483914,0.028732851,0.013531177,0.259981,0.03164675,0.5078884],"study_design_scores_gemma":[0.05730149,0.095553756,0.0772415,0.007864918,0.006100227,0.0022681202,0.0022932438,0.15921192,0.02604047,0.30304602,0.2623886,0.0006897584],"about_ca_topic_score_codex":0.00088273466,"about_ca_topic_score_gemma":0.0014299169,"teacher_disagreement_score":0.7119587,"about_ca_system_score_codex":0.0016249903,"about_ca_system_score_gemma":0.0044802786,"threshold_uncertainty_score":0.8779718},"labels":[],"label_agreement":null},{"id":"W3045859451","doi":"10.1002/sim.8662","title":"A bivariate autoregressive Poisson model and its application to asthma‐related emergency room visits","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Windsor; University of Calgary","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Bivariate analysis; Autoregressive model; Statistics; Poisson regression; Poisson distribution; Overdispersion; Estimator; Bayesian probability; Series (stratigraphy); Econometrics; Time series; Computer science; Mathematics; Count data; Medicine","score_opus":0.019389079551677096,"score_gpt":0.3185714517097249,"score_spread":0.29918237215804777,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3045859451","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01752616,0.00067833666,0.9788215,0.0009720487,0.00004829757,0.00005509258,0.0002591283,0.00016699005,0.0014724366],"genre_scores_gemma":[0.57664984,0.0036248518,0.4092631,0.000399627,0.00033807202,0.0006430751,0.0011253796,0.00018155483,0.007774537],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99745995,0.0015147426,0.000090145295,0.0003185622,0.0004835894,0.00013310913],"domain_scores_gemma":[0.9931855,0.0052668657,0.0005725026,0.0002737175,0.00057048554,0.00013098848],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0075470055,0.00052387355,0.0010534939,0.0015752871,0.0006373264,0.0011656987,0.002550648,0.0014582757,0.0020401923],"category_scores_gemma":[0.026178202,0.000776169,0.0013742566,0.0027297735,0.001176167,0.0011093693,0.0013508636,0.0018098681,0.0004616714],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007497406,0.00007388071,0.011316465,0.00017336826,0.0001820236,0.00036509428,0.0006208272,0.43822563,0.0010149791,0.47134456,0.0040323534,0.0725758],"study_design_scores_gemma":[0.00001910208,0.000023174569,0.0025711427,0.000033642646,0.000037031055,0.000088479974,0.000078046,0.91647255,0.00011949599,0.077492125,0.0030283923,0.000036843125],"about_ca_topic_score_codex":0.08099973,"about_ca_topic_score_gemma":0.059429917,"teacher_disagreement_score":0.08099973,"about_ca_system_score_codex":0.0020078563,"about_ca_system_score_gemma":0.0033543187,"threshold_uncertainty_score":0.16105652},"labels":[],"label_agreement":null},{"id":"W3081371020","doi":"10.1002/sim.8715","title":"A tutorial on dealing with time‐varying eligibility for treatment: Comparing the risk of major bleeding with direct‐acting oral anticoagulant<scp>s</scp> vs warfarin","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre Hospitalier de l’Université de Montréal; McGill University; McGill University Health Centre; Université de Montréal","funders":"Fonds de Recherche du Québec - Santé; Canadian Institutes of Health Research","keywords":"Contraindication; Covariate; Weighting; Warfarin; Computer science; Medicine; Statistics; Econometrics; Mathematics; Internal medicine; Atrial fibrillation; Alternative medicine","score_opus":0.25340054521848643,"score_gpt":0.4219089152046647,"score_spread":0.1685083699861783,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3081371020","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0042837434,0.118618384,0.7381936,0.046100035,0.016448114,0.0010826539,0.011703672,0.0038906443,0.05967918],"genre_scores_gemma":[0.031147256,0.13790406,0.67009175,0.04720532,0.03184769,0.004381038,0.00994084,0.0041732076,0.06330886],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9965783,0.0023296834,0.00025262058,0.0002712977,0.00048057237,0.00008767002],"domain_scores_gemma":[0.9604501,0.037214335,0.0007306582,0.00041054338,0.0008305325,0.00036369942],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006547182,0.0025482858,0.0020181406,0.0024698984,0.00047253177,0.0017015776,0.0016777273,0.0026964878,0.049377304],"category_scores_gemma":[0.047660906,0.00088246755,0.0029967502,0.0026139726,0.0009701392,0.0033411423,0.0014834004,0.0042479583,0.011430844],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00022551876,0.00025207465,0.0019021606,0.003639827,0.00052131433,0.0006707867,0.0003476499,0.012612926,0.0014159834,0.08975469,0.6139659,0.27469113],"study_design_scores_gemma":[0.00018555492,0.00048859196,0.004191304,0.0027455976,0.00026184,0.0018441677,0.00018486146,0.01903033,0.00083606725,0.2659003,0.7041703,0.00016109456],"about_ca_topic_score_codex":0.0028761814,"about_ca_topic_score_gemma":0.0035428891,"teacher_disagreement_score":0.049377304,"about_ca_system_score_codex":0.0011692479,"about_ca_system_score_gemma":0.0015596331,"threshold_uncertainty_score":0.16518354},"labels":[],"label_agreement":null},{"id":"W3081450496","doi":"10.1002/sim.8727","title":"Statistical inference for missing data mechanisms","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Regina","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Missing data; Inference; Statistical inference; Computer science; Statistics; Econometrics; Data mining; Artificial intelligence; Machine learning; Mathematics","score_opus":0.33318924658330423,"score_gpt":0.5021841392437316,"score_spread":0.1689948926604274,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3081450496","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0019257113,0.00065783144,0.9960835,0.0005537399,0.00010252296,0.00007796798,0.0001351362,0.000111702204,0.00035183554],"genre_scores_gemma":[0.30050412,0.0045954194,0.6860967,0.0018636448,0.0012536356,0.0019058866,0.0012316124,0.00023436715,0.0023146807],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9260356,0.050785296,0.004142241,0.011168208,0.006724502,0.0011441404],"domain_scores_gemma":[0.6968322,0.24568598,0.019213775,0.029912025,0.0075384686,0.00081755966],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.12577404,0.0021402347,0.0053884867,0.005133943,0.0018066457,0.004435868,0.0077979085,0.004818941,0.004900499],"category_scores_gemma":[0.3500608,0.002155295,0.0045634573,0.0063378373,0.006370177,0.011486954,0.005490779,0.010263229,0.0009828602],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00021931504,0.00012476607,0.010654941,0.0017120874,0.0016610498,0.0005519546,0.0008503277,0.07050579,0.00082932226,0.7849359,0.004329601,0.12362498],"study_design_scores_gemma":[0.00009071686,0.00010636527,0.0015086004,0.00041047533,0.00027508417,0.00034906573,0.00014773017,0.21251263,0.00093336555,0.7802673,0.0033253778,0.00007325878],"about_ca_topic_score_codex":0.0016694617,"about_ca_topic_score_gemma":0.00089849095,"teacher_disagreement_score":0.874226,"about_ca_system_score_codex":0.0025043457,"about_ca_system_score_gemma":0.0048737987,"threshold_uncertainty_score":0.66516465},"labels":[],"label_agreement":null},{"id":"W3081944489","doi":"10.1002/sim.8710","title":"Impact of discretization of the timeline for longitudinal causal inference methods","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hôpital du Sacré-Cœur de Montréal; Université de Montréal; McGill University","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Timeline; Causal inference; Inference; Computer science; Econometrics; Discretization; Statistics; Mathematics; Artificial intelligence","score_opus":0.22839864773798854,"score_gpt":0.5624154422413042,"score_spread":0.3340167945033157,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3081944489","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06161997,0.0019082795,0.9276191,0.0040429365,0.00048461708,0.00056778477,0.00075713656,0.00044896032,0.0025511954],"genre_scores_gemma":[0.49561742,0.0010410687,0.49884647,0.0011791011,0.00018542426,0.0012938598,0.00090095185,0.00015357815,0.00078217924],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.89526594,0.08817572,0.0060284263,0.0039179227,0.0059027052,0.0007092889],"domain_scores_gemma":[0.3750693,0.56479466,0.015495279,0.038698718,0.004790054,0.001151986],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11885788,0.0008054529,0.0013677367,0.00096878415,0.0014055583,0.0035639158,0.0029886956,0.0024982449,0.0032107153],"category_scores_gemma":[0.4100285,0.0011034281,0.0018487999,0.001986005,0.0033333534,0.004474387,0.003612807,0.006035776,0.00039959175],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0033877962,0.00053194864,0.07314793,0.0017024296,0.0012928165,0.0010182025,0.007141698,0.3666005,0.0034674243,0.33231816,0.005223279,0.20416778],"study_design_scores_gemma":[0.0009246971,0.0013354005,0.014364355,0.0019900769,0.00046133835,0.0006445866,0.0021998342,0.6210288,0.0052254004,0.33395597,0.017644582,0.00022491706],"about_ca_topic_score_codex":0.007926384,"about_ca_topic_score_gemma":0.0055456185,"teacher_disagreement_score":0.11885788,"about_ca_system_score_codex":0.002715884,"about_ca_system_score_gemma":0.0036251713,"threshold_uncertainty_score":0.6285881},"labels":[],"label_agreement":null},{"id":"W3082894717","doi":"10.1002/sim.8722","title":"Multiply robust estimation of causal quantile treatment effects","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; Roche (Canada)","funders":"","keywords":"Quantile; Causal inference; Estimator; Propensity score matching; Econometrics; Statistics; Inverse probability weighting; Empirical likelihood; Weighting; Average treatment effect; Inference; Computer science; Estimation; Mathematics; Artificial intelligence; Medicine; Economics","score_opus":0.1383377641491244,"score_gpt":0.4358404304734299,"score_spread":0.2975026663243055,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3082894717","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0031544173,0.00021280635,0.99585706,0.00014471293,0.000016451184,0.00004616599,0.00006547952,0.00012954712,0.00037321792],"genre_scores_gemma":[0.2945203,0.00095795054,0.7004204,0.00031437541,0.00024092774,0.0006306437,0.00046970005,0.00013628844,0.0023094655],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9934881,0.004347686,0.00024426437,0.0007469879,0.0009773752,0.00019555113],"domain_scores_gemma":[0.9747704,0.019750113,0.0017522662,0.0027254403,0.000843947,0.00015771162],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012897742,0.000689577,0.0018319618,0.0017507917,0.0003985304,0.0011059216,0.0018972225,0.0014303746,0.005400232],"category_scores_gemma":[0.05471754,0.0006043942,0.0015899158,0.0018794655,0.0013743901,0.0016225327,0.002364363,0.002485651,0.00049799186],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00028441023,0.00012944404,0.006560375,0.0003840661,0.00060968,0.00021080379,0.0002454212,0.23805161,0.0031699722,0.49320245,0.0020300506,0.25512183],"study_design_scores_gemma":[0.000095462055,0.00016802724,0.0031955168,0.00007754027,0.00013959353,0.000178787,0.000034513123,0.6596077,0.0025875159,0.32970443,0.004164181,0.000046680223],"about_ca_topic_score_codex":0.0012156415,"about_ca_topic_score_gemma":0.0007749349,"teacher_disagreement_score":0.012897742,"about_ca_system_score_codex":0.0008045588,"about_ca_system_score_gemma":0.0012270041,"threshold_uncertainty_score":0.0682106},"labels":[],"label_agreement":null},{"id":"W3085834915","doi":"10.1002/sim.8688","title":"Genetic association studies with bivariate mixed responses subject to measurement error and misclassification","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic and phenotypic traits in livestock","field":"Biochemistry, Genetics and Molecular Biology","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Bivariate analysis; Univariate; Covariate; Trait; Statistics; Computer science; Pleiotropy; Econometrics; Association (psychology); Binary number; Mathematics; Multivariate statistics; Psychology; Genetics; Biology; Phenotype; Gene","score_opus":0.059095089833382373,"score_gpt":0.31206555386596074,"score_spread":0.2529704640325784,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3085834915","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04006364,0.0019038295,0.9556976,0.0009016309,0.00022626178,0.00014607931,0.00031876608,0.0002458969,0.00049631053],"genre_scores_gemma":[0.66964525,0.0015006458,0.32177067,0.0014613937,0.0003613372,0.0010037768,0.0009783978,0.00013177162,0.0031468414],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9401514,0.043586254,0.0035331775,0.009119478,0.0027727878,0.00083690416],"domain_scores_gemma":[0.81186163,0.14273581,0.020988204,0.019856839,0.003921793,0.0006357623],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09055704,0.0013043078,0.0029571196,0.0017957729,0.0010162537,0.0026264137,0.0034027533,0.0030748714,0.0015137065],"category_scores_gemma":[0.18956237,0.0009715326,0.0023649717,0.0037082199,0.003070492,0.0018286766,0.0024878737,0.002217283,0.0004371537],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015246836,0.00032223546,0.34377423,0.0022400063,0.0057771704,0.0035808103,0.0031745161,0.142946,0.004172133,0.23919968,0.0041079884,0.24918064],"study_design_scores_gemma":[0.00033368278,0.0005386778,0.07090177,0.0008745404,0.0031520003,0.0029457223,0.0006899314,0.52782404,0.0068107913,0.37215906,0.013488387,0.00028129824],"about_ca_topic_score_codex":0.0024072106,"about_ca_topic_score_gemma":0.0017784399,"teacher_disagreement_score":0.09055704,"about_ca_system_score_codex":0.0010422489,"about_ca_system_score_gemma":0.0013483821,"threshold_uncertainty_score":0.47891712},"labels":[],"label_agreement":null},{"id":"W3086774075","doi":"10.1002/sim.8738","title":"Meta‐analysis of quantile intervals from different studies with an application to a pulmonary tuberculosis data","year":2020,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Quantile; Confidence interval; Statistics; Coverage probability; CDF-based nonparametric confidence interval; Meta-analysis; Random effects model; Estimator; Mathematics; Robust confidence intervals; Confidence distribution; Econometrics; Medicine; Internal medicine","score_opus":0.3857157534275251,"score_gpt":0.5229027888145859,"score_spread":0.13718703538706073,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3086774075","genre_codex":"review","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00831725,0.8963886,0.08913118,0.001195957,0.0004918307,0.0010031657,0.0018669105,0.00022386586,0.0013812043],"genre_scores_gemma":[0.3295629,0.50401014,0.15405329,0.0019579015,0.00071056955,0.005054165,0.0034229222,0.00021691047,0.001011142],"study_design_codex":"meta_analysis","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95395225,0.035191383,0.004836248,0.0023218328,0.0034300608,0.0002681934],"domain_scores_gemma":[0.9106427,0.07861858,0.004709466,0.003860101,0.0020284331,0.00014084305],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.053913992,0.001774208,0.0065828986,0.007511768,0.0004318225,0.002833066,0.0019965854,0.0015208372,0.0025720275],"category_scores_gemma":[0.13726908,0.00093818665,0.019866563,0.009208939,0.00066688244,0.0014408441,0.001488931,0.002448655,0.00025949697],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002782547,0.00010403603,0.008714574,0.18661648,0.51545906,0.00039676426,0.00026570843,0.011579098,0.0013202691,0.007882759,0.003471043,0.26140767],"study_design_scores_gemma":[0.0025603655,0.0011143858,0.016352514,0.030526374,0.86260325,0.00080020854,0.00023660506,0.010168588,0.0032049501,0.03879046,0.033439226,0.00020308605],"about_ca_topic_score_codex":0.0019944638,"about_ca_topic_score_gemma":0.002540218,"teacher_disagreement_score":0.053913992,"about_ca_system_score_codex":0.0015748738,"about_ca_system_score_gemma":0.002235382,"threshold_uncertainty_score":0.28512788},"labels":[],"label_agreement":null},{"id":"W3087039075","doi":"10.1002/sim.8752","title":"Bayesian estimation of multivariate Gaussian Markov random fields with constraint","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Spatial and Panel Data Analysis","field":"Economics, Econometrics and Finance","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Multivariate statistics; Deviance information criterion; Computer science; Bayesian probability; Prior probability; Bayesian hierarchical modeling; Markov chain Monte Carlo; Copula (linguistics); Mathematics; Statistics; Bayesian inference; Artificial intelligence; Econometrics","score_opus":0.025308784439702175,"score_gpt":0.2580846507327724,"score_spread":0.23277586629307026,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3087039075","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005983888,0.00021157907,0.9929166,0.00025764265,0.000015470843,0.000025722025,0.000116655225,0.00007929934,0.00039313207],"genre_scores_gemma":[0.45437375,0.00196615,0.5370611,0.000507448,0.00027671762,0.0005182403,0.0012802959,0.00017056713,0.003845759],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99511284,0.003236514,0.00016507234,0.00070297095,0.0005020674,0.00028055662],"domain_scores_gemma":[0.9769974,0.019151434,0.0016044006,0.0010668976,0.0008926445,0.00028718667],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014120947,0.001263516,0.0023117044,0.0015768694,0.00074061786,0.0020984425,0.003168505,0.0023616226,0.0020664071],"category_scores_gemma":[0.0405514,0.0010996294,0.0018559908,0.002084708,0.002655934,0.0035140738,0.0024086116,0.0032804282,0.000425995],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000067496396,0.000032066793,0.0016920461,0.0001076649,0.00008399581,0.00011750169,0.00012329864,0.64836544,0.000551851,0.32480118,0.00095490715,0.0231025],"study_design_scores_gemma":[0.000008520214,0.000013445801,0.00023234355,0.000016081134,0.000009506073,0.000022218073,0.000009309531,0.90542316,0.00014567029,0.09370212,0.00040027907,0.00001735417],"about_ca_topic_score_codex":0.009184508,"about_ca_topic_score_gemma":0.006511465,"teacher_disagreement_score":0.014120947,"about_ca_system_score_codex":0.0019525812,"about_ca_system_score_gemma":0.002826954,"threshold_uncertainty_score":0.07467961},"labels":[],"label_agreement":null},{"id":"W3087499679","doi":"10.1002/sim.8735","title":"Finite sample variance estimation for optimal dynamic treatment regimes of survival outcomes","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Jewish General Hospital; McGill University","funders":"","keywords":"Estimator; Nuisance parameter; Confidence interval; Delta method; Variance (accounting); Statistics; Mathematics; Sample size determination; Econometrics; Computer science; Applied mathematics","score_opus":0.13053276575746314,"score_gpt":0.442096167453149,"score_spread":0.31156340169568586,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3087499679","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009330455,0.00063711265,0.9888101,0.00020858756,0.00003239282,0.00005808713,0.000037455797,0.00016993778,0.0007158702],"genre_scores_gemma":[0.39810696,0.0009145847,0.5988315,0.0003026321,0.0001048439,0.0007759063,0.00039256798,0.00019433432,0.00037669096],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.972139,0.020798516,0.0010592157,0.0019510897,0.0035754722,0.00047665022],"domain_scores_gemma":[0.8146005,0.16431074,0.007049053,0.007748459,0.0057602744,0.00053094],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.045617323,0.00064849993,0.0015662676,0.0036066843,0.0005864056,0.0022219499,0.0022693346,0.0016698599,0.0017896916],"category_scores_gemma":[0.25344872,0.00049479847,0.0016136258,0.0020447401,0.00239402,0.003200581,0.0023098774,0.0032096605,0.00034759459],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004118604,0.00019611871,0.018625006,0.00086523755,0.0006463655,0.00020380267,0.00088435825,0.14814772,0.0031756146,0.36829296,0.0031377766,0.45541325],"study_design_scores_gemma":[0.00014853707,0.00028563716,0.00677414,0.00055835885,0.00017013594,0.00025608623,0.00019494149,0.61862934,0.004448796,0.3635087,0.0049090637,0.00011629846],"about_ca_topic_score_codex":0.0015300004,"about_ca_topic_score_gemma":0.0010845739,"teacher_disagreement_score":0.045617323,"about_ca_system_score_codex":0.0014671339,"about_ca_system_score_gemma":0.002203757,"threshold_uncertainty_score":0.2412504},"labels":[],"label_agreement":null},{"id":"W3089031369","doi":"10.1002/sim.8744","title":"Comparing Kaplan‐Meier curves with the probability of agreement","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Air Force Office of Scientific Research; Natural Sciences and Engineering Research Council of Canada","keywords":"Nonparametric statistics; Similarity (geometry); Statistics; Confidence interval; Computer science; Reliability (semiconductor); Coverage probability; Point estimation; Mathematics; Sample size determination; Contrast (vision); Econometrics; Artificial intelligence","score_opus":0.1459445148099727,"score_gpt":0.39587051930710915,"score_spread":0.24992600449713645,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3089031369","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.023200952,0.00515184,0.9611376,0.0011497968,0.0003993016,0.0006587475,0.0010801356,0.001428422,0.00579317],"genre_scores_gemma":[0.5780479,0.0036433763,0.40992105,0.00064672006,0.00055979285,0.0028488648,0.0016015439,0.00092145253,0.0018093598],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9261653,0.053487815,0.0047975983,0.0057589826,0.008990276,0.0007999876],"domain_scores_gemma":[0.5825381,0.3578052,0.026674861,0.019041134,0.012422038,0.0015186083],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09638296,0.0015251007,0.0017313475,0.00953615,0.00088445377,0.0047741323,0.0024272916,0.002828636,0.008524261],"category_scores_gemma":[0.33320358,0.0007972973,0.0024972588,0.0048460006,0.004299488,0.008416049,0.004961409,0.004761392,0.0011963921],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0030750881,0.00021749471,0.0528262,0.0039645047,0.0039678887,0.00067890494,0.004436661,0.15837546,0.002060551,0.30354014,0.017645063,0.44921204],"study_design_scores_gemma":[0.00037366853,0.0018503007,0.03357028,0.0021306672,0.001057276,0.0022573555,0.0021549822,0.33322248,0.0056611015,0.5616445,0.05530655,0.0007708046],"about_ca_topic_score_codex":0.0014857379,"about_ca_topic_score_gemma":0.0006665435,"teacher_disagreement_score":0.903617,"about_ca_system_score_codex":0.0019800013,"about_ca_system_score_gemma":0.0019897893,"threshold_uncertainty_score":0.5097279},"labels":[],"label_agreement":null},{"id":"W3092385491","doi":"10.1002/sim.8764","title":"The effect of a constraint on the maximum number of controls matched to each treated subject on the performance of full matching on the propensity score when estimating risk differences","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Matching (statistics); Observational study; Calipers; Constraint (computer-aided design); Statistics; Mathematics; Average treatment effect; Econometrics; Medicine","score_opus":0.09042720693450075,"score_gpt":0.3650431871051201,"score_spread":0.2746159801706194,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3092385491","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.10948783,0.002114078,0.8760247,0.003893214,0.00022840145,0.000749913,0.00055745174,0.00045454816,0.0064898725],"genre_scores_gemma":[0.43621588,0.0008343258,0.55847496,0.0014480217,0.00017930211,0.0014613279,0.00044720125,0.00021750352,0.0007215786],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.820243,0.15314928,0.008282544,0.00782053,0.008877868,0.0016267697],"domain_scores_gemma":[0.2438081,0.70485014,0.012190739,0.033601575,0.004547433,0.0010021002],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.20277078,0.0015388187,0.0026365747,0.0015205265,0.0019244519,0.0027396232,0.0031836713,0.0033451398,0.003977214],"category_scores_gemma":[0.6049281,0.0009959048,0.002951044,0.00291398,0.004673641,0.0054106014,0.005352867,0.0047610304,0.0003845816],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.01174413,0.001050148,0.040263142,0.0020286718,0.0026943798,0.001218645,0.0022019648,0.36665905,0.009632327,0.23754828,0.0057771746,0.31918216],"study_design_scores_gemma":[0.0032536834,0.0044032284,0.036068387,0.0017209906,0.0015168771,0.0014473838,0.0005419241,0.6029772,0.016599884,0.3169092,0.014016332,0.00054488133],"about_ca_topic_score_codex":0.0041537285,"about_ca_topic_score_gemma":0.0024490056,"teacher_disagreement_score":0.79722923,"about_ca_system_score_codex":0.001657677,"about_ca_system_score_gemma":0.00427074,"threshold_uncertainty_score":0.98312557},"labels":[],"label_agreement":null},{"id":"W3093256252","doi":"10.1002/sim.8772","title":"Selection models for efficient two‐phase design of family studies","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Canadian Institutes of Health Research; Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada; National Natural Science Foundation of China","keywords":"Simple random sample; Copula (linguistics); Selection (genetic algorithm); Computer science; Sampling (signal processing); Sampling design; Statistics; Sample size determination; Robustness (evolution); Econometrics; Stratified sampling; Mathematics; Machine learning; Biology; Medicine; Population","score_opus":0.3913889358246923,"score_gpt":0.5107849811338429,"score_spread":0.11939604530915054,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3093256252","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0031658707,0.00018983452,0.9952289,0.00018824174,0.00004550366,0.000692511,0.0000737965,0.00008325247,0.00033211138],"genre_scores_gemma":[0.14179713,0.00053480803,0.84474605,0.000343208,0.00013675848,0.010138638,0.0004223504,0.000084835134,0.0017962657],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8511044,0.13888405,0.0020472175,0.00366538,0.0032120568,0.0010869683],"domain_scores_gemma":[0.7740579,0.19882739,0.008102921,0.013574175,0.0042085517,0.0012290718],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.120447226,0.0018929438,0.003765369,0.0018074439,0.0010033576,0.0021113348,0.0040904386,0.0028706521,0.0070154485],"category_scores_gemma":[0.20015912,0.0022160134,0.0020877328,0.0024462151,0.0029942035,0.0027089242,0.00335032,0.004219441,0.0012248292],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0023726984,0.0004380994,0.005607817,0.0006784394,0.00088060397,0.00034040192,0.000860407,0.33893326,0.0010147459,0.50839216,0.003496194,0.13698508],"study_design_scores_gemma":[0.001090846,0.00078899536,0.0010003466,0.00012789729,0.00014334136,0.00008834947,0.00006530118,0.73250026,0.0006008434,0.2589827,0.0045541734,0.00005699496],"about_ca_topic_score_codex":0.0017408958,"about_ca_topic_score_gemma":0.0017107966,"teacher_disagreement_score":0.120447226,"about_ca_system_score_codex":0.0024191146,"about_ca_system_score_gemma":0.0040813386,"threshold_uncertainty_score":0.6369934},"labels":[],"label_agreement":null},{"id":"W3094519324","doi":"10.1002/sim.8786","title":"Spline‐based accelerated failure time model","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre; McGill University","funders":"Fonds de Recherche du Québec - Santé; Natural Sciences and Engineering Research Council of Canada","keywords":"Parametric statistics; Accelerated failure time model; Covariate; Computer science; Spline (mechanical); Proportional hazards model; Parametric model; Semiparametric model; Hazard; Statistics; Mathematics; Engineering; Structural engineering","score_opus":0.4405836122262447,"score_gpt":0.4606245123093229,"score_spread":0.020040900083078184,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3094519324","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02350355,0.001176974,0.9678559,0.0009953945,0.00024747988,0.00016340923,0.0011425306,0.0006133679,0.0043013734],"genre_scores_gemma":[0.8420349,0.0039415914,0.10703881,0.0004839004,0.0006010997,0.0012454344,0.0036972458,0.000258406,0.04069876],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99725217,0.0012032579,0.00015425881,0.00042649964,0.0005695351,0.00039417017],"domain_scores_gemma":[0.99369204,0.0035926981,0.0007523293,0.0004100029,0.001253652,0.00029935013],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0074277716,0.0012633554,0.0023806307,0.0023738523,0.00057815784,0.001412795,0.00402606,0.0023644958,0.009981912],"category_scores_gemma":[0.013361219,0.0007118691,0.0025230064,0.0024587426,0.0012919336,0.0015480593,0.0014343528,0.003370397,0.0022298915],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00034355745,0.00011872297,0.0060408902,0.00031777553,0.00021307173,0.00049860886,0.00027714708,0.70639056,0.0013449786,0.23875237,0.0059083505,0.03979403],"study_design_scores_gemma":[0.000113268405,0.00014263774,0.0011795518,0.0000340463,0.00008482724,0.0001960768,0.000024134757,0.93446946,0.00017106623,0.058899656,0.0046414873,0.0000438134],"about_ca_topic_score_codex":0.009362318,"about_ca_topic_score_gemma":0.0043119253,"teacher_disagreement_score":0.009981912,"about_ca_system_score_codex":0.0010228949,"about_ca_system_score_gemma":0.0023253164,"threshold_uncertainty_score":0.039282262},"labels":[],"label_agreement":null},{"id":"W3094834023","doi":"10.1002/sim.8766","title":"Minimum sample size for external validation of a clinical prediction model with a continuous outcome","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Body Composition Measurement Techniques","field":"Medicine","cited_by":198,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; National Institute for Health and Care Research; NIHR School for Primary Care Research; NIHR Oxford Biomedical Research Centre; Medical Research Council Canada; Wellcome Trust; British Heart Foundation; Cancer Research UK","keywords":"Sample size determination; Variance (accounting); Statistics; Outcome (game theory); Calibration; Sample (material); Range (aeronautics); Cross-validation; Computer science; Mathematics; Econometrics","score_opus":0.1267781237050936,"score_gpt":0.4179327851487913,"score_spread":0.2911546614436977,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3094834023","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.037402533,0.00032961485,0.95297724,0.0021384337,0.00024371025,0.0020471562,0.00041030053,0.00037933586,0.004071654],"genre_scores_gemma":[0.3932779,0.00022293525,0.59620875,0.0012096374,0.00019297384,0.007038581,0.00073711795,0.00018833918,0.0009237418],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.93232334,0.054849602,0.0034648154,0.0036207766,0.005118839,0.0006226348],"domain_scores_gemma":[0.70535177,0.25311267,0.007312169,0.020866951,0.0123051535,0.0010513454],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1428288,0.00090880436,0.0017370789,0.0012755614,0.0009832826,0.001906651,0.00243924,0.002417681,0.0029307518],"category_scores_gemma":[0.37232915,0.0007118922,0.001677879,0.0007819307,0.0019701528,0.0028560641,0.0036599925,0.0033530802,0.0009766378],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008088199,0.0013344403,0.09902495,0.0017924127,0.0011497573,0.0011792026,0.0026593134,0.16094077,0.014480986,0.14467855,0.0164551,0.54821634],"study_design_scores_gemma":[0.0030790393,0.006134079,0.04557443,0.001768026,0.0006428738,0.0012501889,0.0009833634,0.682891,0.028782878,0.20392047,0.024680011,0.00029361766],"about_ca_topic_score_codex":0.0012506525,"about_ca_topic_score_gemma":0.0013545591,"teacher_disagreement_score":0.8571712,"about_ca_system_score_codex":0.0011025948,"about_ca_system_score_gemma":0.003843283,"threshold_uncertainty_score":0.75535995},"labels":[],"label_agreement":null},{"id":"W3101173445","doi":"10.1002/sim.8798","title":"Functional principal component analysis for longitudinal data with informative dropout","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"","keywords":"Functional principal component analysis; Missing data; Principal component analysis; Dropout (neural networks); Functional data analysis; Covariance; Computer science; Orthonormal basis; Statistics; Regression; Mathematics; Artificial intelligence; Machine learning","score_opus":0.2706996628872454,"score_gpt":0.43814257808506935,"score_spread":0.16744291519782395,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3101173445","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0034265192,0.00044585898,0.99532,0.00020651473,0.000036367153,0.00005263607,0.00013100774,0.00022578327,0.00015521838],"genre_scores_gemma":[0.191937,0.002236449,0.79930544,0.00038128157,0.0003362022,0.0012283745,0.0020692926,0.00039231946,0.0021136666],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9929657,0.0049591986,0.00031830822,0.0008046408,0.0007184873,0.00023362847],"domain_scores_gemma":[0.9837712,0.011129745,0.0012849852,0.002293055,0.0012499532,0.00027111234],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016410504,0.0018463993,0.0020249777,0.0027698828,0.0011149633,0.0014282641,0.002347553,0.001771301,0.0023755413],"category_scores_gemma":[0.042443667,0.00081466703,0.0025844479,0.0036390317,0.0018824232,0.002091333,0.0020406593,0.0040857233,0.0010039768],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005072804,0.00023208582,0.012208829,0.0007619332,0.0010426233,0.0006166073,0.00050644384,0.39938465,0.004910542,0.18149531,0.011673473,0.3866602],"study_design_scores_gemma":[0.000029353654,0.00006213208,0.0022989016,0.000050195846,0.000054420343,0.00009928499,0.000036988207,0.91045904,0.0007344838,0.082834356,0.0032967736,0.00004412519],"about_ca_topic_score_codex":0.0055546905,"about_ca_topic_score_gemma":0.0042891465,"teacher_disagreement_score":0.016410504,"about_ca_system_score_codex":0.0010371746,"about_ca_system_score_gemma":0.0036654826,"threshold_uncertainty_score":0.08678806},"labels":[],"label_agreement":null},{"id":"W3105488458","doi":"10.1002/sim.8816","title":"Survival analysis under the Cox proportional hazards model with pooled covariates","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Fonds de Recherche du Québec - Santé; Natural Sciences and Engineering Research Council of Canada","keywords":"Covariate; Statistics; Proportional hazards model; Estimator; Pooling; Accelerated failure time model; Mathematics; Hazard ratio; Confounding; Econometrics; Survival analysis; Random effects model; Medicine; Computer science; Confidence interval; Meta-analysis; Internal medicine","score_opus":0.09750907629122232,"score_gpt":0.4004200705843978,"score_spread":0.3029109942931755,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3105488458","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011367152,0.00046560654,0.9859388,0.0002753972,0.00007156484,0.00023400424,0.00084705354,0.000328451,0.00047191017],"genre_scores_gemma":[0.53701293,0.003361967,0.43977317,0.00056409196,0.00070555456,0.00478651,0.0050400128,0.00024654635,0.008509205],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99161434,0.0054371846,0.00040461018,0.001297509,0.0008093878,0.00043696922],"domain_scores_gemma":[0.99042386,0.006289479,0.001155332,0.0013409368,0.0006568649,0.00013358744],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.021287233,0.0014247444,0.0024800615,0.0023032783,0.00044492187,0.0014730492,0.0024712237,0.001249191,0.0050875666],"category_scores_gemma":[0.03379494,0.0006553499,0.0043862145,0.002933017,0.00079073885,0.0022249604,0.0020484857,0.001750037,0.00092046265],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001716742,0.00024997586,0.038089965,0.0013817872,0.004730174,0.0014142133,0.00083023735,0.49050233,0.002521583,0.17452438,0.007987038,0.27605152],"study_design_scores_gemma":[0.00028592997,0.0010151077,0.008726848,0.00015976945,0.0012500687,0.00048629605,0.00017084407,0.7698161,0.0009128771,0.2085874,0.00847711,0.00011156961],"about_ca_topic_score_codex":0.004701693,"about_ca_topic_score_gemma":0.0025138836,"teacher_disagreement_score":0.021287233,"about_ca_system_score_codex":0.0010504948,"about_ca_system_score_gemma":0.0031530291,"threshold_uncertainty_score":0.11257899},"labels":[],"label_agreement":null},{"id":"W3105768632","doi":"10.1002/sim.8857","title":"Conflating marginal and conditional treatment effects: Comments on “Assessing the performance of population adjustment methods for anchored indirect comparisons: A simulation study”","year":2021,"lang":"en","type":"letter","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; Public Health Ontario; University of Toronto","funders":"Engineering and Physical Sciences Research Council; Canadian Institutes of Health Research","keywords":"Inference; Marginal structural model; Econometrics; Marginal model; Population; Treatment effect; Causal inference; Statistics; Statistical inference; Computer science; Medicine; Mathematics; Artificial intelligence; Regression analysis","score_opus":0.20791701651492495,"score_gpt":0.5391972190249351,"score_spread":0.33128020251001017,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3105768632","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00012291552,0.000663771,0.00049313187,0.9865866,0.011486241,0.000013976152,0.00006250168,0.000024071804,0.00054675137],"genre_scores_gemma":[0.0010370467,0.00029285924,0.0006971334,0.9857789,0.011396575,0.000057934474,0.000013593766,0.000030325764,0.000695595],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.93768835,0.035859052,0.007845438,0.0046985685,0.011826186,0.0020822687],"domain_scores_gemma":[0.7575847,0.20616914,0.006066254,0.003351086,0.023811374,0.0030174446],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0668757,0.0014682127,0.0022997789,0.0011741358,0.006036915,0.0045557246,0.007324765,0.06753303,0.004877028],"category_scores_gemma":[0.32914245,0.0014782111,0.0034227334,0.0018435067,0.009829109,0.0075120064,0.0033686915,0.071105,0.0052292114],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006580007,0.000017929755,0.00027705324,0.000103960294,0.000017673745,0.00033021567,0.0003708015,0.00013142878,0.00008125744,0.006869944,0.986086,0.005648015],"study_design_scores_gemma":[0.00025764084,0.00012490283,0.0013061704,0.0015938642,0.00011794174,0.0013515439,0.00097467983,0.0018592135,0.00081802625,0.045275852,0.9460632,0.00025693123],"about_ca_topic_score_codex":0.020905731,"about_ca_topic_score_gemma":0.028348474,"teacher_disagreement_score":0.9331243,"about_ca_system_score_codex":0.008353424,"about_ca_system_score_gemma":0.013725857,"threshold_uncertainty_score":0.35367674},"labels":[],"label_agreement":null},{"id":"W3112861306","doi":"10.1002/sim.8847","title":"An overview and critique of the use of cumulative sum methods with surgical learning curve data","year":2020,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Healthcare Operations and Scheduling Optimization","field":"Health Professions","cited_by":67,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; Université de Montréal; Hôpital Maisonneuve-Rosemont","funders":"","keywords":"CUSUM; Computer science; Learning curve; Focus (optics); Statistics; Econometrics; Machine learning; Mathematics","score_opus":0.5748082015241052,"score_gpt":0.6488662788855685,"score_spread":0.0740580773614633,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3112861306","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00009740726,0.9871389,0.0054646465,0.0051382575,0.0007930782,0.000026120775,0.00006776984,0.000034696204,0.0012390725],"genre_scores_gemma":[0.0017916702,0.9852304,0.0071346094,0.0034547485,0.0015697555,0.00010617977,0.00009199396,0.00006455618,0.0005560356],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98165673,0.008650057,0.0027286538,0.0012504446,0.0054999497,0.00021426979],"domain_scores_gemma":[0.8627831,0.1169225,0.004046548,0.0021350428,0.013543339,0.0005694647],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.028581323,0.0013508378,0.0033306628,0.00985884,0.0005756606,0.0036123625,0.0040991507,0.0033511736,0.00248584],"category_scores_gemma":[0.08420176,0.0014064909,0.0025092487,0.015037565,0.003409537,0.004155938,0.0016635553,0.0074728066,0.002639158],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000075515905,0.000055702112,0.000679568,0.038635243,0.0005483349,0.00018180936,0.00049414404,0.0027786074,0.00028302844,0.044549875,0.08509558,0.82662266],"study_design_scores_gemma":[0.000028192188,0.00015555129,0.0025941932,0.039550994,0.00037602312,0.0011858566,0.00027033145,0.0014606874,0.0005143508,0.03310285,0.92062104,0.00013988666],"about_ca_topic_score_codex":0.0072049126,"about_ca_topic_score_gemma":0.005610209,"teacher_disagreement_score":0.9714187,"about_ca_system_score_codex":0.0033288035,"about_ca_system_score_gemma":0.006790942,"threshold_uncertainty_score":0.15115428},"labels":[],"label_agreement":null},{"id":"W3112901974","doi":"10.1002/sim.8848","title":"Mean comparisons and power calculations to ensure reproducibility in preclinical drug discovery","year":2020,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Animal testing and alternatives","field":"Veterinary","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"AstraZeneca (Canada)","funders":"","keywords":"Reproducibility; Sample size determination; Statistical power; Statistics; Computer science; Power analysis; Statistical hypothesis testing; Power (physics); Sample (material); Multiple comparisons problem; Data mining; False discovery rate; Reliability engineering; Mathematics; Algorithm; Chemistry","score_opus":0.25756195579739644,"score_gpt":0.48454320906766907,"score_spread":0.22698125327027263,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3112901974","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"reproducibility","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"reproducibility","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0015945983,0.0003854301,0.9944898,0.00019304128,0.00017217392,0.0011206333,0.00012140849,0.00037571148,0.0015472809],"genre_scores_gemma":[0.033814803,0.0004965657,0.9569361,0.00022946988,0.00023274032,0.0066717956,0.00016683004,0.00044830522,0.0010034257],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8948295,0.06980361,0.007162875,0.0061235717,0.021143092,0.0009372679],"domain_scores_gemma":[0.8192377,0.12929317,0.013448384,0.022178099,0.015201661,0.00064104586],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.083590634,0.0022046063,0.0019036586,0.0041937623,0.0012421688,0.002619777,0.0036987418,0.0026351144,0.007747314],"category_scores_gemma":[0.25198567,0.00083287753,0.0025212693,0.0038964418,0.004323175,0.0027742526,0.0030117317,0.0051329103,0.002285831],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014770273,0.00045441464,0.004481021,0.0037705072,0.0007614564,0.00056853966,0.002304376,0.019550243,0.013470681,0.22516969,0.019835293,0.7081569],"study_design_scores_gemma":[0.0010653319,0.0077120736,0.021333294,0.0030074827,0.0012067797,0.0032338048,0.0012845742,0.144926,0.07292785,0.572141,0.17041697,0.00074491365],"about_ca_topic_score_codex":0.0003657366,"about_ca_topic_score_gemma":0.000497168,"teacher_disagreement_score":0.9164094,"about_ca_system_score_codex":0.001622576,"about_ca_system_score_gemma":0.0033069784,"threshold_uncertainty_score":0.44207484},"labels":[],"label_agreement":null},{"id":"W3120069145","doi":"10.1002/sim.10199","title":"Estimating causes of maternal death in data‐sparse contexts","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Global Maternal and Child Health","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"World Health Organization","keywords":"Multinomial distribution; Data quality; Computer science; Missing data; Bayesian probability; Population; Quality (philosophy); Resource allocation; Econometrics; Environmental health; Medicine; Machine learning; Artificial intelligence; Business; Economics","score_opus":0.05196058672600602,"score_gpt":0.4003532393911025,"score_spread":0.3483926526650965,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3120069145","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.68223715,0.0029319937,0.29279882,0.010367812,0.00008248458,0.0003071616,0.006085835,0.00019273545,0.0049960464],"genre_scores_gemma":[0.9498496,0.0008672772,0.04580552,0.00037190822,0.00005433748,0.00015571847,0.0022325339,0.00002110087,0.0006420659],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99320567,0.0053467527,0.00021413625,0.00065450114,0.0002924214,0.00028648585],"domain_scores_gemma":[0.9621393,0.031605657,0.0031872976,0.001654913,0.0010710456,0.00034176942],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014896284,0.0005370344,0.0010501172,0.0016599013,0.0008084414,0.0016409063,0.0014107581,0.0013421937,0.0012417382],"category_scores_gemma":[0.055498663,0.0006674617,0.0011153261,0.0022081765,0.0012517751,0.0026613341,0.0029909117,0.001577202,0.00016150669],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00025010537,0.00015426269,0.3206246,0.0004128654,0.0006141389,0.00091269537,0.0016624724,0.5357558,0.00040505556,0.08102639,0.004779437,0.053402152],"study_design_scores_gemma":[0.000104745595,0.000086972825,0.0625105,0.0002642799,0.00017880877,0.0003520868,0.001919728,0.7961987,0.0003232573,0.13272633,0.0052531296,0.000081403115],"about_ca_topic_score_codex":0.04852635,"about_ca_topic_score_gemma":0.04881776,"teacher_disagreement_score":0.04852635,"about_ca_system_score_codex":0.001421131,"about_ca_system_score_gemma":0.0016060033,"threshold_uncertainty_score":0.09648776},"labels":[],"label_agreement":null},{"id":"W3124976303","doi":"10.1002/sim.8863","title":"Spatial modeling of individual‐level infectious disease transmission: Tuberculosis data in Manitoba, Canada","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"COVID-19 epidemiological studies","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Calgary; University of Manitoba","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Covariate; Infectious disease (medical specialty); Tuberculosis; Infectivity; Statistics; Demography; Demographics; Maximization; Transmission (telecommunications); Population; Estimation; Econometrics; Computer science; Medicine; Disease; Environmental health; Geography; Immunology; Mathematics; Internal medicine","score_opus":0.2830997507269963,"score_gpt":0.4096805293144711,"score_spread":0.1265807785874748,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3124976303","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9780568,0.0013550387,0.008289097,0.0017380717,0.00004407244,0.00011832319,0.007408707,0.00014243498,0.0028476063],"genre_scores_gemma":[0.9886813,0.00068154035,0.004569227,0.00007319999,0.000012400786,0.000041288637,0.0033370948,0.000020194757,0.0025838187],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.999501,0.00014735293,0.000027360658,0.00009747694,0.00008190193,0.00014484972],"domain_scores_gemma":[0.9979348,0.0008242864,0.00020867286,0.00013234626,0.0007373858,0.00016257241],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0012522212,0.0005284794,0.0005907653,0.0014023975,0.0014182318,0.00095796864,0.00228904,0.0007970527,0.001695558],"category_scores_gemma":[0.0045985673,0.00036734785,0.0006164349,0.0037166553,0.00090453593,0.00049408816,0.0008265935,0.00083528645,0.00020593396],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019882624,0.00018451642,0.23319376,0.00015333931,0.00020791491,0.00068285444,0.0006219624,0.7280753,0.0005889066,0.010180466,0.0074274046,0.018484777],"study_design_scores_gemma":[0.000041762007,0.000029093224,0.06973773,0.000040965464,0.000051813757,0.00007281714,0.0007363401,0.92444694,0.00021154137,0.0015049278,0.003076188,0.000049804097],"about_ca_topic_score_codex":0.9897516,"about_ca_topic_score_gemma":0.9854336,"teacher_disagreement_score":0.01794082,"about_ca_system_score_codex":0.01794082,"about_ca_system_score_gemma":0.0157076,"threshold_uncertainty_score":0.13017035},"labels":[],"label_agreement":null},{"id":"W3125785889","doi":"10.1002/sim.1577","title":"The roles of birth inputs and outputs in predicting health, behaviour and test scores in early childhood","year":2003,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Birth, Development, and Health","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Early childhood; Developmental psychology; Birth weight; Predictive value; Psychology; Test (biology); Structural equation modeling; Child development; Medicine; Pregnancy; Statistics; Mathematics","score_opus":0.01593232426620633,"score_gpt":0.3052187598674422,"score_spread":0.28928643560123585,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3125785889","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.93934613,0.0015930983,0.051520176,0.0011868676,0.000068271904,0.00005139598,0.00283864,0.00025941935,0.0031360064],"genre_scores_gemma":[0.99243104,0.00070796994,0.0048051253,0.00004065799,0.000027445843,0.00002637652,0.0009088785,0.000033282715,0.0010191638],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9974716,0.0017247946,0.00012132588,0.00024823757,0.00026443484,0.00016968223],"domain_scores_gemma":[0.9680434,0.026452428,0.002972968,0.0012655101,0.00069722044,0.000568492],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0060248496,0.0014295528,0.000644311,0.0011634812,0.00025289415,0.002295909,0.00072141323,0.00068013516,0.0022198914],"category_scores_gemma":[0.033839095,0.0005799043,0.00092472247,0.0016225846,0.000820559,0.0014572804,0.0015826005,0.0014936599,0.0003508546],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00037170452,0.00007054388,0.9215702,0.00009012939,0.0005350573,0.00024360082,0.00024658022,0.047224652,0.00047186876,0.0052979463,0.00028815947,0.023589572],"study_design_scores_gemma":[0.000040271658,0.0004255724,0.7287098,0.00018609957,0.0006791916,0.00024050023,0.00035925518,0.2334995,0.0023100497,0.03146966,0.0019983675,0.00008164789],"about_ca_topic_score_codex":0.012539976,"about_ca_topic_score_gemma":0.012292116,"teacher_disagreement_score":0.012539976,"about_ca_system_score_codex":0.00078831887,"about_ca_system_score_gemma":0.0009529941,"threshold_uncertainty_score":0.031862855},"labels":[],"label_agreement":null},{"id":"W3125937508","doi":"10.1002/sim.8879","title":"Multiple imputation strategies for a bounded outcome variable in a competing risks analysis","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"University of Bristol; Medical Research Council; Royal Society; NHS Blood and Transplant; Medical Research Council Canada; Wellcome Trust","keywords":"Imputation (statistics); Statistics; Missing data; Cumulative distribution function; Econometrics; Cumulative incidence; Mathematics; Regression analysis; Confidence interval; Computer science; Probability density function","score_opus":0.14089669959972412,"score_gpt":0.4712305057681186,"score_spread":0.3303338061683945,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3125937508","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011359547,0.00032741931,0.9971614,0.0004012991,0.000057092348,0.0003032127,0.00012042637,0.00015062529,0.0003426044],"genre_scores_gemma":[0.03690368,0.0004280283,0.9578379,0.00050809263,0.00011648299,0.0027131746,0.00040991278,0.00019308779,0.0008896866],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.87961257,0.105937086,0.0051313504,0.0042468724,0.0043692146,0.0007029042],"domain_scores_gemma":[0.7772918,0.19092794,0.008292589,0.015469544,0.0070174984,0.0010006555],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1473591,0.0019231079,0.0038078243,0.0032647,0.0012592071,0.0034487792,0.008249113,0.002914939,0.010096566],"category_scores_gemma":[0.24219783,0.0014843125,0.005410573,0.0042516124,0.0014348569,0.0033448269,0.003895263,0.0061120973,0.0019803452],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011197388,0.0004075268,0.01686902,0.001939188,0.0053031845,0.0011220264,0.0024932767,0.13403724,0.0010453999,0.41536352,0.018321373,0.40197843],"study_design_scores_gemma":[0.0005742048,0.0005179828,0.0023848086,0.00084755087,0.0009824722,0.00070468866,0.00024367281,0.49818155,0.0014642235,0.47551095,0.018380895,0.00020701831],"about_ca_topic_score_codex":0.0023613821,"about_ca_topic_score_gemma":0.002735399,"teacher_disagreement_score":0.1473591,"about_ca_system_score_codex":0.0013823997,"about_ca_system_score_gemma":0.0037211338,"threshold_uncertainty_score":0.77931875},"labels":[],"label_agreement":null},{"id":"W3134342065","doi":"10.1002/sim.8922","title":"uTPI: A utility‐based toxicity probability interval design for phase I/II dose‐finding trials","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"National Cancer Institute; National Institutes of Health","keywords":"Maximum tolerated dose; Confidence interval; Clinical trial; Toxicity; Parametric statistics; Computer science; Medicine; Clinical study design; Optimal design; Interval (graph theory); Statistics; Mathematics; Machine learning; Internal medicine","score_opus":0.8161295342178478,"score_gpt":0.6464013638722923,"score_spread":0.16972817034555554,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3134342065","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008210474,0.00032990437,0.9882675,0.00016453545,0.00006137216,0.0017236468,0.00011990626,0.00025903285,0.0008636096],"genre_scores_gemma":[0.34428608,0.0005392196,0.642081,0.0005192032,0.00009815226,0.010531005,0.00034683908,0.00010106515,0.0014974542],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.96838766,0.027183756,0.00084524334,0.0015035262,0.0016305891,0.00044918517],"domain_scores_gemma":[0.9698924,0.0231372,0.0027387948,0.0018132185,0.0015938551,0.00082451367],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.040225882,0.001850285,0.00210598,0.0014548558,0.00044111564,0.001596666,0.0026818349,0.0018291579,0.004297482],"category_scores_gemma":[0.06023899,0.0010189455,0.0014750852,0.0012865271,0.0017787971,0.0015892439,0.0022621697,0.0033595946,0.00053969846],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.010018898,0.0007536415,0.0044114846,0.0013280093,0.00058320654,0.00042776045,0.00039473097,0.54344535,0.007575958,0.17888488,0.003300232,0.24887592],"study_design_scores_gemma":[0.0022843978,0.0039670845,0.0009573036,0.00011478751,0.00023454019,0.00016957716,0.000030988354,0.9128369,0.0033204257,0.072121225,0.00387901,0.000083656196],"about_ca_topic_score_codex":0.00047864116,"about_ca_topic_score_gemma":0.00037745165,"teacher_disagreement_score":0.040225882,"about_ca_system_score_codex":0.0012184475,"about_ca_system_score_gemma":0.0025059353,"threshold_uncertainty_score":0.21273732},"labels":[],"label_agreement":null},{"id":"W3135053972","doi":"10.1002/sim.8906","title":"The impact of adjusting for pure predictors of exposure, mediator, and outcome on the variance of natural direct and indirect effect estimators","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université du Québec à Montréal; Université Laval","funders":"Fonds de Recherche du Québec - Santé; Natural Sciences and Engineering Research Council of Canada","keywords":"Statistics; Estimator; Confounding; Outcome (game theory); Mediation; Inverse probability weighting; Variance (accounting); Weighting; Econometrics; Mathematics; Linear regression; Regression analysis; Medicine; Economics","score_opus":0.05564411871493614,"score_gpt":0.42065357746469295,"score_spread":0.3650094587497568,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3135053972","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.047560785,0.0029284281,0.94448733,0.0015977577,0.00017978375,0.00025150465,0.00020488075,0.00026443414,0.0025251273],"genre_scores_gemma":[0.6266172,0.001898673,0.36835095,0.00068975455,0.00017879298,0.0007234579,0.00034597743,0.0001982825,0.000996863],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8737059,0.10996435,0.0032653606,0.0054358225,0.0064613274,0.0011672875],"domain_scores_gemma":[0.31477886,0.65013283,0.00523968,0.02536105,0.00412829,0.0003593558],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.16134569,0.0013349017,0.0019681735,0.0015368353,0.0008950031,0.002366339,0.002136027,0.002029081,0.0012814074],"category_scores_gemma":[0.44734928,0.0008167618,0.002835346,0.002295651,0.0038417333,0.003252536,0.003345294,0.0031419802,0.00021600172],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014441438,0.00030672038,0.1422686,0.001527808,0.0060389023,0.0009897175,0.0022481538,0.17980126,0.0040649567,0.31097722,0.0023772505,0.3479553],"study_design_scores_gemma":[0.0003675032,0.0013273216,0.05995217,0.0010298614,0.0030184588,0.0012470613,0.0007805545,0.5647208,0.0077447644,0.3515558,0.007952079,0.00030364285],"about_ca_topic_score_codex":0.0075937025,"about_ca_topic_score_gemma":0.006088492,"teacher_disagreement_score":0.16134569,"about_ca_system_score_codex":0.0012407245,"about_ca_system_score_gemma":0.0035852015,"threshold_uncertainty_score":0.85328776},"labels":[],"label_agreement":null},{"id":"W3141196614","doi":"10.1002/sim.8966","title":"Multiparameter one‐sided tests for nonlinear mixed effects models with censored responses","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Nonlinear system; Statistical hypothesis testing; Computer science; Econometrics; Mathematics; Statistics; Physics","score_opus":0.10619714876143097,"score_gpt":0.42130522307059237,"score_spread":0.31510807430916143,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3141196614","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.036108635,0.00074657565,0.959401,0.0006906942,0.00010113646,0.00044541716,0.00074298796,0.00049567455,0.0012679626],"genre_scores_gemma":[0.5464087,0.00047257406,0.4469264,0.0006059246,0.0002095777,0.0025818862,0.0014451543,0.00015877538,0.0011910596],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.89299023,0.0919538,0.002959647,0.006682328,0.0045755506,0.0008385054],"domain_scores_gemma":[0.46850267,0.5015132,0.011384711,0.014904425,0.0026145747,0.0010805016],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.075111546,0.0017264271,0.0035953955,0.0036253678,0.0014372173,0.0030820067,0.005045234,0.0031964642,0.008827999],"category_scores_gemma":[0.3037856,0.0008288737,0.0037931958,0.004432917,0.0049627563,0.0055370205,0.0034719687,0.0047944593,0.000787081],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0032687525,0.00058219954,0.042800155,0.001864513,0.0052283807,0.001426651,0.0010020718,0.28001562,0.0024688102,0.3603823,0.0052364343,0.2957241],"study_design_scores_gemma":[0.0005201539,0.00096886436,0.00919352,0.00025161638,0.0003579469,0.00038788328,0.00023691064,0.7081185,0.0018536431,0.2738297,0.004138643,0.00014261667],"about_ca_topic_score_codex":0.0014230282,"about_ca_topic_score_gemma":0.0013719401,"teacher_disagreement_score":0.075111546,"about_ca_system_score_codex":0.0018334987,"about_ca_system_score_gemma":0.002971267,"threshold_uncertainty_score":0.3972326},"labels":[],"label_agreement":null},{"id":"W3152465836","doi":"10.1002/sim.8949","title":"Optimizing subgroup selection in two‐stage adaptive enrichment and umbrella designs","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Horizon 2020 Framework Programme; H2020 Marie Skłodowska-Curie Actions; Innovative Medicines Initiative; Medical Research Council Canada; National Institute for Health and Care Research; Springworks Therapeutics; Medical Research Council; European Federation of Pharmaceutical Industries and Associations","keywords":"Type I and type II errors; Computer science; Selection (genetic algorithm); Word error rate; Population; Subgroup analysis; Adaptive design; Interim analysis; Bayesian probability; Multiple comparisons problem; Clinical trial; Statistics; Mathematical optimization; Medicine; Mathematics; Machine learning; Artificial intelligence; Internal medicine; Confidence interval","score_opus":0.4502326642966361,"score_gpt":0.553600628135857,"score_spread":0.10336796383922087,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3152465836","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.028595531,0.0004986592,0.96714157,0.00056949817,0.0000965176,0.0017549961,0.00010339414,0.0002148408,0.0010250193],"genre_scores_gemma":[0.27733245,0.00029149707,0.71162665,0.0008463073,0.00010772635,0.007949018,0.00020346593,0.000066817214,0.001576161],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8832103,0.10815408,0.0016263827,0.0034476817,0.0025924689,0.00096903136],"domain_scores_gemma":[0.9016349,0.07825017,0.0063275485,0.009271881,0.0030211064,0.0014943642],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.098996565,0.00231637,0.004372513,0.0014002082,0.0007581822,0.001831518,0.003224246,0.003492286,0.002882243],"category_scores_gemma":[0.1374046,0.0015228629,0.0026631055,0.0012527329,0.0030735268,0.0024922595,0.0036653485,0.0027986825,0.000776768],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.01382952,0.0015187147,0.0074307327,0.0012303967,0.0017962534,0.00030043034,0.0009704288,0.42778823,0.00570844,0.26137477,0.0030206305,0.27503148],"study_design_scores_gemma":[0.0038389592,0.00506588,0.0020759709,0.00028436084,0.000567199,0.00015305051,0.000079601494,0.74220407,0.0043789684,0.23719795,0.0039688786,0.00018513588],"about_ca_topic_score_codex":0.0004715008,"about_ca_topic_score_gemma":0.00045043946,"teacher_disagreement_score":0.9010034,"about_ca_system_score_codex":0.0013532443,"about_ca_system_score_gemma":0.003499291,"threshold_uncertainty_score":0.52355015},"labels":[],"label_agreement":null},{"id":"W3153092758","doi":"10.1002/sim.8991","title":"Computing the polytomous discrimination index","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian VIGOUR Centre; University of Alberta","funders":"Canadian Institutes of Health Research","keywords":"Polytomous Rasch model; Categorical variable; Outcome (game theory); Ranking (information retrieval); Pairwise comparison; Computer science; Statistics; Index (typography); Econometrics; Mathematics; Data mining; Artificial intelligence; Item response theory; Psychometrics","score_opus":0.4309870051715288,"score_gpt":0.5813675000711929,"score_spread":0.15038049489966404,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3153092758","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.070365116,0.00052988675,0.9223504,0.0006891775,0.000098476245,0.0002167241,0.0011015194,0.0017180386,0.0029307858],"genre_scores_gemma":[0.4803889,0.0003025839,0.5140521,0.0002839995,0.00012345704,0.0006500353,0.002653474,0.0004533297,0.001092113],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9942783,0.0024663033,0.00048594153,0.0010923472,0.0014272684,0.0002498066],"domain_scores_gemma":[0.9474782,0.04339297,0.0021530758,0.003633012,0.0026723188,0.0006703258],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011049529,0.0008200251,0.0017024048,0.004116076,0.0007057196,0.0023576692,0.0015783753,0.0011377129,0.0037148478],"category_scores_gemma":[0.085504495,0.000433385,0.0010290615,0.0033782255,0.0014154022,0.0024567554,0.0028349645,0.0024696493,0.0009841345],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006549663,0.00026533345,0.07637327,0.00049345597,0.00056120707,0.00051608897,0.00050478324,0.33070487,0.0025924598,0.084974,0.017961841,0.48439768],"study_design_scores_gemma":[0.0000494596,0.00008844751,0.005398882,0.000037858685,0.0000348865,0.00022679125,0.000080653044,0.85373664,0.0013187655,0.13596782,0.0030162483,0.000043563276],"about_ca_topic_score_codex":0.002035772,"about_ca_topic_score_gemma":0.0017713206,"teacher_disagreement_score":0.011049529,"about_ca_system_score_codex":0.0013253279,"about_ca_system_score_gemma":0.0024652726,"threshold_uncertainty_score":0.058436155},"labels":[],"label_agreement":null},{"id":"W3156818063","doi":"10.1002/sim.9002","title":"Semiparametric recurrent event vs time‐to‐first‐event analyses in randomized trials: Estimands and model misspecification","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Canadian Institutes of Health Research; Natural Sciences and Engineering Research Council of Canada; National Natural Science Foundation of China","keywords":"Censoring (clinical trials); Estimator; Econometrics; Proportional hazards model; Event (particle physics); Semiparametric regression; Semiparametric model; Sample size determination; Statistics; Computer science; Mathematics","score_opus":0.30399888459233215,"score_gpt":0.5326979569422585,"score_spread":0.22869907234992637,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3156818063","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0066505317,0.001382029,0.98955524,0.0011469716,0.00008303695,0.00024343806,0.00009801524,0.00015407086,0.0006867622],"genre_scores_gemma":[0.46472093,0.0029519603,0.52384996,0.0019336051,0.0006534122,0.0033044408,0.0003798239,0.00026652173,0.0019392995],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7655823,0.21593732,0.0051072505,0.005548745,0.006843146,0.0009812468],"domain_scores_gemma":[0.26045296,0.6910486,0.015544456,0.02843304,0.0038134763,0.00070743554],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.23377198,0.0016332826,0.0038620827,0.002451136,0.00054271467,0.003855457,0.0041952003,0.0038128763,0.004038772],"category_scores_gemma":[0.6072423,0.0010280958,0.0041527404,0.0024200617,0.004877081,0.0074556805,0.004515148,0.0060251546,0.0004511386],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011561698,0.00018331701,0.0087849,0.0023749436,0.0024919622,0.0005032517,0.0010387335,0.06938139,0.0012947043,0.7602597,0.0016128843,0.150918],"study_design_scores_gemma":[0.00035595038,0.00071016786,0.0026850181,0.00045640863,0.0008447925,0.0003915671,0.00013105378,0.25809652,0.0016534432,0.7315225,0.003073174,0.00007933757],"about_ca_topic_score_codex":0.0007492693,"about_ca_topic_score_gemma":0.00057882775,"teacher_disagreement_score":0.23377198,"about_ca_system_score_codex":0.001661321,"about_ca_system_score_gemma":0.0029369493,"threshold_uncertainty_score":0.94489557},"labels":[],"label_agreement":null},{"id":"W3157155409","doi":"10.1002/sim.8997","title":"Using generalized linear models to implement g‐estimation for survival data with time‐varying confounding","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"NIHR Cambridge Biomedical Research Centre; Bijzonder Onderzoeksfonds UGent; Medical Research Council Canada; Department of Health and Social Care; Medical Research Council; National Institute for Health and Care Research; UK Research and Innovation","keywords":"Confounding; Computer science; Estimation; Weighting; Statistics; Marginal structural model; Software; Proportional hazards model; Econometrics; Observational study; Data mining; Mathematics; Medicine","score_opus":0.4354915215451955,"score_gpt":0.5352801309584877,"score_spread":0.0997886094132922,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3157155409","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0006032475,0.000088699504,0.9972356,0.00024887495,0.000052562344,0.000094940224,0.00022496397,0.0011419961,0.00030910314],"genre_scores_gemma":[0.021111075,0.0003984038,0.97377723,0.0003823502,0.00008373134,0.0015351304,0.0007249341,0.00087669154,0.0011104478],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96523327,0.029591205,0.0012932854,0.0017289526,0.0017881056,0.00036514402],"domain_scores_gemma":[0.90067935,0.08670541,0.0032642349,0.0070216316,0.002081049,0.00024829328],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04401878,0.0024652,0.0020438586,0.0036176648,0.0008171526,0.0026654152,0.003941889,0.002972892,0.017389731],"category_scores_gemma":[0.16469704,0.0018644973,0.0052795983,0.005416198,0.0020691336,0.0030004208,0.0045107156,0.0065843556,0.005658387],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00030351413,0.0002054369,0.010413267,0.0019239059,0.002623977,0.00086622464,0.0019019521,0.19933102,0.0015193722,0.36267677,0.030997261,0.38723716],"study_design_scores_gemma":[0.00022975916,0.00019600122,0.0024412726,0.0004578301,0.0003436176,0.00032152305,0.00023503933,0.42344576,0.0016394024,0.5324415,0.038098413,0.00014982687],"about_ca_topic_score_codex":0.010417498,"about_ca_topic_score_gemma":0.009737975,"teacher_disagreement_score":0.04401878,"about_ca_system_score_codex":0.0016463294,"about_ca_system_score_gemma":0.0040669944,"threshold_uncertainty_score":0.23279631},"labels":[],"label_agreement":null},{"id":"W3157346880","doi":"10.1002/sim.8944","title":"Comment on Ellenberg and Morris: The role of statisticians in vaccine surveillance","year":2021,"lang":"en","type":"letter","venue":"Statistics in Medicine","topic":"Vaccine Coverage and Hesitancy","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University","funders":"","keywords":"Pandemic; Vaccine trial; Vaccination; Medicine; Observational study; Scope (computer science); Vaccine efficacy; Coronavirus disease 2019 (COVID-19); Virology; Computer science; Disease","score_opus":0.01695799143974529,"score_gpt":0.30909520132909285,"score_spread":0.29213720988934755,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3157346880","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000085145286,0.0010610903,0.000090702226,0.97932315,0.01899571,0.000011952677,0.000060446986,0.00003624864,0.00033560817],"genre_scores_gemma":[0.0006582477,0.00055145135,0.00015039924,0.97564745,0.022334935,0.000028112925,0.000014223357,0.000025672432,0.0005894402],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.97551394,0.008760961,0.0035484526,0.0043572,0.00634264,0.0014768329],"domain_scores_gemma":[0.84211236,0.118308775,0.005520661,0.00411241,0.021369299,0.0085765375],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.038287494,0.0022549145,0.0040706885,0.0021826064,0.0058923424,0.0069098077,0.010718368,0.08978046,0.0075689303],"category_scores_gemma":[0.17926908,0.0025373972,0.0037880386,0.0027116525,0.015272491,0.016743848,0.0034756185,0.12571727,0.009929175],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007828463,0.000018978879,0.00026527737,0.00009822933,0.000017855926,0.00021892306,0.00017664208,0.000051342668,0.00006868233,0.0022930698,0.99395025,0.0027624408],"study_design_scores_gemma":[0.0003035234,0.00013678054,0.0017280015,0.001207631,0.000083003295,0.0011308257,0.0011951044,0.00080917694,0.0006449712,0.01660093,0.97582304,0.00033700594],"about_ca_topic_score_codex":0.02436999,"about_ca_topic_score_gemma":0.019102018,"teacher_disagreement_score":0.08978046,"about_ca_system_score_codex":0.007278706,"about_ca_system_score_gemma":0.009709004,"threshold_uncertainty_score":0.20248604},"labels":[],"label_agreement":null},{"id":"W3158038957","doi":"10.1002/sim.8958","title":"Contamination: How much can an individually randomized trial tolerate?","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Patient Safety and Medication Errors","field":"Health Professions","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital","funders":"Collaboration for Leadership in Applied Health Research and Care - Greater Manchester; National Institute for Health and Care Research","keywords":"Randomized controlled trial; Contamination; Statistics; Computer science; Medicine; Mathematics; Biology; Internal medicine; Ecology","score_opus":0.08878237919998472,"score_gpt":0.4545213410230711,"score_spread":0.3657389618230864,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3158038957","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009298982,0.089283384,0.5008259,0.3531377,0.017045356,0.0052587055,0.0010125637,0.0021987034,0.021938695],"genre_scores_gemma":[0.2965584,0.03599218,0.37708306,0.22861388,0.025514578,0.028089844,0.00088342215,0.0018039952,0.005460655],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.3302398,0.54351807,0.055998713,0.016406547,0.051269423,0.0025674878],"domain_scores_gemma":[0.1375801,0.71723694,0.05263735,0.06454948,0.025043063,0.0029530951],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5099261,0.0024107033,0.009563454,0.002779052,0.0029541026,0.01338956,0.005797962,0.016201153,0.0063025155],"category_scores_gemma":[0.77701604,0.0022885543,0.0069500357,0.003658892,0.02081597,0.022109726,0.009323992,0.01601225,0.0029344314],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0057646437,0.0003625055,0.01387493,0.020915056,0.0063050576,0.00092828006,0.007036872,0.00982109,0.0013200656,0.21924075,0.0940939,0.6203369],"study_design_scores_gemma":[0.0024874138,0.0033447912,0.007371809,0.03139013,0.0039171516,0.002360786,0.0018685526,0.014896527,0.0030960885,0.72562855,0.20287757,0.00076060696],"about_ca_topic_score_codex":0.0015662473,"about_ca_topic_score_gemma":0.00085969607,"teacher_disagreement_score":0.49007392,"about_ca_system_score_codex":0.008958773,"about_ca_system_score_gemma":0.013317419,"threshold_uncertainty_score":0.6043484},"labels":[],"label_agreement":null},{"id":"W3158999510","doi":"10.1002/sim.8918","title":"Confidence interval estimation for treatment effects in cluster randomization trials based on ranks","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Confidence interval; Statistics; Estimator; Restricted randomization; Coverage probability; Randomization; Cluster randomised controlled trial; Mathematics; Interval estimation; Cluster (spacecraft); Computer science; Randomized controlled trial; Medicine","score_opus":0.09668797906989537,"score_gpt":0.46362505929020326,"score_spread":0.3669370802203079,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3158999510","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0031282762,0.0023496589,0.99100643,0.0003243672,0.00017036138,0.001018393,0.000177062,0.0004961396,0.0013292689],"genre_scores_gemma":[0.12714928,0.0020651163,0.8612091,0.00054221804,0.00024702196,0.0073004556,0.0005911031,0.00035240463,0.0005433149],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.61765444,0.3309839,0.014474756,0.011547473,0.023838202,0.0015013051],"domain_scores_gemma":[0.27160916,0.66083133,0.0291577,0.023391012,0.014036732,0.0009740868],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2832662,0.0022307723,0.006153129,0.009404305,0.0011328042,0.005388231,0.006482997,0.004413148,0.005726664],"category_scores_gemma":[0.71740794,0.0015234287,0.0062238486,0.00742477,0.005176414,0.0052828044,0.0045972993,0.008438302,0.0010068988],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0047789807,0.00034670366,0.013134026,0.0086527215,0.006650596,0.00036073243,0.00266379,0.094253026,0.0014033451,0.40315956,0.010590365,0.45400625],"study_design_scores_gemma":[0.0028267228,0.002671106,0.009572246,0.005083777,0.0040388247,0.0007947035,0.00046837676,0.4796988,0.006919869,0.4645078,0.02291457,0.0005031456],"about_ca_topic_score_codex":0.001984963,"about_ca_topic_score_gemma":0.0009595416,"teacher_disagreement_score":0.2832662,"about_ca_system_score_codex":0.002843725,"about_ca_system_score_gemma":0.004878449,"threshold_uncertainty_score":0.8838604},"labels":[],"label_agreement":null},{"id":"W3159307809","doi":"10.1002/sim.9593","title":"Dose finding studies for therapies with late‐onset toxicities: A comparison study of designs","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"NIHR Cambridge Biomedical Research Centre; Medical Research Council Canada; Department of Health and Social Care; Medical Research Council; National Institute for Health and Care Research","keywords":"Maximum tolerated dose; Medicine; Clinical trial; Early stopping; Confidence interval; Set (abstract data type); Toxicity; Interval (graph theory); Computer science; Statistics; Oncology; Mathematics; Internal medicine; Machine learning","score_opus":0.7531072275672495,"score_gpt":0.6337813978697665,"score_spread":0.11932582969748295,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3159307809","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.33249643,0.01791127,0.61340964,0.002137898,0.0013927239,0.019279238,0.0011759708,0.0005260653,0.011670779],"genre_scores_gemma":[0.69494474,0.004256543,0.27665836,0.0009390126,0.00027781934,0.019725224,0.0005541015,0.00013456405,0.002509586],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.83377105,0.15409987,0.0042369626,0.0031316138,0.004083645,0.00067684683],"domain_scores_gemma":[0.60615623,0.3568101,0.014765596,0.016414,0.0045813834,0.0012727522],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.14397055,0.0015762405,0.0032384295,0.0016775844,0.0003820975,0.002238217,0.0022898873,0.0030483494,0.007544538],"category_scores_gemma":[0.3025106,0.0007190066,0.0062913713,0.0010935632,0.0013014197,0.0034626548,0.0016680404,0.0029210725,0.0005337877],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.17859374,0.008092856,0.012688676,0.009741143,0.019826973,0.00022281542,0.0009491368,0.26490146,0.0037052543,0.14202937,0.0033352438,0.3559133],"study_design_scores_gemma":[0.087389484,0.17937373,0.013424087,0.0036437912,0.011291855,0.00055509753,0.0003926867,0.4728271,0.005870342,0.20356743,0.021094123,0.0005703059],"about_ca_topic_score_codex":0.0004239381,"about_ca_topic_score_gemma":0.00035709623,"teacher_disagreement_score":0.85602945,"about_ca_system_score_codex":0.0023567327,"about_ca_system_score_gemma":0.0027583921,"threshold_uncertainty_score":0.76139814},"labels":[],"label_agreement":null},{"id":"W3159320036","doi":"10.1002/sim.9000","title":"Score tests for scale effects, with application to genomic analysis","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Wilfrid Laurier University; University of Toronto; University of Waterloo; Ontario Institute for Cancer Research","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Covariate; Statistics; Mathematics; Type I and type II errors; Permutation (music); Scale (ratio); Gaussian; Applied mathematics","score_opus":0.24350483640801132,"score_gpt":0.5516258013485317,"score_spread":0.3081209649405203,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3159320036","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.012804594,0.00045540967,0.98450667,0.0005213684,0.00007554595,0.00007200496,0.00015103878,0.00034521925,0.0010681964],"genre_scores_gemma":[0.41627494,0.0009115873,0.57736564,0.0005334345,0.0005085116,0.0006677267,0.0007027155,0.0004465232,0.0025889971],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9837208,0.011918856,0.00048826655,0.0011937675,0.0024455192,0.00023274773],"domain_scores_gemma":[0.8833089,0.101909444,0.004391736,0.0063225674,0.003284525,0.0007829082],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016719254,0.0011725437,0.0015391492,0.002663536,0.0005806916,0.0017114149,0.001706612,0.0013954405,0.003428908],"category_scores_gemma":[0.1756498,0.00039604664,0.0012118944,0.0038195073,0.0043118927,0.002212181,0.002328221,0.0027584946,0.000604837],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00049903215,0.00016635189,0.024110204,0.000616193,0.00074574293,0.000952068,0.00047519107,0.13891476,0.0051289247,0.4826513,0.009371463,0.33636883],"study_design_scores_gemma":[0.000118405165,0.000355089,0.0076275785,0.00006244062,0.00008988173,0.000486983,0.00012068593,0.3828008,0.0016214958,0.6024629,0.004182202,0.00007154777],"about_ca_topic_score_codex":0.0011904128,"about_ca_topic_score_gemma":0.0007580691,"teacher_disagreement_score":0.016719254,"about_ca_system_score_codex":0.0008663879,"about_ca_system_score_gemma":0.0015466567,"threshold_uncertainty_score":0.08842093},"labels":[],"label_agreement":null},{"id":"W3161655502","doi":"10.1002/sim.9023","title":"Fine‐Gray subdistribution hazard models to simultaneously estimate the absolute risk of different event types: Cumulative total failure probability may exceed 1","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":88,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Work & Health; Institute for Clinical Evaluative Sciences; Sunnybrook Hospital; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Heart and Stroke Foundation of Canada","keywords":"Covariate; Hazard; Cumulative incidence; Statistics; Hazard ratio; Proportional hazards model; Event (particle physics); Mathematics; Confidence interval","score_opus":0.1656706911041427,"score_gpt":0.42450913207225904,"score_spread":0.25883844096811637,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3161655502","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02616172,0.00044783475,0.9698344,0.00036479803,0.00007687906,0.00032328052,0.0006128032,0.00042265505,0.0017556715],"genre_scores_gemma":[0.77503186,0.0009527904,0.20731302,0.00069714815,0.00018196518,0.0014322761,0.0019443838,0.00025379012,0.012192792],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99509406,0.0028718181,0.00020995058,0.0007830831,0.00068477006,0.00035625452],"domain_scores_gemma":[0.97812825,0.015918545,0.0018781618,0.0028181777,0.0009298403,0.0003271437],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.019898733,0.0016551708,0.0017740104,0.0018370968,0.000371819,0.0015344094,0.003189824,0.0016497745,0.006313096],"category_scores_gemma":[0.02956716,0.00096990675,0.0032286213,0.0014171947,0.001424458,0.0023649603,0.0021707532,0.0035053135,0.00091367774],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000619584,0.00018591828,0.033560567,0.0004092362,0.0010814266,0.0005678591,0.00075046846,0.6892479,0.0025818092,0.1790056,0.006607531,0.08538213],"study_design_scores_gemma":[0.00008407562,0.00018324774,0.003182823,0.000058918795,0.00014963791,0.000338649,0.00006239371,0.89670885,0.0007951592,0.093794174,0.0045875353,0.000054585376],"about_ca_topic_score_codex":0.005576524,"about_ca_topic_score_gemma":0.00412113,"teacher_disagreement_score":0.019898733,"about_ca_system_score_codex":0.0012654581,"about_ca_system_score_gemma":0.001535521,"threshold_uncertainty_score":0.105235815},"labels":[],"label_agreement":null},{"id":"W3173746581","doi":"10.1002/sim.9109","title":"Bayesian variable selection with a pleiotropic loss function in Mendelian randomization","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic Associations and Epidemiology","field":"Biochemistry, Genetics and Molecular Biology","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; National Cancer Institute; National Institutes of Health; National Institute for Health and Care Research; Medical Research Council Canada; NIHR Cambridge Biomedical Research Centre; British Heart Foundation; Wellcome Trust","keywords":"Mendelian randomization; Pleiotropy; Selection (genetic algorithm); Inference; Context (archaeology); Computer science; Statistics; Genetics; Biology; Mathematics; Genetic variants; Machine learning; Artificial intelligence; Phenotype; Gene","score_opus":0.006242120170021306,"score_gpt":0.25977184954295146,"score_spread":0.25352972937293017,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3173746581","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00227772,0.00014625133,0.9969464,0.00014018735,0.000021533595,0.000057789133,0.000020761889,0.00014150003,0.00024782217],"genre_scores_gemma":[0.12664376,0.0005238582,0.8690117,0.00040974378,0.00014615348,0.0009251814,0.00022980016,0.0001730335,0.0019368728],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9850959,0.012139787,0.00031089855,0.0009405799,0.0012007877,0.00031210823],"domain_scores_gemma":[0.9718337,0.024640793,0.000977643,0.0013977389,0.00087505905,0.00027512512],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.023574771,0.0010521379,0.0022562752,0.0014522604,0.0007861381,0.0015035735,0.0023373074,0.0018922818,0.0026072147],"category_scores_gemma":[0.056379832,0.0007837318,0.0013083441,0.0017870085,0.0027344483,0.002035516,0.002452174,0.0027319437,0.0006898682],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00047538662,0.00011950046,0.0036988987,0.00024016181,0.00040620638,0.000273912,0.0001860089,0.4754806,0.0017842018,0.35347068,0.0035645042,0.16029999],"study_design_scores_gemma":[0.0001866564,0.00010939424,0.00045801944,0.000043723092,0.000040008774,0.000105687504,0.000012720772,0.8121161,0.0007204867,0.18407935,0.0020918576,0.00003607298],"about_ca_topic_score_codex":0.0026942973,"about_ca_topic_score_gemma":0.0022354417,"teacher_disagreement_score":0.023574771,"about_ca_system_score_codex":0.001271906,"about_ca_system_score_gemma":0.0028939382,"threshold_uncertainty_score":0.124676764},"labels":[],"label_agreement":null},{"id":"W3176979971","doi":"10.1002/sim.9097","title":"Estimation of diagnostic test accuracy: A “Rule of Three” for data with repeated observations but without a gold standard","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Multinomial distribution; Statistics; Computer science; Goodness of fit; Gold standard (test); Degrees of freedom (physics and chemistry); Test (biology); Mathematics","score_opus":0.1483904432595583,"score_gpt":0.4300721741808364,"score_spread":0.2816817309212781,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3176979971","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007512024,0.0026335898,0.9846795,0.0025697309,0.00045772523,0.0003019015,0.00019938168,0.0003160366,0.0013300241],"genre_scores_gemma":[0.18718146,0.0010978093,0.8051366,0.0033143696,0.00078353885,0.001103933,0.0004590146,0.00030461056,0.0006187695],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5454236,0.3434803,0.030833583,0.029456118,0.04836409,0.0024422912],"domain_scores_gemma":[0.25025406,0.6548974,0.02388303,0.053423055,0.016119534,0.0014230117],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.36545005,0.002946964,0.008780839,0.007854655,0.0030982709,0.008818237,0.007837966,0.010528138,0.0011172509],"category_scores_gemma":[0.7013905,0.0020446524,0.0074823084,0.004918231,0.017364,0.008910326,0.008612079,0.014820624,0.001090484],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020832894,0.00035609456,0.12603267,0.003844402,0.0070017916,0.0022503445,0.004404429,0.059472445,0.0027931703,0.3536062,0.016891865,0.42126337],"study_design_scores_gemma":[0.0003338548,0.0009822726,0.011775108,0.0026039302,0.0011350284,0.00305182,0.0005142202,0.20449264,0.005095776,0.7494755,0.020084405,0.00045543353],"about_ca_topic_score_codex":0.0029847964,"about_ca_topic_score_gemma":0.0017872824,"teacher_disagreement_score":0.63455,"about_ca_system_score_codex":0.0035791798,"about_ca_system_score_gemma":0.0046223737,"threshold_uncertainty_score":0.7825131},"labels":[],"label_agreement":null},{"id":"W3179584355","doi":"10.1002/sim.9126","title":"Fitting marginal models in small samples: A simulation study of marginalized multilevel models and generalized estimating equations","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"National Heart, Lung, and Blood Institute","keywords":"Marginal model; Generalized estimating equation; Gee; Inference; Econometrics; Estimating equations; Sample size determination; Statistics; Multilevel model; Statistical inference; Cluster (spacecraft); Random effects model; Sample (material); Population; Mathematics; Computer science; Regression analysis; Meta-analysis; Estimator; Physics; Demography; Artificial intelligence","score_opus":0.3026635934558983,"score_gpt":0.45129308017905495,"score_spread":0.14862948672315662,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3179584355","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.23654427,0.0036105583,0.74177676,0.005603147,0.00013761531,0.00046123125,0.0005336464,0.00032685767,0.0110059865],"genre_scores_gemma":[0.7508728,0.0018836089,0.24270926,0.0008176817,0.0000973007,0.00079096504,0.0005038746,0.00018994877,0.0021345825],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9754843,0.021940645,0.00036519064,0.00094198907,0.00090189616,0.0003660065],"domain_scores_gemma":[0.6876401,0.29509816,0.005030215,0.0065835263,0.004628503,0.0010194761],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.044569477,0.00091238855,0.0019158253,0.0014894096,0.0010289429,0.002016785,0.002444625,0.0020660441,0.004227848],"category_scores_gemma":[0.19953063,0.00067193713,0.0022373493,0.0027776384,0.00228566,0.004200985,0.0023317747,0.0047431956,0.00036948625],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004179898,0.0002864962,0.026696445,0.00040595088,0.00062165415,0.00066005316,0.001952369,0.5202715,0.0002460942,0.40656763,0.004366571,0.03750721],"study_design_scores_gemma":[0.00013397337,0.0001631768,0.0024227921,0.00015888715,0.00012736663,0.0001384351,0.00043505468,0.8448356,0.00016705501,0.14891657,0.0024548452,0.000046268444],"about_ca_topic_score_codex":0.019649081,"about_ca_topic_score_gemma":0.018256776,"teacher_disagreement_score":0.9554305,"about_ca_system_score_codex":0.0025088033,"about_ca_system_score_gemma":0.0021663578,"threshold_uncertainty_score":0.23570871},"labels":[],"label_agreement":null},{"id":"W3185479023","doi":"10.1002/sim.9135","title":"Optimal design of cluster randomized trials allowing unequal allocation of clusters and unequal cluster size between arms","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada","keywords":"Cluster (spacecraft); Variance (accounting); Range (aeronautics); Sample size determination; Statistics; Cluster size; Statistical power; Computer science; Cluster randomised controlled trial; Mathematics; Randomized controlled trial; Medicine; Physics; Engineering","score_opus":0.22106253953040395,"score_gpt":0.4884150912243174,"score_spread":0.2673525516939135,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3185479023","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0049895165,0.0011908194,0.9847943,0.0005850724,0.00020686263,0.005751492,0.0001473436,0.00022817585,0.0021063902],"genre_scores_gemma":[0.0854411,0.0009539129,0.8864659,0.00054597034,0.00012479276,0.025555085,0.00017157092,0.00010868511,0.00063300546],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7875746,0.19102041,0.005128389,0.006965064,0.00772144,0.001590125],"domain_scores_gemma":[0.8630456,0.114827655,0.0076095825,0.008202544,0.0050878953,0.0012267717],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09977196,0.0025693572,0.0046201097,0.003142007,0.001080224,0.0032362523,0.0031746922,0.0031051831,0.004136187],"category_scores_gemma":[0.2030751,0.0019236414,0.0027628476,0.0025487877,0.005410612,0.0026689642,0.003347177,0.0039484925,0.0008183214],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008946451,0.0006416702,0.0027835506,0.0061574257,0.0020827441,0.0002562739,0.0011135121,0.19035748,0.0025757952,0.5168596,0.008115599,0.26010996],"study_design_scores_gemma":[0.0078000473,0.004839096,0.0019136096,0.0017447742,0.0011877374,0.00022406835,0.0001791125,0.3038446,0.004743527,0.6535648,0.01967381,0.00028481294],"about_ca_topic_score_codex":0.00071888085,"about_ca_topic_score_gemma":0.0007250192,"teacher_disagreement_score":0.900228,"about_ca_system_score_codex":0.0032998996,"about_ca_system_score_gemma":0.008331471,"threshold_uncertainty_score":0.52765083},"labels":[],"label_agreement":null},{"id":"W3189430776","doi":"10.1002/sim.9141","title":"Estimating adjusted risk differences by multiply‐imputing missing control binary potential outcomes following propensity score‐matching","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Statistics; Imputation (statistics); Missing data; Confidence interval; Matching (statistics); Mean squared error; Absolute risk reduction; Regression; Mathematics; Econometrics; Medicine","score_opus":0.0927472377061911,"score_gpt":0.39244463014258196,"score_spread":0.2996973924363909,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3189430776","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.006702872,0.00017359434,0.9922902,0.00011782847,0.000034911605,0.00014865355,0.00008831689,0.00018761524,0.00025601688],"genre_scores_gemma":[0.13476676,0.00036064183,0.86256546,0.00023891352,0.00011325043,0.0008433916,0.00038982913,0.000098764474,0.0006230153],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9745755,0.018984744,0.0011451092,0.0019423268,0.0030113584,0.00034102326],"domain_scores_gemma":[0.9280536,0.057755202,0.0048682177,0.0073032985,0.0018158997,0.00020384304],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03975683,0.000915036,0.0027523076,0.003018562,0.00050926086,0.0013532424,0.0021710435,0.001349987,0.0028341599],"category_scores_gemma":[0.12220735,0.00088280486,0.0028313461,0.0034519518,0.001298815,0.0014829937,0.002310847,0.002412127,0.00034549803],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00082257605,0.00046386366,0.038601067,0.00092204567,0.0047821356,0.00043436675,0.0009769301,0.18910453,0.0044621485,0.1858926,0.0037664967,0.5697713],"study_design_scores_gemma":[0.0005319699,0.00067158975,0.016430626,0.00024343209,0.0010264379,0.00050291547,0.00008642108,0.70131797,0.006017245,0.26288044,0.01011098,0.00018000008],"about_ca_topic_score_codex":0.0029107041,"about_ca_topic_score_gemma":0.0020734721,"teacher_disagreement_score":0.03975683,"about_ca_system_score_codex":0.0007741544,"about_ca_system_score_gemma":0.0019739275,"threshold_uncertainty_score":0.21025676},"labels":[],"label_agreement":null},{"id":"W3191796027","doi":"10.1002/sim.9133","title":"Testing and correcting for weak and pleiotropic instruments in two‐sample multivariable Mendelian randomization","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic Associations and Epidemiology","field":"Biochemistry, Genetics and Molecular Biology","cited_by":630,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada; Wellcome Trust","keywords":"Mendelian randomization; Statistics; Pleiotropy; Test statistic; Instrumental variable; Statistic; Mathematics; Econometrics; Sample size determination; Resampling; Statistical hypothesis testing; Biology; Genetics; Genetic variants","score_opus":0.021971664813142743,"score_gpt":0.32767985268335337,"score_spread":0.30570818787021065,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3191796027","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02917955,0.00018042712,0.96896654,0.00042099232,0.00008459651,0.0001993354,0.00016045134,0.00032863265,0.00047951052],"genre_scores_gemma":[0.38492927,0.00025421914,0.61172354,0.00053285115,0.00012443078,0.0011044128,0.00041331496,0.00016212683,0.0007558428],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.87696105,0.10079015,0.00478314,0.00789833,0.0075758984,0.0019913942],"domain_scores_gemma":[0.5954171,0.35901642,0.013466338,0.026037468,0.005090317,0.00097238366],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.120677896,0.0012733915,0.0032295329,0.0024360744,0.0012538837,0.0022171387,0.0036071953,0.0024338365,0.0032958807],"category_scores_gemma":[0.39661935,0.0008981576,0.0033117773,0.00339949,0.004877315,0.0030869201,0.00407042,0.0031612527,0.00039151817],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019411865,0.00040480358,0.13194817,0.001010042,0.005247696,0.0021143325,0.0017678532,0.09645776,0.004030628,0.4319166,0.0042491783,0.31891176],"study_design_scores_gemma":[0.00057986676,0.001126585,0.02722033,0.00026603346,0.00077389996,0.0008649697,0.00031973282,0.4541307,0.005287883,0.50270087,0.0065193498,0.00020976491],"about_ca_topic_score_codex":0.002755305,"about_ca_topic_score_gemma":0.002271947,"teacher_disagreement_score":0.8793221,"about_ca_system_score_codex":0.0009800218,"about_ca_system_score_gemma":0.004863716,"threshold_uncertainty_score":0.63821334},"labels":[],"label_agreement":null},{"id":"W3193216897","doi":"10.1002/sim.9151","title":"Estimating the marginal effect of a continuous exposure on an ordinal outcome using data subject to covariate‐driven treatment and visit processes","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Fonds de Recherche du Québec - Santé; Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Covariate; Marginal structural model; Outcome (game theory); Confounding; Causal inference; Econometrics; Inference; Statistics; Odds; Generalization; Computer science; Inverse probability weighting; Mathematics; Logistic regression; Artificial intelligence; Propensity score matching","score_opus":0.18254403681751505,"score_gpt":0.48358799182642537,"score_spread":0.30104395500891035,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3193216897","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03690725,0.00056517706,0.9595452,0.000797544,0.00009430249,0.00043172197,0.00063556555,0.00022400933,0.00079924613],"genre_scores_gemma":[0.57685536,0.0014361897,0.4116212,0.00079089036,0.00031598588,0.003055235,0.001712993,0.00011934551,0.004092717],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9744932,0.018798545,0.00097540003,0.0034493024,0.0016785415,0.0006050814],"domain_scores_gemma":[0.85712886,0.11912118,0.008334319,0.012697126,0.001988651,0.000729817],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.052224416,0.0012380995,0.0023729966,0.002947582,0.00090320193,0.0020773693,0.0035757823,0.0026186635,0.006549387],"category_scores_gemma":[0.15587588,0.0010758436,0.004198068,0.0034445864,0.0037257986,0.0034418849,0.0032146394,0.004562186,0.0006595706],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013201019,0.000874282,0.14142749,0.0014927718,0.003383348,0.00067932316,0.0024485413,0.111799106,0.0016527221,0.5472613,0.003127018,0.18453407],"study_design_scores_gemma":[0.0003916723,0.0008775617,0.033603016,0.00048359198,0.0013164587,0.0004403491,0.0004666034,0.37081382,0.0029248307,0.5803251,0.008208599,0.00014844381],"about_ca_topic_score_codex":0.0048220847,"about_ca_topic_score_gemma":0.0032175088,"teacher_disagreement_score":0.052224416,"about_ca_system_score_codex":0.0015272115,"about_ca_system_score_gemma":0.0030416008,"threshold_uncertainty_score":0.27619237},"labels":[],"label_agreement":null},{"id":"W3193380525","doi":"10.1002/sim.9167","title":"A Bayesian nonparametric approach to dynamic item‐response modeling: An application to the GUSTO cohort study","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Douglas Mental Health University Institute","funders":"Medical Research Council; National Medical Research Council; National Research Foundation Singapore; Singapore Institute for Clinical Sciences; National University Health System; National Research Foundation","keywords":"Nonparametric statistics; Respondent; Cluster analysis; Item response theory; Bayesian probability; Cohort; Computer science; Econometrics; Statistics; Psychology; Machine learning; Psychometrics; Mathematics; Artificial intelligence","score_opus":0.05233594052253809,"score_gpt":0.4136245242099736,"score_spread":0.3612885836874355,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3193380525","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0361959,0.00046516256,0.96009177,0.0014625603,0.000055114968,0.00028057734,0.00042197283,0.00014076395,0.00088623026],"genre_scores_gemma":[0.43792167,0.0011031829,0.5544958,0.00053111155,0.00018754431,0.0018363637,0.0008279105,0.00011804527,0.0029783596],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9789596,0.019060802,0.00025417228,0.0007752831,0.0007345698,0.00021558408],"domain_scores_gemma":[0.94414675,0.049817506,0.0015041281,0.0030645172,0.0010710879,0.00039606754],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03674323,0.0007093144,0.0014015196,0.0016164082,0.00093675306,0.0013057988,0.0024710053,0.0016263206,0.0023489224],"category_scores_gemma":[0.07849805,0.00063195796,0.0016274707,0.0020040113,0.0012888083,0.0009057326,0.0021872835,0.0025089646,0.0003544727],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010795388,0.0005743452,0.05706485,0.00042261768,0.0015317983,0.0016119834,0.0030668706,0.26979458,0.0011629976,0.43800002,0.008109298,0.21758111],"study_design_scores_gemma":[0.00017044821,0.00018621252,0.0101700695,0.00009276871,0.00012325973,0.0003237356,0.00030404868,0.7743108,0.00018107299,0.20815934,0.0058934772,0.00008471728],"about_ca_topic_score_codex":0.016189385,"about_ca_topic_score_gemma":0.0122537175,"teacher_disagreement_score":0.03674323,"about_ca_system_score_codex":0.0012727306,"about_ca_system_score_gemma":0.001960493,"threshold_uncertainty_score":0.19431913},"labels":[],"label_agreement":null},{"id":"W3194048807","doi":"10.1002/sim.9159","title":"A Bayesian approach for estimating typhoid fever incidence from large‐scale facility‐based passive surveillance data","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Salmonella and Campylobacter epidemiology","field":"Agricultural and Biological Sciences","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Infection and Immunity","funders":"Public Health Research Programme; Government of the United Kingdom; Department of Health and Social Care; Wellcome Trust; National Institute for Health and Care Research; Bill and Melinda Gates Foundation","keywords":"Typhoid fever; Incidence (geometry); Medicine; Credible interval; Blood culture; Population; Environmental health; Demography; Confidence interval; Virology; Internal medicine; Biology; Mathematics","score_opus":0.05473443333751478,"score_gpt":0.31792407375043935,"score_spread":0.2631896404129246,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3194048807","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02272412,0.00053791085,0.9735448,0.0004233703,0.000028590212,0.00034221757,0.0008841796,0.0003022974,0.0012125645],"genre_scores_gemma":[0.41513962,0.0015799154,0.5728259,0.00044336604,0.00017659359,0.0022146895,0.0044457233,0.0001461107,0.00302807],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9930835,0.004794325,0.0003281721,0.0009840315,0.0005978319,0.00021219159],"domain_scores_gemma":[0.9776315,0.018830612,0.0015911462,0.0006560477,0.0010723881,0.00021829564],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014677088,0.0013008026,0.0020174733,0.0037972177,0.0010026351,0.0023745878,0.0036604253,0.0020456626,0.0023055451],"category_scores_gemma":[0.045796115,0.0020960644,0.0024251242,0.0034366322,0.001250995,0.0025012125,0.0015580195,0.0028212979,0.00064285734],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001813496,0.00013605159,0.012662997,0.00024693224,0.0004833085,0.00026044282,0.00032034755,0.8843335,0.0005139751,0.030529482,0.0019336158,0.06839794],"study_design_scores_gemma":[0.000060393908,0.000066458524,0.003499254,0.000099834964,0.00009815088,0.00015464703,0.00005463487,0.9534774,0.00015182652,0.040552363,0.0017197349,0.00006540023],"about_ca_topic_score_codex":0.041690502,"about_ca_topic_score_gemma":0.03410143,"teacher_disagreement_score":0.041690502,"about_ca_system_score_codex":0.0027897512,"about_ca_system_score_gemma":0.0028005342,"threshold_uncertainty_score":0.08289564},"labels":[],"label_agreement":null},{"id":"W3196973804","doi":"10.1002/sim.3967","title":"Correction","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Toronto Metropolitan University","funders":"","keywords":"Regret; Statistics; Confidence interval; Proofreading; Estimator; Mathematics; Computer science; Medicine","score_opus":0.04686030747778479,"score_gpt":0.41881745912163143,"score_spread":0.37195715164384663,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3196973804","genre_codex":"editorial","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00052110094,0.0012775238,0.006334702,0.0751095,0.8921367,0.00016440573,0.011394729,0.002832376,0.0102289375],"genre_scores_gemma":[0.036192913,0.004004,0.03194685,0.09575424,0.24521688,0.0012111743,0.035989985,0.01356767,0.5361163],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9776774,0.0047504855,0.0039121807,0.0031457925,0.00906611,0.0014481475],"domain_scores_gemma":[0.77332,0.042212814,0.00828737,0.028448118,0.14119236,0.006539377],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.014211453,0.001980152,0.0020594874,0.008229463,0.004094911,0.005394864,0.005077888,0.004916943,0.2821386],"category_scores_gemma":[0.25789562,0.0010224676,0.00263372,0.004637699,0.0025290234,0.004646779,0.004841323,0.009409727,0.13699856],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000043789976,0.000005292666,0.00014457482,0.00014707369,0.000017869925,0.00012971017,0.0000788467,0.000028460085,0.00008746364,0.0015165906,0.9869732,0.010827143],"study_design_scores_gemma":[0.00002641079,0.000013167975,0.00060170214,0.0002808738,0.000034195575,0.00037527672,0.00009968905,0.00013683017,0.00034885423,0.003056287,0.99498737,0.000039402094],"about_ca_topic_score_codex":0.005535771,"about_ca_topic_score_gemma":0.0074233646,"teacher_disagreement_score":0.7178614,"about_ca_system_score_codex":0.00398777,"about_ca_system_score_gemma":0.0075442847,"threshold_uncertainty_score":0},"labels":[],"label_agreement":null},{"id":"W3199775394","doi":"10.1002/sim.9176","title":"Informing power and sample size calculations when using inverse probability of treatment weighting using the propensity score","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":48,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Institute of Health Services and Policy Research; Sunnybrook Hospital; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Observational study; Weighting; Statistics; Sample size determination; Average treatment effect; Mathematics; Statistic; Homogeneity (statistics); Econometrics; Inverse probability weighting; Medicine","score_opus":0.27352121820347886,"score_gpt":0.4268053675944418,"score_spread":0.15328414939096296,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3199775394","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0027815374,0.001171811,0.9867786,0.0016953173,0.00046549854,0.0029085253,0.00035659422,0.0003366933,0.0035054935],"genre_scores_gemma":[0.055964828,0.0013069683,0.92605275,0.0013569895,0.00053100096,0.013274095,0.0003491103,0.00039663343,0.0007676927],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.64859754,0.31124583,0.015231026,0.0069968808,0.01685616,0.0010725169],"domain_scores_gemma":[0.4176637,0.50564605,0.019573424,0.04022536,0.015943129,0.00094843353],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27097908,0.0018234253,0.0033358748,0.0066871857,0.0013079988,0.005947248,0.003912028,0.0036212052,0.010119706],"category_scores_gemma":[0.67251515,0.0018013127,0.0031462633,0.0077788546,0.0043590427,0.0063376124,0.0057327603,0.0066690478,0.0021604903],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012434471,0.00022363475,0.012211488,0.004106768,0.0020192477,0.00035886836,0.00318894,0.021357529,0.0015777939,0.24093144,0.0282728,0.68450814],"study_design_scores_gemma":[0.0019296765,0.0011774981,0.011372444,0.004915281,0.0015870868,0.0007648875,0.0005188005,0.10330348,0.008707074,0.7361843,0.12920035,0.00033916472],"about_ca_topic_score_codex":0.0017346274,"about_ca_topic_score_gemma":0.0015988636,"teacher_disagreement_score":0.72902095,"about_ca_system_score_codex":0.0019158042,"about_ca_system_score_gemma":0.005166983,"threshold_uncertainty_score":0.89901257},"labels":[],"label_agreement":null},{"id":"W3200775649","doi":"10.1002/sim.9562","title":"Network meta‐analysis of rare events using penalized likelihood regression","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":29,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Horizon 2020 Framework Programme; Medical Research Council Canada","keywords":"Overdispersion; Rare events; Statistics; Computer science; Econometrics; Multiplicative function; Likelihood function; Meta-analysis; Mathematics; Maximum likelihood; Count data; Poisson distribution; Medicine","score_opus":0.1664385312629363,"score_gpt":0.44900875208886887,"score_spread":0.2825702208259325,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3200775649","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0077766753,0.026718369,0.96102244,0.0012619931,0.00025323743,0.0007295738,0.0007375359,0.0007136111,0.0007866044],"genre_scores_gemma":[0.3744614,0.01460777,0.6004192,0.0012120908,0.00050989015,0.0048290235,0.0017029566,0.0005314335,0.0017262099],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8947655,0.09701447,0.0024510024,0.003270259,0.0021741574,0.00032464144],"domain_scores_gemma":[0.87254614,0.117873445,0.003916375,0.00411222,0.0013468998,0.00020489984],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06026303,0.0021458361,0.0066769845,0.00545672,0.0006563854,0.0028985417,0.0031064984,0.0018507537,0.0037659325],"category_scores_gemma":[0.13311048,0.0010703132,0.01283436,0.004821256,0.0008121255,0.0023359125,0.0023354825,0.0035677787,0.00038849513],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004771101,0.00020297122,0.013145803,0.027455486,0.18424912,0.0010913899,0.00041761735,0.41801208,0.0015266836,0.06222798,0.008963211,0.27793658],"study_design_scores_gemma":[0.0020544364,0.000708833,0.0040296866,0.0024879556,0.06743093,0.00047263104,0.00007692097,0.77065563,0.0015317006,0.13542554,0.014962393,0.00016338911],"about_ca_topic_score_codex":0.003328291,"about_ca_topic_score_gemma":0.0028797158,"teacher_disagreement_score":0.93973696,"about_ca_system_score_codex":0.0017744562,"about_ca_system_score_gemma":0.001997916,"threshold_uncertainty_score":0.31870514},"labels":[],"label_agreement":null},{"id":"W3203445629","doi":"10.1002/sim.9718","title":"Causal inference in survival analysis using longitudinal observational data: Sequential trials and marginal structural models","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada; Norges Forskningsråd; UK Research and Innovation","keywords":"Marginal structural model; Causal inference; Censoring (clinical trials); Observational study; Inverse probability; Econometrics; Computer science; Confounding; Inference; Statistics; Inverse probability weighting; Propensity score matching; Mathematics; Artificial intelligence; Bayesian probability","score_opus":0.7073736049282677,"score_gpt":0.5705797811041333,"score_spread":0.1367938238241344,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3203445629","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0119984215,0.004240668,0.97703415,0.0028479826,0.00041110587,0.00093165017,0.0005064818,0.00023562119,0.0017938368],"genre_scores_gemma":[0.3922407,0.00719747,0.5871201,0.0021860795,0.001158831,0.0056973174,0.0010172643,0.00020398422,0.0031782633],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.85473335,0.13111381,0.0027321258,0.005642614,0.004795954,0.0009820651],"domain_scores_gemma":[0.6016759,0.36416098,0.013095996,0.014949889,0.0048609604,0.0012562762],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.14201562,0.0023574964,0.005053905,0.0035999597,0.0010830609,0.00332063,0.004800575,0.0039971787,0.007827482],"category_scores_gemma":[0.3508202,0.0015840206,0.006779306,0.0049563907,0.00486403,0.006322307,0.004682478,0.0064848587,0.00062880886],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016619079,0.00030205128,0.015089594,0.0028343487,0.0045557194,0.0005751675,0.0013478887,0.14800885,0.00043722117,0.69343126,0.003467981,0.12828794],"study_design_scores_gemma":[0.0007016281,0.00037829933,0.0016246566,0.00050158834,0.00096313277,0.00016196698,0.0000996638,0.24490388,0.0003736559,0.74619627,0.0040182616,0.000076932294],"about_ca_topic_score_codex":0.007492426,"about_ca_topic_score_gemma":0.005284727,"teacher_disagreement_score":0.14201562,"about_ca_system_score_codex":0.0033244716,"about_ca_system_score_gemma":0.0058080833,"threshold_uncertainty_score":0.75105935},"labels":[],"label_agreement":null},{"id":"W3204069683","doi":"10.1002/sim.9203","title":"A comparison of methods for analyzing a binary composite endpoint with partially observed components in randomized controlled trials","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council Canada; Medical Research Council; National Institute for Health and Care Research; NHS Blood and Transplant","keywords":"Composite number; Binary number; Randomized controlled trial; Statistics; Computer science; Mathematics; Medicine; Algorithm; Internal medicine; Arithmetic","score_opus":0.28725368993790196,"score_gpt":0.5240229339826206,"score_spread":0.2367692440447186,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3204069683","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0070669837,0.014576739,0.97159326,0.0015256156,0.0004922834,0.0026294717,0.00031939018,0.00034113752,0.0014550905],"genre_scores_gemma":[0.09147165,0.008932898,0.8822473,0.0010286081,0.00037152637,0.014575187,0.00038956688,0.00025986088,0.0007234378],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5145625,0.4505728,0.012065898,0.005682296,0.016486026,0.00063044985],"domain_scores_gemma":[0.25049475,0.7111535,0.014208901,0.016432429,0.006830898,0.00087954564],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.32434765,0.0026294908,0.005622129,0.005677556,0.00085878646,0.0038311963,0.0043518962,0.0047714645,0.0052605],"category_scores_gemma":[0.5592189,0.0016309081,0.00943195,0.0052847643,0.0035805968,0.0057060397,0.003347064,0.0062540723,0.0007417557],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.016499847,0.00081234693,0.0072904713,0.019096486,0.024636189,0.00016484488,0.0014106084,0.10071779,0.0009876041,0.349067,0.007279706,0.4720371],"study_design_scores_gemma":[0.012638826,0.007099178,0.007179766,0.00810032,0.0077618337,0.000521887,0.00025243158,0.47592255,0.0017533457,0.46243227,0.015783671,0.0005539509],"about_ca_topic_score_codex":0.0018604349,"about_ca_topic_score_gemma":0.0017094112,"teacher_disagreement_score":0.32434765,"about_ca_system_score_codex":0.004944125,"about_ca_system_score_gemma":0.006662661,"threshold_uncertainty_score":0.8331996},"labels":[],"label_agreement":null},{"id":"W3204449925","doi":"10.1002/sim.9211","title":"Two‐phase sample selection strategies for design and analysis in post‐genome‐wide association fine‐mapping studies","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic and phenotypic traits in livestock","field":"Biochemistry, Genetics and Molecular Biology","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Lunenfeld-Tanenbaum Research Institute; Princess Margaret Cancer Centre; Public Health Ontario; University of Toronto; University Health Network","funders":"National Heart, Lung, and Blood Institute; Canadian Institutes of Health Research; Terveyden ja hyvinvoinnin laitos; Oulun Yliopisto; Government of Ontario; Broad Institute; Compute Canada; University of Toronto; Ontario Institute for Cancer Research","keywords":"Selection (genetic algorithm); Sample (material); Computer science; Sample size determination; Genome-wide association study; Genetic association; Association (psychology); Statistics; Biology; Genetics; Mathematics; Artificial intelligence; Psychology; Single-nucleotide polymorphism; Genotype; Chromatography","score_opus":0.029815570153603213,"score_gpt":0.33703081872226015,"score_spread":0.30721524856865695,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3204449925","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0031174403,0.00011222431,0.99517363,0.00007544795,0.00003831733,0.0010631018,0.000033754608,0.00014826642,0.00023773253],"genre_scores_gemma":[0.055199794,0.0001254844,0.93840283,0.00021495846,0.00004587554,0.0053858673,0.0001289926,0.00006834946,0.0004278565],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9553318,0.03838793,0.0009552073,0.002039607,0.00287328,0.00041206868],"domain_scores_gemma":[0.94810164,0.04137597,0.0024280958,0.004207969,0.0033496923,0.00053666293],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.049026065,0.0019862112,0.0016209888,0.0016160435,0.0006565211,0.0013605101,0.0024457963,0.0024036544,0.0044568367],"category_scores_gemma":[0.09202525,0.0010763023,0.0016229693,0.0012235851,0.001767638,0.001333732,0.0020056828,0.0023986446,0.0010184214],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0045424565,0.0014549632,0.006220768,0.0015807475,0.001115924,0.00036399136,0.0010011271,0.26217282,0.012832342,0.20593669,0.004767764,0.4980105],"study_design_scores_gemma":[0.0038369657,0.005004409,0.0031270995,0.00028044637,0.0003669354,0.00021997957,0.00014472901,0.81545216,0.01004241,0.14630847,0.015057333,0.00015907777],"about_ca_topic_score_codex":0.0006478633,"about_ca_topic_score_gemma":0.00084357226,"teacher_disagreement_score":0.049026065,"about_ca_system_score_codex":0.0009774761,"about_ca_system_score_gemma":0.0035042493,"threshold_uncertainty_score":0.25927764},"labels":[],"label_agreement":null},{"id":"W3205527824","doi":"10.1002/sim.9225","title":"Bayesian consensus clustering for multivariate longitudinal data","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Liver Disease Diagnosis and Treatment","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto; Queen's University","funders":"","keywords":"Cluster analysis; Multivariate statistics; Computer science; Data mining; Fuzzy clustering; Bayesian probability; CURE data clustering algorithm; Correlation clustering; Consensus clustering; Clustering high-dimensional data; Artificial intelligence; Machine learning","score_opus":0.09591034766605583,"score_gpt":0.3992369993620657,"score_spread":0.30332665169600986,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3205527824","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0059348624,0.00035173504,0.99239844,0.00032945257,0.000046362547,0.00008998275,0.00022289473,0.00021970815,0.00040652795],"genre_scores_gemma":[0.3991307,0.0016312907,0.5869797,0.000558534,0.00035341163,0.0012945895,0.0031708053,0.0004163563,0.0064646318],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9927695,0.00429015,0.0003280483,0.0015403838,0.0007750028,0.00029684612],"domain_scores_gemma":[0.9757474,0.017283127,0.0021097003,0.0017114782,0.0026668424,0.00048146458],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01884421,0.0015274915,0.0029510495,0.0031759753,0.001429759,0.0018318502,0.004334168,0.002696362,0.0032757104],"category_scores_gemma":[0.04173208,0.001336348,0.002427616,0.0027400714,0.002337333,0.0026826283,0.0023999745,0.003526403,0.001147454],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00014410376,0.00004425848,0.0027005342,0.00016265512,0.0002252051,0.000113455775,0.00023273042,0.8903609,0.0004779364,0.07265988,0.0021960242,0.030682385],"study_design_scores_gemma":[0.000012094567,0.000014784877,0.00030636863,0.00001550416,0.000013531612,0.000014841228,0.00001953341,0.96615434,0.00011370211,0.032658014,0.00065952406,0.000017806691],"about_ca_topic_score_codex":0.02021977,"about_ca_topic_score_gemma":0.015314622,"teacher_disagreement_score":0.02021977,"about_ca_system_score_codex":0.003248726,"about_ca_system_score_gemma":0.002990596,"threshold_uncertainty_score":0.09965885},"labels":[],"label_agreement":null},{"id":"W3210129212","doi":"10.1002/sim.9246","title":"Weighted generalized estimating equations and unified estimation for longitudinal data with nonmonotone missing data patterns","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Regina","funders":"","keywords":"Missing data; Estimator; Covariate; Generalized estimating equation; Estimating equations; Computer science; Statistics; Mathematics","score_opus":0.22131465097394634,"score_gpt":0.4628701982297937,"score_spread":0.24155554725584735,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3210129212","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0008741748,0.0006995396,0.99780566,0.00019508155,0.000048873146,0.00005374329,0.0001092495,0.00007170233,0.0001419652],"genre_scores_gemma":[0.046663456,0.0036927017,0.94459677,0.0004220287,0.00031620762,0.0015490813,0.0011601098,0.00011169868,0.0014879899],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9514809,0.040320836,0.0019639065,0.0030816938,0.002554948,0.00059764896],"domain_scores_gemma":[0.95122015,0.03682136,0.004589047,0.0045862645,0.0024878245,0.0002952893],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.044447694,0.0024575666,0.003971436,0.0041849297,0.00082566927,0.0021575256,0.0057034083,0.0027758584,0.0033966787],"category_scores_gemma":[0.102370806,0.0023055223,0.0059348573,0.006721614,0.001984996,0.0048003537,0.0040424294,0.005153259,0.0010920417],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00010011088,0.000098207325,0.0072408784,0.00096977275,0.0024630704,0.0004417055,0.00058656425,0.11823199,0.00054687023,0.6715618,0.005769731,0.1919893],"study_design_scores_gemma":[0.000086982946,0.00015243796,0.0020781416,0.00029773198,0.00042470996,0.00024534424,0.00012434744,0.38496497,0.0004695035,0.6003488,0.010687654,0.00011935749],"about_ca_topic_score_codex":0.0062127663,"about_ca_topic_score_gemma":0.006768411,"teacher_disagreement_score":0.044447694,"about_ca_system_score_codex":0.0016593483,"about_ca_system_score_gemma":0.003422883,"threshold_uncertainty_score":0.23506469},"labels":[],"label_agreement":null},{"id":"W3212678746","doi":"10.1002/sim.9233","title":"Estimation of standard deviations and inverse‐variance weights from an observed range","year":2021,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Estimator; Statistics; Standard deviation; Standard error; Contrast (vision); Variance (accounting); Mathematics; Sample size determination; Inverse; Range (aeronautics); Mean squared error; Population variance; Sample (material); Econometrics; Population; Computer science; Medicine","score_opus":0.6336011764395778,"score_gpt":0.6039632280292192,"score_spread":0.029637948410358605,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3212678746","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0030964476,0.19947842,0.7854878,0.0022388618,0.0012978036,0.00070781686,0.0008693498,0.0003361091,0.0064874277],"genre_scores_gemma":[0.11416852,0.26381811,0.60833126,0.0017410009,0.00179708,0.0050199362,0.001678556,0.00040345782,0.0030421154],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9453702,0.03900313,0.004528548,0.0029246218,0.007880368,0.00029320025],"domain_scores_gemma":[0.8894318,0.09596554,0.006274427,0.0044411803,0.0037033276,0.00018369843],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.052204236,0.0018054174,0.0038202403,0.0078124506,0.0003183757,0.0033966382,0.0030680133,0.002085483,0.003985348],"category_scores_gemma":[0.17992164,0.0010289086,0.0037629213,0.007671818,0.0023818451,0.0038812454,0.002027804,0.003448776,0.0013269129],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002715338,0.000060529626,0.0026511173,0.021540916,0.004289879,0.00007059697,0.0003442908,0.017290194,0.00061193627,0.15395425,0.008004192,0.7909105],"study_design_scores_gemma":[0.00063724426,0.0007499867,0.0113311,0.027869537,0.0072140093,0.0010698531,0.00042850623,0.056938704,0.0056752455,0.67231905,0.21532044,0.00044637534],"about_ca_topic_score_codex":0.0011049854,"about_ca_topic_score_gemma":0.00096292875,"teacher_disagreement_score":0.94779575,"about_ca_system_score_codex":0.0014035358,"about_ca_system_score_gemma":0.0027884925,"threshold_uncertainty_score":0.27608567},"labels":[],"label_agreement":null},{"id":"W3216992363","doi":"10.1002/sim.9259","title":"Using fractional polynomials and restricted cubic splines to model non‐proportional hazards or time‐varying covariate effects in the Cox regression model","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":94,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University Health Network; Institute for Clinical Evaluative Sciences; Sunnybrook Hospital; University of Toronto","funders":"Canadian Institutes of Health Research; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Covariate; Proportional hazards model; Mathematics; Statistics; Regression; Regression analysis; Applied mathematics; Econometrics","score_opus":0.16597718893485655,"score_gpt":0.48343134940256793,"score_spread":0.31745416046771135,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3216992363","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.006809481,0.00043218627,0.99087393,0.0003271142,0.0001255136,0.00007490458,0.00022981537,0.0002644639,0.000862581],"genre_scores_gemma":[0.28010976,0.0020199167,0.70768887,0.0004401116,0.00033354523,0.0010020237,0.00081065873,0.0004792155,0.007115842],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9937185,0.004268976,0.00023146479,0.000506312,0.00084129535,0.0004333687],"domain_scores_gemma":[0.9872228,0.009943498,0.0011303467,0.00092370744,0.00058915163,0.00019050512],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011729181,0.0009902084,0.0010854101,0.0016088493,0.00053943647,0.0012927166,0.0017420108,0.0010344633,0.00424082],"category_scores_gemma":[0.030424748,0.00057183433,0.0025941199,0.0029405863,0.0011683996,0.0013733564,0.0018201771,0.0033546998,0.0008030097],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005292408,0.00018653943,0.013344208,0.0005462308,0.00049994484,0.0007905031,0.0008872795,0.5019093,0.0018852605,0.264685,0.01222819,0.20250835],"study_design_scores_gemma":[0.00012789997,0.0002496339,0.003195401,0.00012863883,0.00013178351,0.0003686617,0.00014216294,0.81223094,0.0007402039,0.1633354,0.019226076,0.00012324608],"about_ca_topic_score_codex":0.013848467,"about_ca_topic_score_gemma":0.012221049,"teacher_disagreement_score":0.013848467,"about_ca_system_score_codex":0.001170849,"about_ca_system_score_gemma":0.003096545,"threshold_uncertainty_score":0.062030554},"labels":[],"label_agreement":null},{"id":"W4200422816","doi":"10.1002/sim.9270","title":"Methods to assess evidence consistency in <scp>dose‐response</scp> model based network <scp>meta‐analysis</scp>","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Mental Health Research Topics","field":"Psychology","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada; Department of Health and Social Care; University of Bristol; National Institute for Health and Care Research; University Hospitals Bristol NHS Foundation Trust; Pfizer","keywords":"Consistency (knowledge bases); Function (biology); Meta-analysis; Computer science; Node (physics); Mathematics; Medicine; Artificial intelligence; Biology","score_opus":0.34084893848496445,"score_gpt":0.5589487149081261,"score_spread":0.2180997764231617,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4200422816","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0067878347,0.016162978,0.9524344,0.0022282745,0.0006140772,0.008967687,0.0067383815,0.0025425497,0.003523723],"genre_scores_gemma":[0.20842348,0.0033301394,0.74143547,0.0016707665,0.0002913795,0.0388321,0.003990116,0.0007653282,0.001261298],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6629129,0.28802827,0.021287952,0.010062143,0.017205587,0.0005031902],"domain_scores_gemma":[0.31729662,0.61924374,0.01973595,0.034996863,0.008112405,0.00061442197],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.23078097,0.002793561,0.006054519,0.010383765,0.00092135725,0.005584957,0.0051019937,0.0038538456,0.011914915],"category_scores_gemma":[0.53690875,0.0021253019,0.016174719,0.009237664,0.002266786,0.0024898755,0.0042280867,0.0043369387,0.0011726136],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00674751,0.00033463739,0.023919547,0.08626381,0.30503523,0.0010510185,0.0005996678,0.14383908,0.0037835476,0.069901995,0.026420789,0.33210313],"study_design_scores_gemma":[0.007421458,0.0024908618,0.019913312,0.018888498,0.11788314,0.0010992246,0.00027121115,0.4227364,0.007768568,0.34876335,0.052055404,0.0007085996],"about_ca_topic_score_codex":0.002958173,"about_ca_topic_score_gemma":0.0038203022,"teacher_disagreement_score":0.76921904,"about_ca_system_score_codex":0.0030253602,"about_ca_system_score_gemma":0.00489816,"threshold_uncertainty_score":0.948584},"labels":[],"label_agreement":null},{"id":"W4200447663","doi":"10.1002/sim.9285","title":"Ranked set sampling in finite populations with bivariate responses: An application to an osteoporosis study","year":2021,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Manitoba","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Bivariate analysis; Statistics; Univariate; Estimator; Mathematics; Sampling (signal processing); Sampling design; Poisson distribution; Sample size determination; Poisson regression; Population; Econometrics; Multivariate statistics; Computer science; Medicine","score_opus":0.2533609540721423,"score_gpt":0.5232595687678369,"score_spread":0.26989861469569454,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4200447663","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029279584,0.0002901508,0.9680873,0.00035992014,0.000043720243,0.00045894017,0.00008925788,0.000079915226,0.001311351],"genre_scores_gemma":[0.3018968,0.00045993685,0.6944711,0.00027528105,0.000061246945,0.0015795403,0.0001609058,0.000027673848,0.0010674356],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9218191,0.072840385,0.00068961555,0.0011011558,0.0031349736,0.00041480822],"domain_scores_gemma":[0.8216785,0.15983078,0.0043633184,0.008492505,0.0048072548,0.0008275548],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.059512936,0.0007367981,0.0013698414,0.0015652141,0.00088725274,0.0011472496,0.0018105194,0.0014514046,0.0029777607],"category_scores_gemma":[0.12917961,0.0004983508,0.0016606062,0.0022598978,0.0024011116,0.0009934499,0.0021239917,0.0019767042,0.00028826628],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010647958,0.000630229,0.021619096,0.0008622585,0.0005472421,0.0009299554,0.0020592816,0.19577663,0.0024185565,0.55846417,0.002151708,0.2134761],"study_design_scores_gemma":[0.0005990803,0.0017382806,0.0065392554,0.0002085339,0.00019583876,0.00040371917,0.00061069976,0.6741159,0.002064974,0.30753085,0.005887586,0.00010529283],"about_ca_topic_score_codex":0.0044384194,"about_ca_topic_score_gemma":0.00446635,"teacher_disagreement_score":0.059512936,"about_ca_system_score_codex":0.0015934061,"about_ca_system_score_gemma":0.002551228,"threshold_uncertainty_score":0.3147382},"labels":[],"label_agreement":null},{"id":"W4200631872","doi":"10.1002/sim.9439","title":"Using principal stratification in analysis of clinical trials","year":2022,"lang":"en","type":"preprint","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Bayer (Canada)","funders":"","keywords":"Principal (computer security); Computer science; Variety (cybernetics); Observational study; Set (abstract data type); Addendum; Key (lock); Data science; Stratification (seeds); Management science; Machine learning; Risk analysis (engineering); Artificial intelligence; Mathematics; Engineering; Statistics; Medicine; Programming language; Biology","score_opus":0.9143698492728977,"score_gpt":0.7492055652700645,"score_spread":0.16516428400283312,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4200631872","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011571592,0.0011415258,0.9941268,0.0011858416,0.0002553212,0.00036121937,0.0002630098,0.00058092305,0.0009281858],"genre_scores_gemma":[0.06633434,0.0022634196,0.92348844,0.0010723216,0.00092327775,0.0039363066,0.0007278374,0.0004023181,0.0008517635],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.79089683,0.18628162,0.00660211,0.006700332,0.008401618,0.0011175476],"domain_scores_gemma":[0.6662491,0.28380015,0.013661325,0.027409922,0.007608581,0.0012709111],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.13373226,0.0023372665,0.0033844314,0.0063501704,0.0013499928,0.0063582477,0.0022163095,0.0028097779,0.0059645483],"category_scores_gemma":[0.3669349,0.0019564722,0.0049298992,0.0060312147,0.0059622303,0.005829196,0.0061442354,0.007146493,0.0023271723],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00064942153,0.000079280995,0.007251163,0.0022544158,0.0020986188,0.00022935476,0.0012197552,0.051795825,0.0005462824,0.6771648,0.016058307,0.24065275],"study_design_scores_gemma":[0.00022228817,0.00015049754,0.00085258146,0.00039805533,0.00018505035,0.00007265602,0.00005268058,0.07685173,0.0005610946,0.90886503,0.011715709,0.000072530296],"about_ca_topic_score_codex":0.0026687153,"about_ca_topic_score_gemma":0.0019533955,"teacher_disagreement_score":0.86626774,"about_ca_system_score_codex":0.0032521992,"about_ca_system_score_gemma":0.008804154,"threshold_uncertainty_score":0.70725226},"labels":[],"label_agreement":null},{"id":"W4205106986","doi":"10.1002/sim.8607","title":"Issue Information","year":2021,"lang":"en","type":"paratext","venue":"Statistics in Medicine","topic":"Human auditory perception and evaluation","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"National Institutes of Health","keywords":"Computer science; Citation; Information retrieval; Library science","score_opus":0.023449971841725304,"score_gpt":0.3298981782937842,"score_spread":0.3064482064520589,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4205106986","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00023997553,0.0007395172,0.0027597197,0.004184224,0.013330168,0.0004943469,0.033538964,0.005747188,0.9389659],"genre_scores_gemma":[0.0006026672,0.0005021049,0.0011065544,0.0015335211,0.0021948377,0.000157025,0.013073642,0.0014330358,0.9793967],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99843353,0.00020824722,0.00012079626,0.00019700905,0.00090306764,0.00013741427],"domain_scores_gemma":[0.9907658,0.0018378622,0.00039823164,0.00076608674,0.0043718275,0.0018601578],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0018995942,0.0014715499,0.0015835667,0.0062035937,0.0012727894,0.007986914,0.001650347,0.0025588558,0.91248673],"category_scores_gemma":[0.015244462,0.0006738462,0.0010164222,0.004815782,0.0005663659,0.0029997046,0.0023110546,0.0018920818,0.88214207],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000010664726,0.000016405915,0.000037200803,0.00011457693,0.0000014682337,0.000007785569,0.0000052291584,0.000035241508,0.00006125105,0.0011491814,0.9704157,0.028145364],"study_design_scores_gemma":[0.000018637458,0.000013493291,0.00020508176,0.00011573645,0.0000022268482,0.000019000483,0.000013632633,0.00011143108,0.00007921429,0.002110303,0.9973047,0.0000064625174],"about_ca_topic_score_codex":0.0020141753,"about_ca_topic_score_gemma":0.004038212,"teacher_disagreement_score":0.08751327,"about_ca_system_score_codex":0.0012296905,"about_ca_system_score_gemma":0.003144828,"threshold_uncertainty_score":0.12482709},"labels":[],"label_agreement":null},{"id":"W4205333182","doi":"10.1002/sim.9320","title":"Profile‐likelihood Bayesian model averaging for two‐sample summary data Mendelian randomization in the presence of horizontal pleiotropy","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic Associations and Epidemiology","field":"Biochemistry, Genetics and Molecular Biology","cited_by":44,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council Canada; Medical Research Council; University of Bristol; University of Exeter; Bureau for Economic Growth, Education, and Environment, United States Agency for International Development","keywords":"Mendelian randomization; Posterior probability; Bayesian probability; Computer science; Pleiotropy; Econometrics; Sample size determination; Bayesian inference; Statistics; Mathematics; Artificial intelligence; Biology; Genetic variants; Genetics","score_opus":0.02507433636254932,"score_gpt":0.3259323980784495,"score_spread":0.3008580617159002,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4205333182","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0038206375,0.00012019698,0.9954224,0.000093306204,0.000013457372,0.0000473761,0.00007506858,0.00023703257,0.00017048899],"genre_scores_gemma":[0.19613755,0.00061108207,0.79886925,0.0002329901,0.00015688663,0.00075743214,0.00088502816,0.00039244856,0.0019573956],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98823303,0.009020362,0.0005028611,0.0011154397,0.0008159541,0.0003123387],"domain_scores_gemma":[0.94467294,0.04745343,0.002235586,0.0036149295,0.0015120357,0.00051095657],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.029215407,0.0011414797,0.0036941636,0.0019284878,0.00091966626,0.001908196,0.0041815094,0.0017682214,0.0034666078],"category_scores_gemma":[0.084671065,0.001311536,0.002545065,0.0025467493,0.002315775,0.002688915,0.0022716832,0.0026984198,0.0007584073],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023658795,0.000051832933,0.0028078116,0.0003074771,0.00046030147,0.0003075409,0.00024469892,0.69837505,0.0014787699,0.22887635,0.0018309171,0.06502265],"study_design_scores_gemma":[0.000035858182,0.00003650267,0.00035211272,0.000020787114,0.000034256733,0.000044350283,0.0000127491985,0.91922677,0.00034371566,0.0790127,0.00085260783,0.000027504202],"about_ca_topic_score_codex":0.007844347,"about_ca_topic_score_gemma":0.0067405635,"teacher_disagreement_score":0.029215407,"about_ca_system_score_codex":0.0016346889,"about_ca_system_score_gemma":0.0029416992,"threshold_uncertainty_score":0.1545077},"labels":[],"label_agreement":null},{"id":"W4206399182","doi":"10.1002/sim.8615","title":"Issue Information","year":2021,"lang":"en","type":"paratext","venue":"Statistics in Medicine","topic":"Human auditory perception and evaluation","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"National Institutes of Health","keywords":"Computer science; Citation; Information retrieval; World Wide Web","score_opus":0.023449971841725304,"score_gpt":0.3298981782937842,"score_spread":0.3064482064520589,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4206399182","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00023997553,0.0007395172,0.0027597197,0.004184224,0.013330168,0.0004943469,0.033538964,0.005747188,0.9389659],"genre_scores_gemma":[0.0006026672,0.0005021049,0.0011065544,0.0015335211,0.0021948377,0.000157025,0.013073642,0.0014330358,0.9793967],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99843353,0.00020824722,0.00012079626,0.00019700905,0.00090306764,0.00013741427],"domain_scores_gemma":[0.9907658,0.0018378622,0.00039823164,0.00076608674,0.0043718275,0.0018601578],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0018995942,0.0014715499,0.0015835667,0.0062035937,0.0012727894,0.007986914,0.001650347,0.0025588558,0.91248673],"category_scores_gemma":[0.015244462,0.0006738462,0.0010164222,0.004815782,0.0005663659,0.0029997046,0.0023110546,0.0018920818,0.88214207],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000010664726,0.000016405915,0.000037200803,0.00011457693,0.0000014682337,0.000007785569,0.0000052291584,0.000035241508,0.00006125105,0.0011491814,0.9704157,0.028145364],"study_design_scores_gemma":[0.000018637458,0.000013493291,0.00020508176,0.00011573645,0.0000022268482,0.000019000483,0.000013632633,0.00011143108,0.00007921429,0.002110303,0.9973047,0.0000064625174],"about_ca_topic_score_codex":0.0020141753,"about_ca_topic_score_gemma":0.004038212,"teacher_disagreement_score":0.08751327,"about_ca_system_score_codex":0.0012296905,"about_ca_system_score_gemma":0.003144828,"threshold_uncertainty_score":0.12482709},"labels":[],"label_agreement":null},{"id":"W4206839457","doi":"10.1002/sim.7888","title":"Issue Information","year":2019,"lang":"en","type":"paratext","venue":"Statistics in Medicine","topic":"Diverse Scientific and Economic Studies","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"National Institutes of Health","keywords":"Computer science; Statistics; Mathematics","score_opus":0.0312994404200627,"score_gpt":0.2620493222487607,"score_spread":0.23074988182869802,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4206839457","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0003471095,0.0011246237,0.0011790283,0.00791146,0.025987772,0.00056077034,0.05821612,0.002524764,0.9021482],"genre_scores_gemma":[0.0013727715,0.000717855,0.00044163683,0.002090268,0.0023901807,0.0001848555,0.018080432,0.0006974225,0.97402453],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9983228,0.00023773161,0.00014916784,0.00022164157,0.00084786187,0.00022086066],"domain_scores_gemma":[0.99165463,0.0014089635,0.00034711338,0.00086499285,0.0040083355,0.0017160336],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0018105186,0.0010803033,0.0014337028,0.0035348383,0.0013621754,0.005704831,0.001958533,0.0022807536,0.94294125],"category_scores_gemma":[0.015353859,0.0004924766,0.0009283933,0.0031268068,0.00045307385,0.0031361429,0.0021682077,0.00209056,0.87084365],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00002762384,0.000017055225,0.000038286893,0.0001363776,0.0000017823364,0.000012925173,0.000007248699,0.000018397655,0.00005110533,0.0009604327,0.9780391,0.020689659],"study_design_scores_gemma":[0.000013064673,0.000013232411,0.00015283222,0.00012880642,0.0000016676405,0.000022806717,0.000015902277,0.000032524502,0.000061443396,0.00064341025,0.9989108,0.0000035435635],"about_ca_topic_score_codex":0.001988834,"about_ca_topic_score_gemma":0.00403882,"teacher_disagreement_score":0.05705875,"about_ca_system_score_codex":0.0015312567,"about_ca_system_score_gemma":0.0027661,"threshold_uncertainty_score":0.0813874},"labels":[],"label_agreement":null},{"id":"W4206897392","doi":"10.1002/sim.9317","title":"Response adaptive intervention allocation in stepped‐wedge cluster randomized trials","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada","keywords":"Intervention (counseling); Randomized controlled trial; Cluster (spacecraft); Adaptability; Computer science; Medicine; Cluster randomised controlled trial; Research design; Risk analysis (engineering); Reliability engineering; Operations management; Statistics; Mathematics; Engineering; Nursing; Surgery","score_opus":0.5051541458839005,"score_gpt":0.5887984476152415,"score_spread":0.08364430173134096,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4206897392","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.018045394,0.0023879947,0.94824505,0.0019861779,0.0006851225,0.024697931,0.00026426488,0.00064921274,0.003038846],"genre_scores_gemma":[0.29669458,0.00070769613,0.6599796,0.0015586312,0.00019579718,0.039648727,0.0001936854,0.000116523464,0.00090477004],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.48619905,0.48825333,0.007825055,0.0076580313,0.008781672,0.0012829474],"domain_scores_gemma":[0.63034725,0.32977638,0.015920814,0.015203957,0.006487255,0.0022643618],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.28737393,0.0027886797,0.0045970045,0.002852636,0.00084646855,0.002702736,0.0058517717,0.0043551847,0.0074090427],"category_scores_gemma":[0.34494877,0.001961922,0.004311156,0.0022848013,0.0055874367,0.0036645562,0.0040747076,0.0053680274,0.0011694875],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.033326514,0.00218177,0.006423221,0.015770271,0.0067333863,0.00049296534,0.0023520808,0.34716383,0.0014852916,0.31606653,0.007069179,0.2609349],"study_design_scores_gemma":[0.029386055,0.011573806,0.0014726878,0.003096606,0.0015997458,0.00023592584,0.0002595771,0.5956557,0.0026914056,0.34263682,0.01110195,0.00028972144],"about_ca_topic_score_codex":0.0009384844,"about_ca_topic_score_gemma":0.00065806624,"teacher_disagreement_score":0.7126261,"about_ca_system_score_codex":0.0040724967,"about_ca_system_score_gemma":0.0073661357,"threshold_uncertainty_score":0.8787948},"labels":[],"label_agreement":null},{"id":"W4210609813","doi":"10.1002/sim.9318","title":"Preserving data privacy when using multi‐site data to estimate individualized treatment rules","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Fonds de Recherche du Québec - Santé; Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Computer science; Pooling; Robustness (evolution); Data mining; Covariate; Regression; Context (archaeology); Confidentiality; Machine learning; Artificial intelligence; Statistics; Mathematics","score_opus":0.5117738302480719,"score_gpt":0.5546970299518682,"score_spread":0.04292319970379632,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4210609813","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.045095135,0.00033861274,0.9516599,0.0010122779,0.00003470954,0.00012035752,0.00051921955,0.00025030083,0.0009694781],"genre_scores_gemma":[0.72153425,0.00029764776,0.27619377,0.000342597,0.00007863005,0.00026298,0.0006470804,0.00007101893,0.0005720276],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9720167,0.021208564,0.0011274562,0.002433889,0.002688613,0.0005247675],"domain_scores_gemma":[0.85347277,0.10833555,0.008694361,0.026577875,0.0024648649,0.00045459997],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04267474,0.000646438,0.0013947091,0.0013115897,0.0009758515,0.002829742,0.0022708531,0.001725694,0.0009988273],"category_scores_gemma":[0.11094617,0.0007514242,0.0016451824,0.0025331962,0.0026369118,0.003934979,0.002540143,0.0026481282,0.0002625996],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009202228,0.00016297655,0.037868362,0.0003285237,0.00075268303,0.00063819165,0.00095213,0.6666054,0.0027075007,0.095445246,0.0022394056,0.19137923],"study_design_scores_gemma":[0.00012434236,0.0002113582,0.0060612494,0.000077144694,0.00013257387,0.00041300472,0.0002897931,0.77431506,0.004772721,0.20977817,0.0037604354,0.00006405121],"about_ca_topic_score_codex":0.0030921632,"about_ca_topic_score_gemma":0.0024643675,"teacher_disagreement_score":0.9573253,"about_ca_system_score_codex":0.0010791483,"about_ca_system_score_gemma":0.0019437326,"threshold_uncertainty_score":0.22568828},"labels":[],"label_agreement":null},{"id":"W4214682600","doi":"10.1002/sim.8233","title":"Issue Information","year":2020,"lang":"en","type":"paratext","venue":"Statistics in Medicine","topic":"Human auditory perception and evaluation","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Citation; Information retrieval; Library science","score_opus":0.026466283137171996,"score_gpt":0.32556839600590054,"score_spread":0.29910211286872856,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4214682600","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00023056837,0.0008246917,0.002584563,0.004398188,0.014554502,0.0005144441,0.030854395,0.0045710616,0.9414676],"genre_scores_gemma":[0.00051349937,0.00047552335,0.0009439818,0.0014811117,0.0016859153,0.00014168433,0.011311654,0.00092556555,0.9825211],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99824274,0.00023613541,0.0001278102,0.00021511903,0.0010209094,0.0001572599],"domain_scores_gemma":[0.9893885,0.0016811463,0.00042114034,0.0007590908,0.0054471837,0.0023029484],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0023441457,0.0014101088,0.0014623948,0.006028374,0.0012730561,0.008461499,0.0016874053,0.0029332065,0.8969989],"category_scores_gemma":[0.016923161,0.0006249296,0.0010215153,0.0041817417,0.0006304504,0.0030057887,0.0023121862,0.0019864985,0.8633027],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000011573213,0.000015959036,0.00003907733,0.00011789477,0.0000013070835,0.000007143471,0.0000050266926,0.000040134368,0.000061479084,0.0016063204,0.9674511,0.030642886],"study_design_scores_gemma":[0.000016382757,0.00001345445,0.00019173003,0.00011692389,0.000001767309,0.000015508293,0.000012067418,0.00010773989,0.000068495334,0.0023038485,0.9971462,0.0000058368355],"about_ca_topic_score_codex":0.0027643365,"about_ca_topic_score_gemma":0.005063108,"teacher_disagreement_score":0.10300112,"about_ca_system_score_codex":0.0015374128,"about_ca_system_score_gemma":0.0040387623,"threshold_uncertainty_score":0.14691854},"labels":[],"label_agreement":null},{"id":"W4220708167","doi":"10.1002/sim.9049","title":"Issue Information","year":2022,"lang":"en","type":"paratext","venue":"Statistics in Medicine","topic":"Human auditory perception and evaluation","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science","score_opus":0.022882675844517697,"score_gpt":0.3267985601856628,"score_spread":0.3039158843411451,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4220708167","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00024921866,0.00077700755,0.0028531428,0.0043589496,0.014159214,0.0005107679,0.036412165,0.0060264785,0.9346531],"genre_scores_gemma":[0.0006420745,0.0005271655,0.0011808183,0.0016672509,0.0023203047,0.00016631401,0.014081275,0.0015391092,0.9778758],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99849725,0.0001984397,0.00011663798,0.00019369721,0.00086021953,0.00013369716],"domain_scores_gemma":[0.99097097,0.0017903633,0.00039778074,0.00076131127,0.0042254976,0.0018540877],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0018688705,0.0014550623,0.0015597296,0.005969684,0.0012297829,0.00770422,0.001646639,0.0024317398,0.91504025],"category_scores_gemma":[0.015052027,0.0006651357,0.0010112696,0.0046658134,0.00055092503,0.0028892732,0.0022885327,0.0018761373,0.88225067],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000011118785,0.00001698641,0.00003801092,0.00011723071,0.000001532071,0.000007755015,0.000005204306,0.000033959976,0.000063316715,0.0011029323,0.9712478,0.027354147],"study_design_scores_gemma":[0.000019103516,0.000013457484,0.00020765614,0.00011331262,0.0000023295872,0.000018876937,0.000013604057,0.000104489176,0.000079811885,0.002062335,0.9973586,0.000006374355],"about_ca_topic_score_codex":0.0019144753,"about_ca_topic_score_gemma":0.003922408,"teacher_disagreement_score":0.084959745,"about_ca_system_score_codex":0.00115201,"about_ca_system_score_gemma":0.003072405,"threshold_uncertainty_score":0.12118471},"labels":[],"label_agreement":null},{"id":"W4220757114","doi":"10.1002/sim.9367","title":"A mixture distribution approach for assessing genetic impact from twin study","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Statistics; Restricted maximum likelihood; Estimator; Bivariate analysis; Consistency (knowledge bases); Inference; Dizygotic twins; Twin study; Econometrics; Correlation; Monozygotic twin; Genetic correlation; Mathematics; Maximum likelihood; Computer science; Heritability; Biology; Genetics; Genetic variation; Artificial intelligence; Medicine","score_opus":0.030329456089655076,"score_gpt":0.36346358413950375,"score_spread":0.33313412804984865,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4220757114","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.006330075,0.00047491703,0.99200237,0.00015333953,0.000039835748,0.00013545789,0.00016526978,0.0001659356,0.0005327576],"genre_scores_gemma":[0.17657842,0.0009985836,0.8176766,0.00025715408,0.00015359283,0.0011917829,0.00094500853,0.00021638899,0.001982489],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97744435,0.017416902,0.00072883366,0.0023291046,0.0017685771,0.00031234097],"domain_scores_gemma":[0.9524008,0.039880104,0.0018277025,0.003691807,0.0016804106,0.0005191968],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03988807,0.0015324216,0.0027779154,0.0061553633,0.001245033,0.002566742,0.0036405968,0.0026797568,0.004699947],"category_scores_gemma":[0.0977244,0.0015201276,0.0033550076,0.004685537,0.002301508,0.0033361414,0.0038544112,0.0036185412,0.0007527561],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010558247,0.0002554065,0.06710983,0.000766785,0.003717097,0.001406091,0.0016497923,0.15080334,0.0035301014,0.39773557,0.0055053476,0.36646488],"study_design_scores_gemma":[0.00012220624,0.00019342412,0.01078847,0.00021940934,0.0007127061,0.0008356858,0.00019288181,0.5788713,0.0010709072,0.39963654,0.0072273454,0.00012913068],"about_ca_topic_score_codex":0.006836809,"about_ca_topic_score_gemma":0.005167459,"teacher_disagreement_score":0.03988807,"about_ca_system_score_codex":0.0012838274,"about_ca_system_score_gemma":0.0021729057,"threshold_uncertainty_score":0.21095079},"labels":[],"label_agreement":null},{"id":"W4220980829","doi":"10.1002/sim.9299","title":"Sensitivity to missing not at random dropout in clinical trials: Use and interpretation of the trimmed means estimator","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Health Technology Assessment Programme; Medical Research Council; Medical Research Council Canada; University of Exeter; University of Bristol; Department of Health and Social Care; Elizabeth Blackwell Institute for Health Research, University of Bristol; University Hospitals Bristol NHS Foundation Trust; UK Research and Innovation; National Institute for Health and Care Research; Cancer Research UK; Wellcome Trust","keywords":"Estimator; Missing data; Statistics; Dropout (neural networks); Randomized controlled trial; Mathematics; Econometrics; Sensitivity (control systems); Imputation (statistics); Medicine; Computer science; Internal medicine","score_opus":0.20896445576742714,"score_gpt":0.5020260976696712,"score_spread":0.29306164190224404,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4220980829","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0151809845,0.020912103,0.9445459,0.009851051,0.0012802253,0.0026184001,0.00062190025,0.0008203754,0.0041690003],"genre_scores_gemma":[0.55331796,0.00610721,0.41337976,0.012709215,0.0015557407,0.010720538,0.0005708263,0.00040866484,0.0012300614],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.29523873,0.6481547,0.027460275,0.012823747,0.01500236,0.0013202413],"domain_scores_gemma":[0.09392046,0.83851624,0.03589512,0.02560438,0.0055774446,0.00048634253],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.56198114,0.0024872725,0.00773687,0.0072497115,0.0015523747,0.006956242,0.0063810186,0.007943279,0.0040023723],"category_scores_gemma":[0.8416746,0.00253669,0.0099516995,0.0064010397,0.0063839485,0.0072929775,0.008388259,0.009553875,0.0005923807],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0057388037,0.00027628458,0.043484244,0.026628071,0.040924106,0.0024279947,0.00759145,0.12453506,0.0013031926,0.32061446,0.012606359,0.4138699],"study_design_scores_gemma":[0.0019134772,0.0016999142,0.0127894655,0.01155319,0.012306906,0.0020658313,0.0006319156,0.18663791,0.003539562,0.7475809,0.018617837,0.00066300936],"about_ca_topic_score_codex":0.0023373899,"about_ca_topic_score_gemma":0.0015628084,"teacher_disagreement_score":0.43801886,"about_ca_system_score_codex":0.005291652,"about_ca_system_score_gemma":0.0048993463,"threshold_uncertainty_score":0.5401553},"labels":[],"label_agreement":null},{"id":"W4221004841","doi":"10.1002/sim.9364","title":"Assessing complier average causal effects from longitudinal trials with multiple endpoints and treatment noncompliance: An application to a study of Arthritis Health Journal","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; Research Canada; Simon Fraser University","funders":"National Cancer Institute; Social Sciences and Humanities Research Council of Canada; Natural Sciences and Engineering Research Council of Canada; National Institutes of Health","keywords":"Econometrics; Medicine; Clinical trial; Arthritis; Internal medicine; Economics","score_opus":0.19042324253128365,"score_gpt":0.49042503604727183,"score_spread":0.3000017935159882,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4221004841","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01780595,0.0026441864,0.97403324,0.0024609193,0.00017037717,0.0012114317,0.00040748043,0.00027095852,0.0009955368],"genre_scores_gemma":[0.34415865,0.0022637444,0.6441267,0.0016655963,0.0004469577,0.0050215945,0.0006243753,0.00013418632,0.0015581511],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.7610776,0.2166703,0.005924141,0.008436138,0.0070414613,0.0008504051],"domain_scores_gemma":[0.2528913,0.7010043,0.021125507,0.019258754,0.004668809,0.0010513414],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2528539,0.0024089026,0.0051779267,0.0047755647,0.0013972197,0.003457586,0.0035827975,0.0044706143,0.006818573],"category_scores_gemma":[0.5138026,0.0014205755,0.009411936,0.0059006796,0.004618411,0.0048040953,0.0045850193,0.0071284785,0.00043212372],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0032619222,0.0007034911,0.07984366,0.0045060394,0.017157413,0.002117036,0.002834792,0.20212203,0.0014539876,0.37234777,0.0056386385,0.30801314],"study_design_scores_gemma":[0.0011306929,0.0015493693,0.013196445,0.0007307184,0.0028973494,0.00067097135,0.00021154294,0.550816,0.000831735,0.42107803,0.0067150164,0.00017210987],"about_ca_topic_score_codex":0.0048325183,"about_ca_topic_score_gemma":0.0036787852,"teacher_disagreement_score":0.7471461,"about_ca_system_score_codex":0.0024618078,"about_ca_system_score_gemma":0.0056141666,"threshold_uncertainty_score":0.9213642},"labels":[],"label_agreement":null},{"id":"W4221129397","doi":"10.1002/sim.9386","title":"Multiply robust subgroup analysis based on a single‐index threshold linear marginal model for longitudinal data with dropouts","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"National Natural Science Foundation of China","keywords":"Longitudinal data; Statistics; Index (typography); Mathematics; Single-index model; Linear model; Marginal model; Subgroup analysis; Econometrics; Computer science; Applied mathematics; Regression analysis; Data mining","score_opus":0.22592500866140303,"score_gpt":0.4120618597321267,"score_spread":0.1861368510707237,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4221129397","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017033424,0.00030191094,0.9818138,0.0002068243,0.000036849135,0.00016743569,0.000119354874,0.0001686112,0.00015179068],"genre_scores_gemma":[0.5109439,0.00085964706,0.48181963,0.000410046,0.00021420888,0.0017488064,0.0011403378,0.00020240061,0.0026609893],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99189866,0.005885565,0.0003173418,0.0010201636,0.00057497737,0.00030317463],"domain_scores_gemma":[0.9836646,0.012042169,0.0010449789,0.002193492,0.0008218573,0.00023294915],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.022511754,0.0010173024,0.0027716667,0.0016998452,0.00058001844,0.0012755856,0.0029300707,0.0015290922,0.002334083],"category_scores_gemma":[0.047294464,0.00058251596,0.0038160726,0.0016575238,0.0013096274,0.0017182283,0.0024499658,0.002341492,0.0003399772],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015812708,0.00040561854,0.03437774,0.00096948765,0.0034006361,0.0012220623,0.001321457,0.44571033,0.0065940847,0.17640851,0.0041555506,0.32385322],"study_design_scores_gemma":[0.00010687542,0.00027634084,0.0025728166,0.000039680395,0.0003831459,0.0000964224,0.00007561276,0.9326448,0.00082458364,0.061250035,0.0016877563,0.000041927244],"about_ca_topic_score_codex":0.0042306515,"about_ca_topic_score_gemma":0.0035703292,"teacher_disagreement_score":0.022511754,"about_ca_system_score_codex":0.0008926534,"about_ca_system_score_gemma":0.0024572918,"threshold_uncertainty_score":0.11905491},"labels":[],"label_agreement":null},{"id":"W4224718928","doi":"10.1002/sim.9417","title":"Variable selection in semiparametric regression models for longitudinal data with informative observation times","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada","keywords":"Semiparametric regression; Covariate; Estimator; Regression; Semiparametric model; Regression analysis; Statistics; Feature selection; Econometrics; Outcome (game theory); Flexibility (engineering); Computer science; Mathematics; Artificial intelligence","score_opus":0.20494563651052067,"score_gpt":0.43039976701503624,"score_spread":0.22545413050451557,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4224718928","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005163982,0.00034477998,0.99355423,0.0003139232,0.000026892776,0.000063824766,0.00012915871,0.00015728158,0.00024605944],"genre_scores_gemma":[0.39469182,0.002442165,0.5919689,0.00079766463,0.0004041331,0.002285995,0.001976924,0.00037029217,0.005062108],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9697955,0.025972482,0.00075274444,0.0017462769,0.0012499964,0.00048304867],"domain_scores_gemma":[0.8872856,0.1003639,0.004781847,0.0048428504,0.0022016359,0.0005241663],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03536799,0.0013732022,0.002981102,0.002005991,0.00076122826,0.0022366885,0.0033884242,0.0018894658,0.0037134225],"category_scores_gemma":[0.1015017,0.0013644301,0.0028998223,0.0030085822,0.0030469254,0.0032767134,0.0028258397,0.003656885,0.0009308015],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00039741246,0.00016803286,0.008871399,0.00085259206,0.0007624277,0.00071647094,0.00069434993,0.40472645,0.0013388529,0.46840593,0.0048069325,0.10825921],"study_design_scores_gemma":[0.00008282312,0.0000987083,0.0011766147,0.000071264796,0.00008863881,0.0001227482,0.000050938623,0.79614997,0.00044732806,0.19956578,0.0020978074,0.00004734816],"about_ca_topic_score_codex":0.002990896,"about_ca_topic_score_gemma":0.0027834426,"teacher_disagreement_score":0.03536799,"about_ca_system_score_codex":0.0011275483,"about_ca_system_score_gemma":0.0021225086,"threshold_uncertainty_score":0.18704605},"labels":[],"label_agreement":null},{"id":"W4228997420","doi":"10.1002/sim.9429","title":"Analyzing cohort studies with interval‐censored data: A new model‐based linear rank‐type test","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; Centre Hospitalier Universitaire Sainte-Justine","funders":"","keywords":"Statistics; Nonparametric statistics; Test statistic; Interval (graph theory); Hazard ratio; Mathematics; Hazard; Confidence interval; Rank (graph theory); Statistical hypothesis testing; Computer science; Econometrics","score_opus":0.21045099987993063,"score_gpt":0.4656520468827665,"score_spread":0.25520104700283586,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4228997420","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0412037,0.0011844499,0.9535472,0.0007813136,0.00019863431,0.00028682884,0.0007990372,0.00044855004,0.0015503366],"genre_scores_gemma":[0.6821934,0.0006933815,0.3118292,0.00052943564,0.00053267693,0.0011206389,0.0011690306,0.00012820959,0.0018041069],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9773466,0.017428027,0.00085797376,0.001463665,0.0025592905,0.00034450003],"domain_scores_gemma":[0.89987457,0.08479033,0.0060340664,0.0042653633,0.004085671,0.0009500549],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.021647122,0.0006887935,0.0014079242,0.0037030408,0.00043654742,0.0018001485,0.0023642876,0.0013033833,0.0046235383],"category_scores_gemma":[0.08368155,0.00031592025,0.0016236076,0.0028102603,0.0013549615,0.001956339,0.0016971235,0.0019531716,0.0005822218],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002168949,0.0007216556,0.17912817,0.0022767738,0.0044451705,0.0016100785,0.00045847605,0.107325226,0.00417526,0.11593556,0.014227522,0.5675271],"study_design_scores_gemma":[0.00050338707,0.0033910114,0.02413309,0.00027833006,0.0007763816,0.0020515982,0.00037273654,0.8133311,0.0035127606,0.14027695,0.01114711,0.00022550205],"about_ca_topic_score_codex":0.0007820309,"about_ca_topic_score_gemma":0.00055535283,"teacher_disagreement_score":0.021647122,"about_ca_system_score_codex":0.0007334393,"about_ca_system_score_gemma":0.0018813163,"threshold_uncertainty_score":0.11448228},"labels":[],"label_agreement":null},{"id":"W4231258378","doi":"10.1002/sim.822.abs","title":"Simultaneous modelling of operative mortality and long‐term survival after coronary artery bypass surgery","year":2001,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Insurance, Mortality, Demography, Risk Management","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"BC Cancer Agency; Simon Fraser University; University of Alberta","funders":"","keywords":"Covariate; Poisson regression; Proportional hazards model; Coronary artery bypass surgery; Medicine; Survival analysis; Accelerated failure time model; Bypass surgery; Term (time); Regression analysis; Poisson distribution; Artery; Surgery; Cardiology; Statistics; Mathematics; Population","score_opus":0.05708445331418451,"score_gpt":0.3571871711612406,"score_spread":0.3001027178470561,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4231258378","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.64787215,0.0011270801,0.34152633,0.002774926,0.00012601305,0.00016296373,0.0017445447,0.00034581195,0.0043200576],"genre_scores_gemma":[0.9788134,0.000618999,0.012271924,0.000070227215,0.00006729229,0.00030897354,0.00075221295,0.000033994205,0.007062995],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9976699,0.0013076313,0.00010201314,0.00034282357,0.0002336356,0.00034398847],"domain_scores_gemma":[0.9907656,0.006756375,0.0012752538,0.00041812667,0.00035363386,0.00043110785],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0045238254,0.0009902229,0.0011542279,0.00087709504,0.00052160124,0.0017337614,0.0020885407,0.0021343802,0.0025601913],"category_scores_gemma":[0.017806783,0.00086059433,0.0019169716,0.0011569121,0.0014511393,0.0015767703,0.002970092,0.0019697149,0.0004828714],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007794752,0.00026325387,0.043585353,0.00019534133,0.00044045414,0.0007512375,0.0013706675,0.87615544,0.0021038298,0.04891614,0.000961933,0.024476852],"study_design_scores_gemma":[0.00007367542,0.00031605328,0.016664354,0.000035095673,0.00014384571,0.0001739881,0.0001557142,0.93893766,0.0005461068,0.041525133,0.0013821943,0.00004615888],"about_ca_topic_score_codex":0.008280212,"about_ca_topic_score_gemma":0.008521244,"teacher_disagreement_score":0.008280212,"about_ca_system_score_codex":0.0013585985,"about_ca_system_score_gemma":0.0022043858,"threshold_uncertainty_score":0.02392459},"labels":[],"label_agreement":null},{"id":"W4232296411","doi":"10.1002/sim.3571","title":"Correction","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Delphi Technique in Research","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hospital for Sick Children","funders":"","keywords":"Regret; Acknowledgement; Spirometry; Statistics; Psychology; Asthma; Medicine; Computer science; Mathematics; Internal medicine","score_opus":0.09469270781760097,"score_gpt":0.5091439785365464,"score_spread":0.4144512707189454,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4232296411","genre_codex":"editorial","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0008534826,0.002360512,0.00751634,0.1332229,0.69072956,0.0003578703,0.062411148,0.007040789,0.09550735],"genre_scores_gemma":[0.020084862,0.0045726434,0.012314696,0.050780304,0.089741796,0.00087439554,0.063288756,0.012151844,0.7461908],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9901698,0.0017651271,0.0015654797,0.0015551238,0.0041010343,0.00084325584],"domain_scores_gemma":[0.9323037,0.012241653,0.0026987211,0.010008588,0.040423434,0.0023238496],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0061256024,0.0016973021,0.0012607162,0.0077532916,0.0031150898,0.006443519,0.004013828,0.0035241519,0.4684044],"category_scores_gemma":[0.115481354,0.00076765224,0.0018265462,0.00524417,0.0017194387,0.0052427594,0.0039638137,0.0061443374,0.28429624],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000015034139,0.0000026266869,0.00008485508,0.00008643006,0.000004761866,0.00006745831,0.00004145358,0.000019452875,0.000022394372,0.0015148312,0.98843306,0.0097077],"study_design_scores_gemma":[0.000010043313,0.000004872769,0.00033233897,0.0001985308,0.0000072562093,0.00015929186,0.000060399514,0.000040105686,0.0000893289,0.0014665261,0.99761826,0.000012893949],"about_ca_topic_score_codex":0.009382427,"about_ca_topic_score_gemma":0.006752648,"teacher_disagreement_score":0.4684044,"about_ca_system_score_codex":0.004110999,"about_ca_system_score_gemma":0.0070830924,"threshold_uncertainty_score":0},"labels":[],"label_agreement":null},{"id":"W4232817067","doi":"10.1002/sim.8559","title":"Authors' response","year":2020,"lang":"en","type":"letter","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"","keywords":"Library science; Lien; The arts; Citation; Sociology; Political science; Law; Computer science","score_opus":0.37558666727407897,"score_gpt":0.47380113509376554,"score_spread":0.09821446781968657,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4232817067","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00016228437,0.00043957517,0.00008500762,0.934994,0.0603294,0.000036962214,0.000166698,0.00006779382,0.003718338],"genre_scores_gemma":[0.00069959025,0.00014630362,0.00008952952,0.9728123,0.013214885,0.0000634817,0.000032653876,0.000032813892,0.012908419],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98272127,0.003989755,0.0019157189,0.0020342134,0.006126805,0.0032122636],"domain_scores_gemma":[0.9480439,0.023426142,0.0038828205,0.0018726254,0.015905812,0.0068687685],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012055749,0.0010662011,0.002824592,0.0020320322,0.007367824,0.0108429035,0.0036436422,0.095832184,0.036354173],"category_scores_gemma":[0.10310053,0.0015105099,0.0028284597,0.0013024278,0.0042024716,0.0038060632,0.0051216413,0.06362048,0.034155186],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00002046327,0.000006262211,0.00010558335,0.000028753144,0.000006883179,0.00014585712,0.00007472415,0.00001752556,0.00003264475,0.0006726697,0.99768424,0.0012044924],"study_design_scores_gemma":[0.00006467479,0.000022877623,0.0005854672,0.00035522014,0.00002480651,0.00023529567,0.00045683043,0.00016136587,0.0001484985,0.0024196156,0.9954652,0.00006013057],"about_ca_topic_score_codex":0.013649253,"about_ca_topic_score_gemma":0.01708531,"teacher_disagreement_score":0.095832184,"about_ca_system_score_codex":0.010215339,"about_ca_system_score_gemma":0.014274331,"threshold_uncertainty_score":0.12161684},"labels":[],"label_agreement":null},{"id":"W4234921364","doi":"10.1002/sim.2781","title":"The performance of different propensity score methods for estimating marginal odds ratios","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":244,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Canadian Institutes of Health Research; Institute for Clinical Evaluative Sciences","keywords":"Propensity score matching; Covariate; Statistics; Odds ratio; Odds; Estimator; Matching (statistics); Medicine; Mathematics; Logistic regression","score_opus":0.16363719544178307,"score_gpt":0.47372662365848633,"score_spread":0.31008942821670327,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4234921364","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.018347679,0.0022331222,0.97727,0.00031145514,0.00008146587,0.00032955123,0.0002892025,0.00029465772,0.0008429358],"genre_scores_gemma":[0.20714252,0.0031940392,0.7859131,0.00023492113,0.00016446278,0.0012984221,0.0009474799,0.0002972546,0.00080776744],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.93211895,0.05553478,0.0032220695,0.0035285666,0.0052170325,0.00037856406],"domain_scores_gemma":[0.7956067,0.17431645,0.00991007,0.013201423,0.0063590617,0.0006063514],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07765374,0.0018424876,0.0019094661,0.005909416,0.00064507074,0.0024748715,0.0023781485,0.0018511235,0.0030297036],"category_scores_gemma":[0.26014405,0.00090888405,0.0033946729,0.0051962603,0.0014983638,0.0031809288,0.0023108942,0.0020831393,0.00073593814],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018392295,0.00036405367,0.058345526,0.0015995405,0.005712131,0.00022578861,0.00084325875,0.3148263,0.0022086555,0.120061964,0.0038391661,0.4901344],"study_design_scores_gemma":[0.0006096481,0.0006888587,0.02177115,0.0005251091,0.00084596477,0.0005154789,0.0001579737,0.83451545,0.004067289,0.12890494,0.007118532,0.00027951066],"about_ca_topic_score_codex":0.0029077344,"about_ca_topic_score_gemma":0.0023901716,"teacher_disagreement_score":0.92234623,"about_ca_system_score_codex":0.0015311678,"about_ca_system_score_gemma":0.0019736285,"threshold_uncertainty_score":0.41067713},"labels":[],"label_agreement":null},{"id":"W4240995029","doi":"10.1002/sim.8237","title":"Issue Information","year":2020,"lang":"en","type":"paratext","venue":"Statistics in Medicine","topic":"Human auditory perception and evaluation","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Information retrieval; Citation; Library science","score_opus":0.026466283137171996,"score_gpt":0.32556839600590054,"score_spread":0.29910211286872856,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4240995029","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00023056837,0.0008246917,0.002584563,0.004398188,0.014554502,0.0005144441,0.030854395,0.0045710616,0.9414676],"genre_scores_gemma":[0.00051349937,0.00047552335,0.0009439818,0.0014811117,0.0016859153,0.00014168433,0.011311654,0.00092556555,0.9825211],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99824274,0.00023613541,0.0001278102,0.00021511903,0.0010209094,0.0001572599],"domain_scores_gemma":[0.9893885,0.0016811463,0.00042114034,0.0007590908,0.0054471837,0.0023029484],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0023441457,0.0014101088,0.0014623948,0.006028374,0.0012730561,0.008461499,0.0016874053,0.0029332065,0.8969989],"category_scores_gemma":[0.016923161,0.0006249296,0.0010215153,0.0041817417,0.0006304504,0.0030057887,0.0023121862,0.0019864985,0.8633027],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000011573213,0.000015959036,0.00003907733,0.00011789477,0.0000013070835,0.000007143471,0.0000050266926,0.000040134368,0.000061479084,0.0016063204,0.9674511,0.030642886],"study_design_scores_gemma":[0.000016382757,0.00001345445,0.00019173003,0.00011692389,0.000001767309,0.000015508293,0.000012067418,0.00010773989,0.000068495334,0.0023038485,0.9971462,0.0000058368355],"about_ca_topic_score_codex":0.0027643365,"about_ca_topic_score_gemma":0.005063108,"teacher_disagreement_score":0.10300112,"about_ca_system_score_codex":0.0015374128,"about_ca_system_score_gemma":0.0040387623,"threshold_uncertainty_score":0.14691854},"labels":[],"label_agreement":null},{"id":"W4244582227","doi":"10.1002/sim.2880","title":"Authors' reply","year":2007,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Healthcare Policy and Management","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Biostatistics; Library science; Public health; Medical school; Health care; Medicine; Political science; Medical education; Nursing","score_opus":0.06658054735097424,"score_gpt":0.3524056193680195,"score_spread":0.28582507201704527,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4244582227","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00024148806,0.00088387134,0.00008743432,0.89076895,0.10669119,0.000013355855,0.00009425348,0.000037421007,0.0011821878],"genre_scores_gemma":[0.0020358744,0.0004069728,0.00011459495,0.95328665,0.038376234,0.00004212084,0.000032415133,0.000036835558,0.005668338],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98555017,0.0036899613,0.0021397495,0.0022685912,0.00411166,0.0022397924],"domain_scores_gemma":[0.93360895,0.037825115,0.004319333,0.0029281618,0.017797044,0.0035213619],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010948279,0.0008839677,0.002300306,0.0019899048,0.0048499643,0.008710116,0.0032099735,0.06370605,0.020610962],"category_scores_gemma":[0.12806477,0.0013585836,0.0024837188,0.0013524541,0.005851206,0.004671519,0.0046019102,0.050297346,0.013751947],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000033753855,0.000005590041,0.00022179841,0.00005941237,0.000023575723,0.0002580394,0.00020713633,0.00002583404,0.00006816543,0.0016174334,0.9957638,0.001715559],"study_design_scores_gemma":[0.00009735529,0.000023783768,0.000991456,0.0004461556,0.00008122091,0.00056816527,0.0009151473,0.00013493764,0.0004048266,0.0053457553,0.9909037,0.00008757369],"about_ca_topic_score_codex":0.0053022346,"about_ca_topic_score_gemma":0.004772069,"teacher_disagreement_score":0.06370605,"about_ca_system_score_codex":0.006575914,"about_ca_system_score_gemma":0.007847082,"threshold_uncertainty_score":0.068950534},"labels":[],"label_agreement":null},{"id":"W4245910571","doi":"10.1002/sim.6278","title":"Correction","year":2014,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Computer science; Smoothing; Multivariate statistics; Bayesian probability; Statistics; Data mining; B-spline; Artificial intelligence; Mathematics; Machine learning","score_opus":0.08967552808242954,"score_gpt":0.4643787968405947,"score_spread":0.3747032687581652,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4245910571","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00033922427,0.0014110884,0.007652214,0.12217315,0.8296323,0.00018133485,0.02059811,0.0038963144,0.01411614],"genre_scores_gemma":[0.022995425,0.006051948,0.030696377,0.17381412,0.28213924,0.0018663146,0.051728606,0.0147384135,0.41596952],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.97270566,0.0059872465,0.004605826,0.0039328565,0.010931025,0.0018373115],"domain_scores_gemma":[0.7656293,0.055522684,0.011538076,0.024859013,0.13696115,0.005489758],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017122513,0.0023698572,0.0025869035,0.008535796,0.0031517327,0.008213534,0.0058725863,0.006214701,0.30733246],"category_scores_gemma":[0.3202802,0.0015279774,0.0029107388,0.0067058518,0.0031240303,0.0057595205,0.0061874697,0.012770821,0.202708],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000026160553,0.0000034990048,0.000107229316,0.00012779747,0.0000096285,0.000054137057,0.00005013482,0.000027834061,0.000028758555,0.0012361355,0.99105877,0.0072699394],"study_design_scores_gemma":[0.000034165318,0.000012423662,0.00047661274,0.00050434767,0.000024031311,0.00026981265,0.00011009259,0.00013984714,0.00016917665,0.0035159364,0.9947044,0.000039302708],"about_ca_topic_score_codex":0.006275761,"about_ca_topic_score_gemma":0.007229186,"teacher_disagreement_score":0.30733246,"about_ca_system_score_codex":0.0044609155,"about_ca_system_score_gemma":0.009035342,"threshold_uncertainty_score":0},"labels":[],"label_agreement":null},{"id":"W4248104173","doi":"10.1002/sim.3486","title":"Correction","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"","keywords":"Inflation (cosmology); Sample size determination; Computer science; Statistics; Mathematics; Applied mathematics; Physics; Theoretical physics","score_opus":0.4513497466859385,"score_gpt":0.6047742542354313,"score_spread":0.15342450754949277,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4248104173","genre_codex":"editorial","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0005324126,0.0023959302,0.016544353,0.07556466,0.71085006,0.0008570305,0.14743528,0.00871234,0.037107892],"genre_scores_gemma":[0.021018535,0.005127487,0.04837759,0.062487535,0.15517206,0.004536424,0.14993127,0.028943488,0.52440566],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98138076,0.0054751453,0.0038250203,0.0018192442,0.006531534,0.0009683191],"domain_scores_gemma":[0.74760056,0.081678495,0.009712803,0.024853168,0.13223219,0.0039228094],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.016267164,0.0025172194,0.0022664561,0.010912179,0.0019393342,0.0051612216,0.004521923,0.0040013096,0.5334648],"category_scores_gemma":[0.30932072,0.001376117,0.0030576044,0.008275847,0.0014085806,0.0045712814,0.0035709976,0.00720701,0.26092094],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000028742632,0.0000022948943,0.00006352825,0.00021792806,0.000012351847,0.000025943758,0.000015327882,0.00003458615,0.000016339834,0.0006981773,0.99185985,0.007024912],"study_design_scores_gemma":[0.000076314864,0.000012958692,0.0007898374,0.00077644933,0.000044960463,0.00016333362,0.00004238138,0.00016313248,0.00016693276,0.0035072677,0.9942227,0.00003359457],"about_ca_topic_score_codex":0.009366445,"about_ca_topic_score_gemma":0.00812991,"teacher_disagreement_score":0.4665352,"about_ca_system_score_codex":0.0040112836,"about_ca_system_score_gemma":0.009017718,"threshold_uncertainty_score":0},"labels":[],"label_agreement":null},{"id":"W4252029050","doi":"10.1002/sim.3909","title":"Construction of confidence limits about effect measures: A general approach—Reply","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Biostatistics; Epidemiology; Library science; Citation; Confidence interval; Medicine; History; Demography; Family medicine; Sociology; Computer science; Pathology","score_opus":0.29723315980429194,"score_gpt":0.5271650463082589,"score_spread":0.22993188650396695,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4252029050","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0002487505,0.002568236,0.007312143,0.9785421,0.010774144,0.000015578908,0.00012465662,0.000054629792,0.00035980198],"genre_scores_gemma":[0.010233162,0.002847008,0.009747967,0.9125966,0.062319495,0.00018557513,0.000087726716,0.0001302232,0.0018521568],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9810739,0.008074465,0.003909579,0.0032906325,0.0029701395,0.0006812034],"domain_scores_gemma":[0.8231303,0.14850573,0.004396639,0.006332507,0.015585157,0.0020496405],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.037997458,0.0021925098,0.0042204247,0.002387993,0.002552472,0.0050318674,0.0074259154,0.04495062,0.005447882],"category_scores_gemma":[0.22233956,0.0020328958,0.0040858085,0.0026748283,0.01871223,0.013153202,0.0049357223,0.076362774,0.0031385496],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002660488,0.00008925434,0.0014202354,0.0009596377,0.0005328377,0.0010305179,0.0015725221,0.0015418377,0.00063558907,0.10344043,0.85272753,0.03578365],"study_design_scores_gemma":[0.000814581,0.00018076188,0.004874833,0.0016850701,0.00061896216,0.003876981,0.0014695714,0.006083331,0.0011568894,0.5942347,0.38433954,0.00066474144],"about_ca_topic_score_codex":0.0073492,"about_ca_topic_score_gemma":0.003529994,"teacher_disagreement_score":0.9620025,"about_ca_system_score_codex":0.00611783,"about_ca_system_score_gemma":0.0047341622,"threshold_uncertainty_score":0.20095217},"labels":[],"label_agreement":null},{"id":"W4252233634","doi":"10.1002/sim.8229","title":"Issue Information","year":2020,"lang":"en","type":"paratext","venue":"Statistics in Medicine","topic":"Human auditory perception and evaluation","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Citation; Information retrieval; World Wide Web","score_opus":0.026466283137171996,"score_gpt":0.32556839600590054,"score_spread":0.29910211286872856,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4252233634","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00023056837,0.0008246917,0.002584563,0.004398188,0.014554502,0.0005144441,0.030854395,0.0045710616,0.9414676],"genre_scores_gemma":[0.00051349937,0.00047552335,0.0009439818,0.0014811117,0.0016859153,0.00014168433,0.011311654,0.00092556555,0.9825211],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99824274,0.00023613541,0.0001278102,0.00021511903,0.0010209094,0.0001572599],"domain_scores_gemma":[0.9893885,0.0016811463,0.00042114034,0.0007590908,0.0054471837,0.0023029484],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0023441457,0.0014101088,0.0014623948,0.006028374,0.0012730561,0.008461499,0.0016874053,0.0029332065,0.8969989],"category_scores_gemma":[0.016923161,0.0006249296,0.0010215153,0.0041817417,0.0006304504,0.0030057887,0.0023121862,0.0019864985,0.8633027],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000011573213,0.000015959036,0.00003907733,0.00011789477,0.0000013070835,0.000007143471,0.0000050266926,0.000040134368,0.000061479084,0.0016063204,0.9674511,0.030642886],"study_design_scores_gemma":[0.000016382757,0.00001345445,0.00019173003,0.00011692389,0.000001767309,0.000015508293,0.000012067418,0.00010773989,0.000068495334,0.0023038485,0.9971462,0.0000058368355],"about_ca_topic_score_codex":0.0027643365,"about_ca_topic_score_gemma":0.005063108,"teacher_disagreement_score":0.10300112,"about_ca_system_score_codex":0.0015374128,"about_ca_system_score_gemma":0.0040387623,"threshold_uncertainty_score":0.14691854},"labels":[],"label_agreement":null},{"id":"W4252992953","doi":"10.1002/(sici)1097-0258(20000415)19:7<975::aid-sim381>3.3.co;2-0","title":"A robust mixed linear model analysis for longitudinal data","year":2000,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Okanagan University College; University of British Columbia","funders":"","keywords":"Mixed model; Random effects model; Generalized linear mixed model; Bayesian probability; Autocorrelation; Computer science; Longitudinal data; Linear model; Statistics; Econometrics; Linear regression; Robust regression; Mathematics; Data mining; Medicine","score_opus":0.3914669943620445,"score_gpt":0.473223682916968,"score_spread":0.08175668855492346,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4252992953","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00014912053,0.0004098068,0.9982364,0.000103692495,0.000059877526,0.000089028305,0.00021907907,0.0005131477,0.00021982624],"genre_scores_gemma":[0.005495176,0.0007246196,0.99039245,0.00013069437,0.00016666116,0.001351935,0.0006897857,0.0002877593,0.0007610076],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97134215,0.022310546,0.00096933154,0.002062896,0.003045767,0.00026928584],"domain_scores_gemma":[0.97450507,0.019289292,0.001783654,0.0026107957,0.0014998402,0.0003113233],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.023640957,0.002431696,0.003198434,0.0033583366,0.0008613384,0.002604504,0.0039173868,0.0021505859,0.008513041],"category_scores_gemma":[0.06604922,0.0012761451,0.004522503,0.0037722364,0.0011250762,0.0020946467,0.003688394,0.004253989,0.0049421857],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004120371,0.00020456224,0.0017978308,0.0020593358,0.0030336168,0.0006815653,0.00046516984,0.08603367,0.004835864,0.3169284,0.026513726,0.55703425],"study_design_scores_gemma":[0.00029914276,0.0009083408,0.0019213727,0.00074010907,0.0010111086,0.0010814387,0.000115739414,0.48905194,0.0037680804,0.37328842,0.12747785,0.0003365332],"about_ca_topic_score_codex":0.001617292,"about_ca_topic_score_gemma":0.0018874239,"teacher_disagreement_score":0.023640957,"about_ca_system_score_codex":0.0010115183,"about_ca_system_score_gemma":0.0033161985,"threshold_uncertainty_score":0.12502682},"labels":[],"label_agreement":null},{"id":"W4254487627","doi":"10.1002/sim.8864","title":"Editorial","year":2020,"lang":"en","type":"editorial","venue":"Statistics in Medicine","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science","score_opus":0.012351039779580427,"score_gpt":0.34109523348290754,"score_spread":0.32874419370332714,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4254487627","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000014735852,0.0013505695,0.00007980033,0.026882289,0.96921283,0.000029138975,0.00009178789,0.000094930045,0.0022439866],"genre_scores_gemma":[0.00030469897,0.0018977233,0.00012589767,0.029927948,0.9531012,0.000049117425,0.00011092877,0.00008013189,0.014402367],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98542285,0.0028170974,0.0016052976,0.0014379835,0.0078545045,0.0008623125],"domain_scores_gemma":[0.9246649,0.022383614,0.0044129067,0.0027513886,0.037111226,0.008675994],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011855578,0.0035885284,0.0044917115,0.005187719,0.0041480083,0.012874439,0.0050583496,0.01566228,0.059014235],"category_scores_gemma":[0.08768968,0.0014935575,0.0032402433,0.0019580594,0.0029743183,0.005974519,0.0022535801,0.020121077,0.062505],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00002332534,0.0000053622243,0.000008953655,0.00012432502,0.000009322029,0.00004272173,0.0000038736857,0.000008710683,0.000011458585,0.00019583036,0.9971361,0.0024299955],"study_design_scores_gemma":[0.00008278388,0.00002023271,0.00010520569,0.00067475374,0.000036133188,0.00016943921,0.00002275135,0.000102564176,0.00007062685,0.0013474807,0.99735093,0.000017127773],"about_ca_topic_score_codex":0.0019645407,"about_ca_topic_score_gemma":0.004056148,"teacher_disagreement_score":0.059014235,"about_ca_system_score_codex":0.0037358177,"about_ca_system_score_gemma":0.0063816747,"threshold_uncertainty_score":0},"labels":[],"label_agreement":null},{"id":"W4254597931","doi":"10.1002/sim.8247","title":"Issue Information","year":2020,"lang":"en","type":"paratext","venue":"Statistics in Medicine","topic":"Human auditory perception and evaluation","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"National Institutes of Health","keywords":"Computer science; Citation; Information retrieval; Library science","score_opus":0.026466283137171996,"score_gpt":0.32556839600590054,"score_spread":0.29910211286872856,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4254597931","genre_codex":"other","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"editorial","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00023056837,0.0008246917,0.002584563,0.004398188,0.014554502,0.0005144441,0.030854395,0.0045710616,0.9414676],"genre_scores_gemma":[0.00051349937,0.00047552335,0.0009439818,0.0014811117,0.0016859153,0.00014168433,0.011311654,0.00092556555,0.9825211],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99824274,0.00023613541,0.0001278102,0.00021511903,0.0010209094,0.0001572599],"domain_scores_gemma":[0.9893885,0.0016811463,0.00042114034,0.0007590908,0.0054471837,0.0023029484],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0023441457,0.0014101088,0.0014623948,0.006028374,0.0012730561,0.008461499,0.0016874053,0.0029332065,0.8969989],"category_scores_gemma":[0.016923161,0.0006249296,0.0010215153,0.0041817417,0.0006304504,0.0030057887,0.0023121862,0.0019864985,0.8633027],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000011573213,0.000015959036,0.00003907733,0.00011789477,0.0000013070835,0.000007143471,0.0000050266926,0.000040134368,0.000061479084,0.0016063204,0.9674511,0.030642886],"study_design_scores_gemma":[0.000016382757,0.00001345445,0.00019173003,0.00011692389,0.000001767309,0.000015508293,0.000012067418,0.00010773989,0.000068495334,0.0023038485,0.9971462,0.0000058368355],"about_ca_topic_score_codex":0.0027643365,"about_ca_topic_score_gemma":0.005063108,"teacher_disagreement_score":0.10300112,"about_ca_system_score_codex":0.0015374128,"about_ca_system_score_gemma":0.0040387623,"threshold_uncertainty_score":0.14691854},"labels":[],"label_agreement":null},{"id":"W4280491254","doi":"10.1002/sim.9431","title":"The impact of methodological choices when developing predictive models using urinary metabolite data","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Metabolomics and Mass Spectrometry Studies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta; University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Canada Foundation for Innovation; National Institute of Diabetes and Digestive and Kidney Diseases; Kidney Foundation of Canada","keywords":"Metabolomics; Computer science; Generalizability theory; Database normalization; Normalization (sociology); Linear discriminant analysis; Metabolite; Urinary system; Data mining; Computational biology; Bioinformatics; Medicine; Artificial intelligence; Biology; Internal medicine; Pattern recognition (psychology); Statistics; Mathematics","score_opus":0.18211004271361309,"score_gpt":0.43133334131057005,"score_spread":0.24922329859695697,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4280491254","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.062348343,0.0035249698,0.9223659,0.00597685,0.00052469806,0.0011227176,0.0016319881,0.0006880251,0.0018165689],"genre_scores_gemma":[0.53174704,0.0020613358,0.45732173,0.0022528782,0.0004238096,0.0023278429,0.0028342777,0.00043191304,0.00059924106],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.85863584,0.1221062,0.007985769,0.0052842586,0.0053356835,0.00065220374],"domain_scores_gemma":[0.6046245,0.358834,0.007889806,0.017815923,0.01018127,0.00065441383],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.24758536,0.0016926355,0.0016630177,0.0021790618,0.0013174872,0.00613237,0.004000113,0.0020372595,0.0011609995],"category_scores_gemma":[0.41487953,0.0006658991,0.0033763556,0.0030179066,0.0020236955,0.0035288988,0.0027472733,0.0042086677,0.0005322772],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0037317888,0.0007570719,0.13181628,0.0025584497,0.010159636,0.0010797959,0.00089727,0.4334115,0.0062894863,0.043439537,0.0068199206,0.35903928],"study_design_scores_gemma":[0.00085436914,0.0010036896,0.026946627,0.0011856025,0.0013934033,0.0005939154,0.0006945092,0.84919286,0.010170371,0.09820911,0.00950698,0.00024850262],"about_ca_topic_score_codex":0.004669586,"about_ca_topic_score_gemma":0.0034685174,"teacher_disagreement_score":0.75241464,"about_ca_system_score_codex":0.0013507516,"about_ca_system_score_gemma":0.0037606661,"threshold_uncertainty_score":0.9278612},"labels":[],"label_agreement":null},{"id":"W4280543579","doi":"10.1002/sim.9433","title":"Dynamic prediction with time‐dependent marker in survival analysis using supervised functional principal component analysis","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"National Cancer Institute; Natural Sciences and Engineering Research Council of Canada; Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada; Division of Cancer Prevention, National Cancer Institute","keywords":"Principal component analysis; Computer science; Functional principal component analysis; Event (particle physics); Artificial intelligence; Independent component analysis; Data mining; Machine learning; Pattern recognition (psychology); Feature (linguistics)","score_opus":0.05897852154404278,"score_gpt":0.35533086880594217,"score_spread":0.2963523472618994,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4280543579","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.032918748,0.00036396238,0.9654665,0.00023252155,0.000035771653,0.000057815363,0.00019724703,0.0004802927,0.00024709926],"genre_scores_gemma":[0.69917995,0.00058224716,0.29669562,0.00013354412,0.0001809057,0.0004459914,0.0012940029,0.00016820652,0.0013195915],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9984237,0.00080701767,0.000085662665,0.0004021439,0.00018783874,0.00009367896],"domain_scores_gemma":[0.99625313,0.0022282142,0.00038920974,0.00048080154,0.0005203746,0.00012823654],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0043332935,0.0010005268,0.0011866473,0.0015949013,0.00049001054,0.00093886134,0.0010904346,0.0008280188,0.0007813925],"category_scores_gemma":[0.010563569,0.00042112567,0.0012652604,0.0015004328,0.0007267922,0.0011411335,0.0010397413,0.0018324326,0.00037173627],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004445009,0.00027443154,0.022197314,0.00016589198,0.00034724185,0.00023333657,0.00025407868,0.59022874,0.0046454286,0.010663957,0.0057831323,0.36476192],"study_design_scores_gemma":[0.000011878234,0.00003709275,0.001795468,0.000008674883,0.000021129692,0.00004804541,0.0000115744715,0.9881633,0.00061503786,0.008675633,0.00059442426,0.000017770397],"about_ca_topic_score_codex":0.0042807786,"about_ca_topic_score_gemma":0.005069374,"teacher_disagreement_score":0.0043332935,"about_ca_system_score_codex":0.00052381336,"about_ca_system_score_gemma":0.0017469948,"threshold_uncertainty_score":0.022916913},"labels":[],"label_agreement":null},{"id":"W4281251350","doi":"10.1002/sim.9441","title":"The polytomous discrimination index for prediction involving multistate processes under intermittent observation","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Waterloo","funders":"National Cancer Institute; Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Polytomous Rasch model; Multinomial logistic regression; Context (archaeology); Econometrics; Computer science; Logistic regression; Index (typography); Time horizon; Statistics; Machine learning; Mathematics; Item response theory; Psychometrics; Mathematical optimization","score_opus":0.1312956431616157,"score_gpt":0.40548558923975936,"score_spread":0.2741899460781436,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4281251350","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06890711,0.00087955856,0.9252093,0.0015612339,0.0001957775,0.00019290113,0.00048050238,0.00027150495,0.0023020066],"genre_scores_gemma":[0.85586,0.00060094404,0.13962401,0.0005787225,0.0004549503,0.00053401483,0.0008584941,0.00008973578,0.001399205],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9858716,0.007938171,0.0007817203,0.0026347067,0.0022176427,0.00055607606],"domain_scores_gemma":[0.8187799,0.15384413,0.010559251,0.011835452,0.0031302513,0.0018510715],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.036290716,0.0011149915,0.0019162082,0.0038848897,0.0010200795,0.002424761,0.003379653,0.0024423073,0.0026011546],"category_scores_gemma":[0.13334,0.00043924438,0.0019181279,0.0035721178,0.0046521104,0.004822512,0.0036167107,0.0052463044,0.00039597417],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005442718,0.00039894486,0.12618022,0.0004334972,0.0010286093,0.0007589612,0.0006143665,0.43401167,0.0012310488,0.3249493,0.003740217,0.10610885],"study_design_scores_gemma":[0.000029197026,0.00017261184,0.007212774,0.00006253286,0.000049306767,0.00015450457,0.00006532534,0.8286583,0.0002729607,0.16236587,0.0009163998,0.00004021623],"about_ca_topic_score_codex":0.0030869583,"about_ca_topic_score_gemma":0.001726487,"teacher_disagreement_score":0.036290716,"about_ca_system_score_codex":0.0016358622,"about_ca_system_score_gemma":0.0021334107,"threshold_uncertainty_score":0.19192594},"labels":[],"label_agreement":null},{"id":"W4281383153","doi":"10.1002/sim.9442","title":"A classification for complex imbalanced data in disease screening and early diagnosis","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Imbalanced Data Classification Techniques","field":"Computer Science","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Institute on Aging; Canadian Institutes of Health Research","keywords":"Computer science; Artificial intelligence; Machine learning","score_opus":0.1130163235397242,"score_gpt":0.3732825950494658,"score_spread":0.26026627150974163,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4281383153","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029039958,0.000713818,0.9672573,0.0012692409,0.00017755781,0.00006841724,0.00035463896,0.0003363978,0.0007825855],"genre_scores_gemma":[0.58062893,0.0012189774,0.41173786,0.0005848628,0.0007406054,0.00039365623,0.0016150303,0.00012437244,0.0029557094],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9975957,0.0009999267,0.00017776297,0.0005378913,0.00052988844,0.00015869692],"domain_scores_gemma":[0.99349505,0.0036084095,0.00080670137,0.00092199206,0.00095329806,0.00021467207],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005415646,0.0007350893,0.0011037282,0.0024473208,0.0010020097,0.0017651764,0.0011967397,0.0014404003,0.0013654337],"category_scores_gemma":[0.017268823,0.00026976474,0.0008965365,0.0022746972,0.0012699915,0.0018076428,0.0017218131,0.0018730441,0.00071716605],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005858224,0.0003198982,0.026824685,0.00040616692,0.00019578569,0.0005724767,0.0005287989,0.16788113,0.0076994067,0.082144864,0.02039997,0.69244105],"study_design_scores_gemma":[0.000016719836,0.000091865404,0.0045824703,0.00004908917,0.000024149023,0.0002700443,0.000119647455,0.9240052,0.0016648325,0.06493822,0.0042068297,0.000030983072],"about_ca_topic_score_codex":0.0009825587,"about_ca_topic_score_gemma":0.00087463023,"teacher_disagreement_score":0.005415646,"about_ca_system_score_codex":0.0006871055,"about_ca_system_score_gemma":0.0011354857,"threshold_uncertainty_score":0.028641045},"labels":[],"label_agreement":null},{"id":"W4283706657","doi":"10.1002/sim.9509","title":"Augmented weighting estimators for the additive rates model under multivariate recurrent event data with missing event type","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto","funders":"Fundamental Research Funds for the Central Universities; Natural Sciences and Engineering Research Council of Canada; National Natural Science Foundation of China","keywords":"Inverse probability weighting; Missing data; Multivariate statistics; Estimator; Event (particle physics); Nonparametric statistics; Computer science; Inverse probability; Statistics; Weighting; Econometrics; Data mining; Mathematics; Medicine; Bayesian probability","score_opus":0.19762742846424525,"score_gpt":0.47804157139208986,"score_spread":0.2804141429278446,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4283706657","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0037639795,0.00010301253,0.99567527,0.00005961077,0.000017313514,0.000040515166,0.000054924818,0.000071858325,0.00021351491],"genre_scores_gemma":[0.24582183,0.0011919019,0.74501485,0.00023005258,0.0002556697,0.0009854625,0.000964621,0.00017861683,0.005356964],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9887861,0.007353333,0.00056738843,0.001474343,0.001405125,0.0004136591],"domain_scores_gemma":[0.95720273,0.029116506,0.004408886,0.0058820504,0.0029450345,0.0004447879],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02733592,0.001551485,0.0024630409,0.0020077482,0.00048181275,0.0019680855,0.0053324304,0.0020345952,0.004069656],"category_scores_gemma":[0.07496189,0.00096177665,0.002332567,0.0023768356,0.0017762267,0.0055123423,0.0030042373,0.0038598655,0.0012147594],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00025559007,0.00022964431,0.0062280367,0.00051578815,0.00050255004,0.00032822875,0.00052913104,0.24612717,0.0022969984,0.51498294,0.0024776033,0.22552624],"study_design_scores_gemma":[0.000059733557,0.00015541994,0.0013488041,0.00007964402,0.00013875864,0.0002277616,0.00007393167,0.80290127,0.00089428195,0.19098413,0.0030622703,0.00007394506],"about_ca_topic_score_codex":0.0015048004,"about_ca_topic_score_gemma":0.0014152102,"teacher_disagreement_score":0.02733592,"about_ca_system_score_codex":0.0008852699,"about_ca_system_score_gemma":0.0018276453,"threshold_uncertainty_score":0.1445679},"labels":[],"label_agreement":null},{"id":"W4283804698","doi":"10.1002/sim.9512","title":"Unified estimation for Cox regression model with nonmonotone missing at random covariates","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University; University of Regina","funders":"","keywords":"Covariate; Estimator; Missing data; Computer science; Proportional hazards model; Imputation (statistics); Statistics; Mathematics; Econometrics","score_opus":0.06925157385471965,"score_gpt":0.40727885511625317,"score_spread":0.33802728126153353,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4283804698","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0025698452,0.00037364292,0.9963856,0.00012340274,0.000028470085,0.000052118037,0.00010464199,0.00009217924,0.0002701401],"genre_scores_gemma":[0.23014486,0.001793195,0.76186484,0.00045436458,0.00034078854,0.0011718032,0.0016270225,0.00012659022,0.0024764999],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98000526,0.01461467,0.00068887195,0.0020236822,0.0021308826,0.00053654826],"domain_scores_gemma":[0.9724204,0.018522652,0.0028315561,0.0038952786,0.0019803771,0.00034974408],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.025119184,0.0010414892,0.0025604018,0.0022825315,0.00068239454,0.00196885,0.003919985,0.0014053796,0.0033047367],"category_scores_gemma":[0.059831966,0.0010393942,0.0027809183,0.0037448248,0.0010596855,0.0030733559,0.0029178206,0.0026856968,0.0006227695],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00020999463,0.00015365497,0.018211264,0.00070195145,0.0012992258,0.00036712288,0.0005653497,0.11709067,0.0012694919,0.59103173,0.008299593,0.2607999],"study_design_scores_gemma":[0.000119629854,0.00030448558,0.0055781277,0.00016848213,0.00049892766,0.00033978253,0.00014577188,0.74346006,0.001142462,0.23742333,0.010717652,0.00010126079],"about_ca_topic_score_codex":0.004428439,"about_ca_topic_score_gemma":0.0046846163,"teacher_disagreement_score":0.025119184,"about_ca_system_score_codex":0.0013123439,"about_ca_system_score_gemma":0.0038371196,"threshold_uncertainty_score":0.13284457},"labels":[],"label_agreement":null},{"id":"W4284892632","doi":"10.1002/sim.9516","title":"Adaptive response‐dependent two‐phase designs: Some results on robustness and efficiency","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Robustness (evolution); Computer science; Covariate; Biomarker; Statistics; Quantile; Data mining; Econometrics; Machine learning; Mathematics","score_opus":0.47951310210464493,"score_gpt":0.5641417443802038,"score_spread":0.08462864227555889,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284892632","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005287645,0.00079259736,0.9916379,0.0003472486,0.000055788503,0.00046627066,0.00008801105,0.000106946885,0.0012175415],"genre_scores_gemma":[0.13786055,0.0018644565,0.8526862,0.00055510877,0.00033711395,0.0038483692,0.00037820236,0.00023165172,0.0022382482],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9017719,0.086626746,0.0023397775,0.0045013786,0.003934878,0.00082530396],"domain_scores_gemma":[0.43405443,0.51873314,0.0134429,0.02485097,0.0076316264,0.001286927],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.17996301,0.0032480205,0.0040472336,0.0028807223,0.0012620084,0.0020307319,0.0039823493,0.0035225162,0.006713701],"category_scores_gemma":[0.34459692,0.0017891412,0.005194171,0.0027424125,0.0064318315,0.0043863454,0.00447227,0.00601311,0.0009952646],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0025311816,0.00057003775,0.005699174,0.00111368,0.0015093366,0.00030412074,0.0012401312,0.16529949,0.0035415739,0.66645736,0.002165346,0.14956854],"study_design_scores_gemma":[0.0013733942,0.0022963432,0.002648746,0.00044096095,0.0006339745,0.0002920928,0.0001844072,0.5002386,0.003737946,0.48019627,0.0077812304,0.00017597534],"about_ca_topic_score_codex":0.0013221715,"about_ca_topic_score_gemma":0.00072256604,"teacher_disagreement_score":0.17996301,"about_ca_system_score_codex":0.001963608,"about_ca_system_score_gemma":0.0031783355,"threshold_uncertainty_score":0.9517467},"labels":[],"label_agreement":null},{"id":"W4285592479","doi":"10.1002/sim.9519","title":"Bootstrap vs asymptotic variance estimation when using propensity score weighting with continuous and binary outcomes","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":43,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Sunnybrook Hospital; University of Toronto","funders":"Canadian Institutes of Health Research; Ministerio de Sanidad, Consumo y Bienestar Social; Heart and Stroke Foundation of Canada","keywords":"Statistics; Mathematics; Propensity score matching; Estimator; Confidence interval; Weighting; Average treatment effect; Sample size determination; Delta method; Matching (statistics); Inverse probability weighting; Medicine","score_opus":0.14787825533505633,"score_gpt":0.3957896493172065,"score_spread":0.24791139398215017,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4285592479","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.034661956,0.0007184908,0.9621612,0.00026117408,0.00006789542,0.0002575507,0.00009234129,0.0002878013,0.0014916097],"genre_scores_gemma":[0.5012664,0.00058091624,0.49533364,0.0002825253,0.00009254265,0.0013716981,0.00032201275,0.00024256363,0.00050774304],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9187398,0.072013006,0.0018738831,0.0022011932,0.004449434,0.00072260055],"domain_scores_gemma":[0.69021803,0.281695,0.008022352,0.014831424,0.004794616,0.00043857377],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10069794,0.0008880088,0.0017914183,0.0026829143,0.0005499455,0.0018325923,0.0018646066,0.0020890445,0.0024782196],"category_scores_gemma":[0.39423493,0.00076731027,0.0018094073,0.002517194,0.0020922658,0.003921348,0.0025301205,0.0024039608,0.00035725284],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018348887,0.00047459383,0.036521792,0.0013170798,0.0018071609,0.0006261098,0.0014166728,0.45148188,0.0021615303,0.26753426,0.003771671,0.23105246],"study_design_scores_gemma":[0.00043307358,0.00056375837,0.0066005085,0.0005050956,0.0002521352,0.00036048633,0.00023620244,0.7745339,0.0023292857,0.2100247,0.004071845,0.00008907843],"about_ca_topic_score_codex":0.0027557139,"about_ca_topic_score_gemma":0.0020874022,"teacher_disagreement_score":0.89930207,"about_ca_system_score_codex":0.0009282522,"about_ca_system_score_gemma":0.0018560913,"threshold_uncertainty_score":0.53254795},"labels":[],"label_agreement":null},{"id":"W4285732379","doi":"10.1002/sim.9520","title":"Bayesian unanchored additive models for component network meta‐analysis","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Notation; Component (thermodynamics); Bayesian probability; Bayesian network; Statistical model; Additive model; Contrast (vision); Econometrics; Machine learning; Data mining; Artificial intelligence; Mathematics","score_opus":0.6624841692767993,"score_gpt":0.5228154949789899,"score_spread":0.13966867429780938,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4285732379","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0022084892,0.0022879147,0.9934105,0.0004854031,0.00013755029,0.00022828257,0.00034114323,0.00017195458,0.0007287236],"genre_scores_gemma":[0.2170327,0.0053093457,0.7672122,0.00094153266,0.00052252325,0.00442822,0.0010623052,0.00018954341,0.0033017052],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9479439,0.045154,0.0012694759,0.0029646559,0.0023150234,0.0003529779],"domain_scores_gemma":[0.9006473,0.08834883,0.0034823092,0.0053663333,0.0018816219,0.00027372097],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.052761067,0.002547652,0.0038918806,0.0044051376,0.0010511652,0.0035551786,0.0050858026,0.0025857044,0.0050631706],"category_scores_gemma":[0.12344465,0.0013907214,0.0057645305,0.004820101,0.0019509902,0.003652584,0.00313902,0.0045139734,0.00086029427],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005636803,0.00010649797,0.003717586,0.0022586586,0.006300485,0.00025988262,0.00047557038,0.3750198,0.0005441,0.48995033,0.004993745,0.11580975],"study_design_scores_gemma":[0.0001807677,0.00017844203,0.00076101796,0.00036267066,0.002318211,0.00013185805,0.000036296282,0.55970615,0.00036617173,0.42784372,0.008038895,0.00007579665],"about_ca_topic_score_codex":0.005944165,"about_ca_topic_score_gemma":0.007445839,"teacher_disagreement_score":0.052761067,"about_ca_system_score_codex":0.002264279,"about_ca_system_score_gemma":0.0023302266,"threshold_uncertainty_score":0.2790305},"labels":[],"label_agreement":null},{"id":"W4286209045","doi":"10.1002/sim.9529","title":"Modeling and forecasting of at home activity in older adults using passive sensor technology","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Context-Aware Activity Recognition Systems","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"Engineering and Physical Sciences Research Council","keywords":"Categorical variable; Quarter (Canadian coin); Life expectancy; Computer science; Gerontology; Medicine; Machine learning; Geography; Environmental health","score_opus":0.04229577504528906,"score_gpt":0.2954046475400237,"score_spread":0.2531088724947347,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4286209045","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9542433,0.0010012133,0.03645468,0.0006978626,0.0001048129,0.00007209307,0.005865207,0.0004008627,0.0011599502],"genre_scores_gemma":[0.9797493,0.00053224846,0.011535566,0.00006536801,0.00006744803,0.000063884836,0.0067954003,0.0000138721225,0.0011767875],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99978775,0.000050458646,0.000023698689,0.00007260725,0.000039331317,0.00002610092],"domain_scores_gemma":[0.99943143,0.00030873084,0.00006500527,0.00004869761,0.00010309953,0.000043038406],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0006103538,0.00075693976,0.00058722193,0.000746947,0.00019772234,0.0006485642,0.00078429666,0.00074841623,0.0006756967],"category_scores_gemma":[0.0018152047,0.00023023436,0.0006899577,0.00077170786,0.00019813546,0.0005845351,0.00038306369,0.0006965986,0.00043884385],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000615163,0.0009893421,0.18998608,0.0002458033,0.00020701833,0.0003330346,0.00026283096,0.6843318,0.0044247974,0.0010158464,0.006085431,0.11150291],"study_design_scores_gemma":[0.000013886292,0.00013267697,0.0278303,0.000015604352,0.000017379032,0.00004605282,0.00007174546,0.9699084,0.000489312,0.00077338226,0.00069065276,0.00001050482],"about_ca_topic_score_codex":0.025861306,"about_ca_topic_score_gemma":0.028995022,"teacher_disagreement_score":0.025861306,"about_ca_system_score_codex":0.00048507605,"about_ca_system_score_gemma":0.00032400156,"threshold_uncertainty_score":0.051421523},"labels":[],"label_agreement":null},{"id":"W4290033699","doi":"10.1002/sim.9538","title":"Calculating the power to examine treatment‐covariate interactions when planning an individual participant data meta‐analysis of randomized trials with a binary outcome","year":2022,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"NIHR Oxford Biomedical Research Centre; Medical Research Council; National Institute for Health and Care Research; Medical Research Council Canada; Cancer Research UK","keywords":"Covariate; Statistics; Outcome (game theory); Variance (accounting); Econometrics; Logistic regression; Computer science; Statistical power; Wald test; Binary data; Type I and type II errors; Binary number; Mathematics; Statistical hypothesis testing","score_opus":0.9718818217885831,"score_gpt":0.689846894753815,"score_spread":0.28203492703476807,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4290033699","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008091311,0.019430256,0.9363319,0.0057030963,0.0020194238,0.016820822,0.004641254,0.0018210623,0.0051408503],"genre_scores_gemma":[0.14539573,0.004444742,0.7663972,0.0041012657,0.0005855913,0.07483794,0.0017880558,0.0010823387,0.0013671634],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.56536496,0.38702652,0.022030411,0.011787385,0.012605068,0.0011856836],"domain_scores_gemma":[0.33154175,0.6298409,0.0093151,0.023681508,0.0048869415,0.0007337564],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4285697,0.0038191227,0.013009856,0.007130497,0.0009966115,0.0072167884,0.005280107,0.0071254116,0.011959633],"category_scores_gemma":[0.681321,0.0029821596,0.03882508,0.0061518317,0.0033000722,0.00835879,0.006680632,0.009692952,0.0020430137],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.021728724,0.00045027107,0.025834536,0.07232811,0.24139817,0.0017403556,0.0025882833,0.049500316,0.0043390403,0.0758115,0.026366195,0.47791442],"study_design_scores_gemma":[0.024572346,0.0057842024,0.016000964,0.023169814,0.19599953,0.0019911544,0.00077156536,0.11969366,0.019423336,0.48242348,0.10908245,0.001087536],"about_ca_topic_score_codex":0.0014632376,"about_ca_topic_score_gemma":0.0013067257,"teacher_disagreement_score":0.5714303,"about_ca_system_score_codex":0.0017784169,"about_ca_system_score_gemma":0.0035278746,"threshold_uncertainty_score":0.7046753},"labels":[],"label_agreement":null},{"id":"W4290034027","doi":"10.1002/sim.9547","title":"Use of copula to model within‐study association in bivariate meta‐analysis of binomial data at the aggregate level: A Bayesian approach and application to surrogate endpoint evaluation","year":2022,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Chronic Myeloid Leukemia Treatments","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada; University of Leicester","keywords":"Bivariate analysis; Copula (linguistics); Statistics; Beta-binomial distribution; Econometrics; Random effects model; Continuity correction; Surrogate endpoint; Binomial distribution; Bayesian probability; Computer science; Meta-analysis; Mathematics; Negative binomial distribution; Medicine; Internal medicine; Poisson distribution","score_opus":0.3400497914709901,"score_gpt":0.44790679767266445,"score_spread":0.10785700620167432,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4290034027","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0018105336,0.0039116032,0.993178,0.00032008937,0.000063695115,0.00016612449,0.00018834212,0.00016848168,0.0001930582],"genre_scores_gemma":[0.12637518,0.009403955,0.8582372,0.0007580069,0.00029662708,0.0027281833,0.000848857,0.00036837463,0.0009836043],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.936529,0.055125605,0.0021950984,0.002893243,0.0027494985,0.000507543],"domain_scores_gemma":[0.88587224,0.10168618,0.0046578054,0.004700364,0.002671459,0.0004119662],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.103614025,0.0029817058,0.0073841275,0.0077523305,0.0009438818,0.004195481,0.004078829,0.0032236257,0.0026411577],"category_scores_gemma":[0.15351586,0.0022013162,0.012808709,0.007561502,0.0019976527,0.003350702,0.0034434963,0.005382875,0.000591077],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008329855,0.00017602686,0.013833181,0.0054564555,0.028506542,0.001080351,0.0007099014,0.5948707,0.0016660623,0.14737761,0.00521265,0.20027757],"study_design_scores_gemma":[0.00034812695,0.00028172063,0.0026966163,0.0008212459,0.006069676,0.00048048387,0.00007485095,0.8087787,0.00078741694,0.17157853,0.007897457,0.00018515985],"about_ca_topic_score_codex":0.0064660776,"about_ca_topic_score_gemma":0.0059402892,"teacher_disagreement_score":0.89638597,"about_ca_system_score_codex":0.0019456558,"about_ca_system_score_gemma":0.003914751,"threshold_uncertainty_score":0.5479699},"labels":[],"label_agreement":null},{"id":"W4297271369","doi":"10.1002/sim.9549","title":"Substantive model compatible multilevel multiple imputation: A joint modeling approach","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada","keywords":"Imputation (statistics); Covariate; Computer science; Missing data; Statistics; Multilevel model; Data mining; Econometrics; Mathematics; Machine learning","score_opus":0.19222843533564607,"score_gpt":0.40369911782895407,"score_spread":0.211470682493308,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4297271369","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0008944612,0.00008711181,0.9979936,0.00030316782,0.000015471862,0.000046094734,0.000088695335,0.00016627775,0.000404987],"genre_scores_gemma":[0.05945329,0.00019687567,0.93870103,0.00023699344,0.000082606304,0.00033196408,0.00037054758,0.00014751311,0.00047926232],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9730576,0.02228673,0.0007082682,0.0012393426,0.0024031943,0.0003048306],"domain_scores_gemma":[0.95212364,0.03402144,0.0028249968,0.006873973,0.0034794975,0.0006765082],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026746912,0.0006822515,0.0014364758,0.002033247,0.00096595194,0.002606472,0.0045401813,0.0016103219,0.00499276],"category_scores_gemma":[0.073917836,0.0007935,0.0025373418,0.0035889666,0.001143918,0.0017870705,0.0040338645,0.0037934426,0.0012754527],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00021080059,0.00019181106,0.012555591,0.0008246595,0.0013212566,0.0004718941,0.0012866739,0.10728419,0.0013604899,0.5324106,0.016845703,0.32523632],"study_design_scores_gemma":[0.0000732151,0.00009494709,0.0018956569,0.00020606391,0.00018084465,0.00029834133,0.00011679928,0.5823267,0.00078945345,0.40067866,0.013290715,0.000048665694],"about_ca_topic_score_codex":0.0020128072,"about_ca_topic_score_gemma":0.0031886827,"teacher_disagreement_score":0.026746912,"about_ca_system_score_codex":0.0011965054,"about_ca_system_score_gemma":0.0043570506,"threshold_uncertainty_score":0.14145291},"labels":[],"label_agreement":null},{"id":"W4309515696","doi":"10.1002/sim.9608","title":"Double robust estimation of optimal partially adaptive treatment strategies: An application to breast cancer treatment using hormonal therapy","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre hospitalier de l'Université Laval; McGill University; Université Laval","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Estimator; Breast cancer; Covariate; Computer science; Robustness (evolution); Ordinary least squares; Confounding; Estimation; Econometrics; Statistics; Machine learning; Cancer; Medicine; Mathematics; Internal medicine","score_opus":0.49093574846171567,"score_gpt":0.5630733962666775,"score_spread":0.07213764780496185,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4309515696","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.026145054,0.00091275526,0.9708085,0.00081375334,0.00004154771,0.00018245899,0.00014285631,0.00019916066,0.0007539224],"genre_scores_gemma":[0.55236,0.0007682479,0.44396356,0.00052680343,0.00006876813,0.00056839536,0.00032093457,0.000115769275,0.0013076721],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9879408,0.010429534,0.00026854908,0.0006406219,0.0005141879,0.00020632477],"domain_scores_gemma":[0.89866614,0.09382739,0.0030976343,0.0026285497,0.0014013062,0.0003789028],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.019445807,0.0008989589,0.0017233596,0.00078930205,0.00031588922,0.0011651858,0.001277082,0.0014918064,0.0020163273],"category_scores_gemma":[0.11748376,0.0005002922,0.0013845777,0.0010276233,0.0010735993,0.0010745575,0.0018551889,0.0018022826,0.00019238298],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009985247,0.00023154073,0.00874282,0.00042490222,0.0006975018,0.00030888309,0.00020882879,0.7347401,0.0020933498,0.05771266,0.0022189838,0.19162197],"study_design_scores_gemma":[0.00022476539,0.00031354322,0.0018978586,0.000064247426,0.00012255317,0.00008470137,0.00004967389,0.95282745,0.0010845157,0.04154848,0.0017357289,0.000046520156],"about_ca_topic_score_codex":0.0042577274,"about_ca_topic_score_gemma":0.0026227627,"teacher_disagreement_score":0.019445807,"about_ca_system_score_codex":0.0009233975,"about_ca_system_score_gemma":0.0017351675,"threshold_uncertainty_score":0.10284048},"labels":[],"label_agreement":null},{"id":"W4309699570","doi":"10.1002/sim.9611","title":"Measures of explained variation under the mixture cure model for survival data","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"","keywords":"Variation (astronomy); Statistics; Regression; Residual; Regression analysis; Computer science; Econometrics; Mathematics; Algorithm","score_opus":0.32615973814007776,"score_gpt":0.4525609731484041,"score_spread":0.12640123500832634,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4309699570","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.045186847,0.0013046114,0.95078576,0.00046635367,0.00006328398,0.000085430576,0.0005997834,0.00028137176,0.0012265365],"genre_scores_gemma":[0.8019419,0.0014080323,0.18899813,0.00049042684,0.00022257047,0.00081669085,0.002815346,0.0002811733,0.0030257304],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9878079,0.0066701514,0.0006592482,0.0020459527,0.002351325,0.00046541452],"domain_scores_gemma":[0.93780214,0.046496857,0.005197024,0.0071270633,0.0029120792,0.00046499068],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.027612772,0.0014201156,0.0019712097,0.0041460725,0.0007915146,0.0022641206,0.0021550623,0.0021788161,0.0016461038],"category_scores_gemma":[0.07852644,0.00045568973,0.0020258133,0.0030266219,0.0037370885,0.004631375,0.0031380262,0.0024875754,0.0004832952],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00047695875,0.00015147957,0.034473598,0.0005653243,0.0014504041,0.0004435754,0.0013212025,0.33103237,0.0033573287,0.47698548,0.0032712116,0.14647107],"study_design_scores_gemma":[0.000030329085,0.00028540177,0.016431313,0.00016752996,0.00023460171,0.0005191828,0.0002376311,0.47091582,0.0014913209,0.5059592,0.0035449911,0.00018271446],"about_ca_topic_score_codex":0.0012591687,"about_ca_topic_score_gemma":0.0010447785,"teacher_disagreement_score":0.027612772,"about_ca_system_score_codex":0.0012326093,"about_ca_system_score_gemma":0.0010838065,"threshold_uncertainty_score":0.14603204},"labels":[],"label_agreement":null},{"id":"W4309915237","doi":"10.1002/sim.9071","title":"Issue Information","year":2022,"lang":"en","type":"paratext","venue":"Statistics in Medicine","topic":"Transportation Systems and Logistics","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Citation; Information retrieval; Library science","score_opus":0.013242268203910631,"score_gpt":0.2767656958766696,"score_spread":0.263523427672759,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4309915237","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0002429961,0.00086283823,0.0011985447,0.0039553167,0.013341431,0.0004670936,0.05044645,0.0031598553,0.92632544],"genre_scores_gemma":[0.00061487255,0.0005226748,0.0005609739,0.0014376909,0.0017971414,0.00017190202,0.017664308,0.00084237993,0.976388],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99825853,0.00023930559,0.00015570664,0.00024068478,0.0009396986,0.00016607606],"domain_scores_gemma":[0.9925041,0.0014401767,0.0003816407,0.00071971316,0.00343428,0.0015201408],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0016151266,0.0017538838,0.0017263367,0.007481356,0.0014903034,0.008029563,0.0019181746,0.0029180008,0.9218754],"category_scores_gemma":[0.012760762,0.00076597556,0.0009672629,0.0065336977,0.00056054856,0.0035781665,0.0024892797,0.0020574722,0.8958569],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000012785625,0.000014831047,0.000036234625,0.00013028768,0.000001105675,0.000008823397,0.0000060555985,0.000024101513,0.000046598874,0.0008243783,0.9801338,0.018761057],"study_design_scores_gemma":[0.00001908165,0.000012918632,0.00022267783,0.000118246506,0.000001996151,0.000017112889,0.000017800174,0.000064817104,0.000056332065,0.0011047953,0.998359,0.000005240603],"about_ca_topic_score_codex":0.0024429173,"about_ca_topic_score_gemma":0.0044648917,"teacher_disagreement_score":0.07812458,"about_ca_system_score_codex":0.0012849431,"about_ca_system_score_gemma":0.0032192203,"threshold_uncertainty_score":0.111435235},"labels":[],"label_agreement":null},{"id":"W4309915763","doi":"10.1002/sim.9606","title":"Adjusting for treatment selection in phase II/III clinical trials with time to event data","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Novartis Pharma; Medical Research Council Canada","keywords":"Estimator; Statistics; Selection (genetic algorithm); Stage (stratigraphy); U-statistic; Selection bias; Mean squared error; Variance (accounting); Clinical trial; Efficiency; Mathematics; Computer science; Econometrics; Medicine; Internal medicine; Artificial intelligence","score_opus":0.7994143033413676,"score_gpt":0.7010730436421652,"score_spread":0.09834125969920249,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4309915763","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010263113,0.0020236908,0.98376715,0.0008010339,0.00025381014,0.0015414915,0.00018365397,0.0004572793,0.0007087656],"genre_scores_gemma":[0.30638248,0.0012258136,0.68390226,0.0021057755,0.00021030079,0.004691396,0.00049245224,0.00021318009,0.0007762996],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7749473,0.20211956,0.007909524,0.0051813116,0.008671832,0.0011705387],"domain_scores_gemma":[0.6412518,0.30474213,0.022145724,0.021227391,0.009565152,0.0010677534],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.18356226,0.0011678537,0.00216229,0.0018827234,0.0007578102,0.0020578112,0.0021891587,0.0023025658,0.002270646],"category_scores_gemma":[0.47758085,0.0008197512,0.003017829,0.0028271433,0.0015358712,0.002362091,0.0016850453,0.003053832,0.0004069824],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0052302135,0.00052406435,0.060234368,0.0039379946,0.0071002063,0.00035650263,0.0010665123,0.1318882,0.0036323061,0.104597434,0.010360976,0.67107123],"study_design_scores_gemma":[0.0047255037,0.00499663,0.028279128,0.002534615,0.0054646046,0.000626543,0.00023025862,0.6712813,0.0153647065,0.22736931,0.038749058,0.0003784102],"about_ca_topic_score_codex":0.0014424586,"about_ca_topic_score_gemma":0.0018312478,"teacher_disagreement_score":0.18356226,"about_ca_system_score_codex":0.0015267143,"about_ca_system_score_gemma":0.0044425507,"threshold_uncertainty_score":0.9707816},"labels":[],"label_agreement":null},{"id":"W4310022099","doi":"10.1002/sim.9594","title":"Practical implementation of the partial ordering continual reassessment method in a Phase I combination‐schedule dose‐finding trial","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"NIHR Cambridge Biomedical Research Centre; Medical Research Council Canada; Department of Health and Social Care; Merck KGaA; Medical Research Council; National Institute for Health and Care Research","keywords":"Computer science; Dosing; Maximum tolerated dose; Schedule; Scheduling (production processes); A priori and a posteriori; Operations research; Medical physics; Clinical trial; Risk analysis (engineering); Management science; Medicine; Mathematical optimization; Mathematics","score_opus":0.5226493986877256,"score_gpt":0.6801128053508417,"score_spread":0.1574634066631161,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4310022099","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.013151239,0.00016543736,0.9834614,0.00040529782,0.00005228286,0.00067830354,0.000079989906,0.0009077587,0.0010983017],"genre_scores_gemma":[0.1687051,0.000096853786,0.82838005,0.0002199895,0.00002974802,0.0014162721,0.00008872382,0.000119335404,0.00094398117],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.985637,0.011958262,0.0005309919,0.0006392636,0.001034831,0.0001996759],"domain_scores_gemma":[0.96481764,0.028371455,0.0020411466,0.0027807315,0.0013635549,0.00062546565],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02792236,0.0007289616,0.0009714688,0.0007007435,0.0003419342,0.0011121633,0.0014322194,0.001242341,0.004955563],"category_scores_gemma":[0.044037364,0.00071625185,0.0011807437,0.00049779774,0.0009725476,0.00095995626,0.0013263222,0.0016934236,0.00057726033],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.005353953,0.000562154,0.0037924468,0.0008368608,0.0003028196,0.00038040252,0.0008396588,0.4779752,0.009852424,0.061872303,0.0040156334,0.4342161],"study_design_scores_gemma":[0.0011674884,0.0018161403,0.0005481129,0.0001026199,0.000089132125,0.00016658967,0.000045328678,0.9564052,0.004710323,0.027577393,0.0073063006,0.0000653857],"about_ca_topic_score_codex":0.0009034898,"about_ca_topic_score_gemma":0.0010194678,"teacher_disagreement_score":0.02792236,"about_ca_system_score_codex":0.0008560369,"about_ca_system_score_gemma":0.0024570639,"threshold_uncertainty_score":0.14766932},"labels":[],"label_agreement":null},{"id":"W4310462658","doi":"10.1002/sim.9618","title":"A nonparametric simultaneous confidence band for biomarker effect on the restricted mean survival time","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian Cancer Society; Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Biomarker; Censoring (clinical trials); Confidence interval; Statistics; Proportional hazards model; Breast cancer; Nonparametric statistics; Survival analysis; Oncology; Smoothing; Sample size determination; Medicine; Econometrics; Mathematics; Internal medicine; Cancer; Biology","score_opus":0.2991692867616709,"score_gpt":0.5212572817545618,"score_spread":0.22208799499289084,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4310462658","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008222563,0.0004671176,0.98968494,0.00024020173,0.00006181475,0.00012273886,0.00009741181,0.00031940953,0.0007837436],"genre_scores_gemma":[0.4291754,0.001061026,0.56319547,0.0006399273,0.00044227153,0.002188788,0.00077829586,0.0003695153,0.002149292],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.956972,0.026704453,0.0018933837,0.0064600254,0.0064352113,0.0015348655],"domain_scores_gemma":[0.60892224,0.33773264,0.0147025315,0.024067089,0.012629737,0.0019457848],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06896922,0.0023582075,0.004037635,0.0041172686,0.0011426662,0.0040220483,0.004902762,0.0049479697,0.005549291],"category_scores_gemma":[0.2964082,0.0017070344,0.0040264213,0.0029101602,0.006154444,0.0063443785,0.0072073275,0.006992513,0.0010770233],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0030617993,0.0005699599,0.013750609,0.0018218829,0.0016142458,0.0009945051,0.0020411313,0.20466544,0.010411715,0.48336464,0.0041426024,0.27356148],"study_design_scores_gemma":[0.00041215666,0.0012568504,0.00846592,0.00046579345,0.0006228474,0.0006763525,0.0002569274,0.7606239,0.008414637,0.21188511,0.006684402,0.00023524014],"about_ca_topic_score_codex":0.0016359859,"about_ca_topic_score_gemma":0.00065239205,"teacher_disagreement_score":0.06896922,"about_ca_system_score_codex":0.0014961532,"about_ca_system_score_gemma":0.0027453927,"threshold_uncertainty_score":0.36474848},"labels":[],"label_agreement":null},{"id":"W4310566398","doi":"10.1002/sim.9605","title":"Point estimation for adaptive trial designs I: A methodological review","year":2022,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":49,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"NIHR Cambridge Biomedical Research Centre; Medical Research Council; National Institute for Health and Care Research; Medical Research Council Canada; Department of Health and Social Care; Cancer Research UK; Health and Care Research Wales","keywords":"Estimator; Computer science; Estimation; Type I and type II errors; Point estimation; Point (geometry); Contrast (vision); Set (abstract data type); Econometrics; Statistics; Artificial intelligence; Mathematics","score_opus":0.9563494966257471,"score_gpt":0.7322871607745378,"score_spread":0.22406233585120927,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4310566398","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00007331183,0.9742372,0.021662522,0.0018151179,0.000726849,0.00033844713,0.000115310286,0.000043349082,0.0009878963],"genre_scores_gemma":[0.0029752993,0.94516695,0.046305075,0.0019652592,0.0010379982,0.001911658,0.00016107556,0.000070776485,0.00040606537],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.92307454,0.050787088,0.011634984,0.0030394148,0.01103113,0.00043284384],"domain_scores_gemma":[0.7705557,0.20050737,0.010838248,0.005085126,0.012559023,0.00045460372],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07908149,0.0021873,0.0073278407,0.009158058,0.000783755,0.004763234,0.004243191,0.0046538883,0.005398276],"category_scores_gemma":[0.23690993,0.0017924384,0.007332678,0.010575122,0.0033262372,0.005099765,0.0026749836,0.006416859,0.0022741605],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00027394324,0.00005849606,0.00047373073,0.20088077,0.0022351027,0.00008644704,0.00033453738,0.0026157328,0.00027582364,0.057341646,0.027471336,0.70795244],"study_design_scores_gemma":[0.00048540535,0.00058681116,0.0020156605,0.35284936,0.006154067,0.0011170686,0.00021692838,0.004306053,0.0009683248,0.11222052,0.51880425,0.00027554203],"about_ca_topic_score_codex":0.0028833996,"about_ca_topic_score_gemma":0.0025209072,"teacher_disagreement_score":0.9209185,"about_ca_system_score_codex":0.0057830177,"about_ca_system_score_gemma":0.0112802535,"threshold_uncertainty_score":0.4182279},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4311277361","doi":"10.1002/sim.9629","title":"Consistent inverse probability of treatment weighted estimation of the average treatment effect with mismeasured time‐dependent confounders","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"National Natural Science Foundation of China","keywords":"Observational error; Computer science; Confounding; Inverse probability; Inverse; Statistics; Function (biology); Estimation; Econometrics; Algorithm; Mathematics; Bayesian probability","score_opus":0.0685040974191771,"score_gpt":0.3618328318137441,"score_spread":0.29332873439456697,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4311277361","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0030602517,0.00053924695,0.9952852,0.0003623946,0.000074077805,0.00011695224,0.00014507657,0.00011948193,0.0002973498],"genre_scores_gemma":[0.24313618,0.0020653207,0.7471474,0.0012026338,0.00046489688,0.0018538592,0.0011404435,0.00022922087,0.0027600366],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97775686,0.016387403,0.0008052796,0.0028465118,0.0017968996,0.00040716268],"domain_scores_gemma":[0.91832274,0.0663108,0.004533238,0.0076109944,0.0029236395,0.00029855664],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.038565397,0.0013737568,0.0031146116,0.0024788338,0.0007921691,0.002157422,0.0035104793,0.002766263,0.0027218033],"category_scores_gemma":[0.12303817,0.00086412305,0.0031819118,0.0036093313,0.0025278763,0.003278545,0.0019712932,0.003929929,0.00061872107],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004017968,0.00026295902,0.024347931,0.0015747034,0.0023228251,0.00046907566,0.0008124863,0.17631502,0.0018850409,0.31517884,0.008494057,0.46793526],"study_design_scores_gemma":[0.00024621072,0.00032364376,0.007238534,0.00037779196,0.0009612926,0.00044902688,0.00018769625,0.5351745,0.0028694326,0.43918306,0.012856536,0.0001322305],"about_ca_topic_score_codex":0.003744056,"about_ca_topic_score_gemma":0.0025853037,"teacher_disagreement_score":0.038565397,"about_ca_system_score_codex":0.0013740727,"about_ca_system_score_gemma":0.0045259693,"threshold_uncertainty_score":0.20395577},"labels":[],"label_agreement":null},{"id":"W4311277378","doi":"10.1002/sim.9621","title":"An exact regression‐based approach for the estimation of natural direct and indirect effects with a binary outcome and a continuous mediator","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; Université du Québec à Montréal","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Outcome (game theory); Binary number; Regression; Estimation; Econometrics; Statistics; Regression analysis; Computer science; Natural (archaeology); Mathematics; Mathematical economics; Economics; Biology","score_opus":0.035421275230070264,"score_gpt":0.38196971108714584,"score_spread":0.34654843585707557,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4311277378","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0007346316,0.0001133994,0.99870515,0.000054401633,0.000013900672,0.00004128877,0.00004060363,0.00006261799,0.00023400877],"genre_scores_gemma":[0.07089354,0.000558095,0.92578703,0.00028372795,0.000079154044,0.0007546179,0.00022700832,0.00008935647,0.0013274646],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9716186,0.022051306,0.0010342641,0.0024672756,0.0024910271,0.00033755833],"domain_scores_gemma":[0.9566014,0.034794323,0.0023886722,0.004862504,0.0011679779,0.00018512234],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03455978,0.0012520726,0.0022120327,0.0022012198,0.0004888197,0.0017079471,0.003815175,0.0016264885,0.006470163],"category_scores_gemma":[0.09445259,0.0010051879,0.0026814435,0.0024510685,0.0020461883,0.0029522004,0.0033934533,0.0038143126,0.0009677892],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00018468249,0.00015215359,0.0073191766,0.00084324833,0.0010497927,0.0004304452,0.0007652183,0.0927529,0.001711436,0.62386405,0.001953853,0.26897302],"study_design_scores_gemma":[0.0001258825,0.00037038256,0.0040706703,0.00027522896,0.00035776556,0.0007079869,0.00014430944,0.4243093,0.0018116747,0.5567995,0.010888942,0.0001383757],"about_ca_topic_score_codex":0.0024176168,"about_ca_topic_score_gemma":0.0020244352,"teacher_disagreement_score":0.03455978,"about_ca_system_score_codex":0.0009431428,"about_ca_system_score_gemma":0.0024653818,"threshold_uncertainty_score":0.18277174},"labels":[],"label_agreement":null},{"id":"W4312118888","doi":"10.1002/sim.9632","title":"Power analyses for stepped wedge designs with multivariate continuous outcomes","year":2022,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"National Center for Advancing Translational Sciences; National Institute on Aging","keywords":"Multivariate statistics; Univariate; Statistics; Multivariate analysis; Estimator; Sample size determination; Multivariate normal distribution; Computer science; Linear model; Econometrics; Mathematics","score_opus":0.6093308194721376,"score_gpt":0.6148700768653204,"score_spread":0.005539257393182728,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4312118888","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0034401927,0.0005883817,0.9894627,0.0003868029,0.00011383561,0.004191334,0.00023303351,0.0002199873,0.0013638288],"genre_scores_gemma":[0.109120734,0.00068365125,0.8612203,0.0006371504,0.00014904949,0.026767237,0.00034705055,0.00024019068,0.00083460455],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8062272,0.17211503,0.0059965723,0.005838934,0.009161959,0.00066032953],"domain_scores_gemma":[0.63322175,0.32692674,0.0130545255,0.0208291,0.005376348,0.00059150154],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.16224812,0.002132346,0.0037389034,0.0028964116,0.0007744982,0.0022911166,0.0035501048,0.0027818244,0.014081974],"category_scores_gemma":[0.45875508,0.001281596,0.003928703,0.0032443933,0.0032164168,0.0046021985,0.0037564752,0.004973347,0.0014812901],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.003687436,0.0004473103,0.0051122825,0.006296412,0.0029860344,0.00033711662,0.0010709896,0.04777887,0.0019063146,0.46653727,0.008195458,0.4556444],"study_design_scores_gemma":[0.0038091613,0.004322881,0.0035021948,0.0019585055,0.0012135532,0.0004740976,0.00029558057,0.19865222,0.004600619,0.7597153,0.021311922,0.00014396115],"about_ca_topic_score_codex":0.00040940157,"about_ca_topic_score_gemma":0.00043404897,"teacher_disagreement_score":0.83775187,"about_ca_system_score_codex":0.0013237788,"about_ca_system_score_gemma":0.0039832084,"threshold_uncertainty_score":0.85806036},"labels":[],"label_agreement":null},{"id":"W4312132984","doi":"10.1002/sim.9451","title":"Issue Information","year":2022,"lang":"en","type":"paratext","venue":"Statistics in Medicine","topic":"Human auditory perception and evaluation","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Citation; Information retrieval; Library science; World Wide Web","score_opus":0.022882675844517697,"score_gpt":0.3267985601856628,"score_spread":0.3039158843411451,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4312132984","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00024921866,0.00077700755,0.0028531428,0.0043589496,0.014159214,0.0005107679,0.036412165,0.0060264785,0.9346531],"genre_scores_gemma":[0.0006420745,0.0005271655,0.0011808183,0.0016672509,0.0023203047,0.00016631401,0.014081275,0.0015391092,0.9778758],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99849725,0.0001984397,0.00011663798,0.00019369721,0.00086021953,0.00013369716],"domain_scores_gemma":[0.99097097,0.0017903633,0.00039778074,0.00076131127,0.0042254976,0.0018540877],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0018688705,0.0014550623,0.0015597296,0.005969684,0.0012297829,0.00770422,0.001646639,0.0024317398,0.91504025],"category_scores_gemma":[0.015052027,0.0006651357,0.0010112696,0.0046658134,0.00055092503,0.0028892732,0.0022885327,0.0018761373,0.88225067],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000011118785,0.00001698641,0.00003801092,0.00011723071,0.000001532071,0.000007755015,0.000005204306,0.000033959976,0.000063316715,0.0011029323,0.9712478,0.027354147],"study_design_scores_gemma":[0.000019103516,0.000013457484,0.00020765614,0.00011331262,0.0000023295872,0.000018876937,0.000013604057,0.000104489176,0.000079811885,0.002062335,0.9973586,0.000006374355],"about_ca_topic_score_codex":0.0019144753,"about_ca_topic_score_gemma":0.003922408,"teacher_disagreement_score":0.084959745,"about_ca_system_score_codex":0.00115201,"about_ca_system_score_gemma":0.003072405,"threshold_uncertainty_score":0.12118471},"labels":[],"label_agreement":null},{"id":"W4317359118","doi":"10.1002/sim.9650","title":"Practical strategies for operationalizing optimal allocation in stratified cluster‐based outcome‐dependent sampling designs","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Institute of Environmental Health Sciences; National Institutes of Health; Grand Challenges Canada; Harvard T.H. Chan School of Public Health; Government of the United Kingdom; Bill and Melinda Gates Foundation; National Heart, Lung, and Blood Institute; United States Agency for International Development","keywords":"Computer science; Statistics; Sampling design; Covariate; Sampling (signal processing); Sample size determination; Cluster sampling; Inverse probability weighting; Outcome (game theory); Weighting; Missing data; Stratified sampling; Adaptive sampling; Mathematical optimization; Mathematics; Estimator; Filter (signal processing); Population; Monte Carlo method","score_opus":0.3702937272011516,"score_gpt":0.5299317334396962,"score_spread":0.15963800623854463,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4317359118","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0027615197,0.000093991686,0.99495584,0.00014937649,0.000023846938,0.0011908123,0.00003912313,0.00008757805,0.00069790264],"genre_scores_gemma":[0.048139922,0.0001346409,0.9474599,0.00012579314,0.000021911124,0.0037623853,0.00008793267,0.000026271695,0.00024112778],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8984617,0.09213491,0.0023237853,0.0024193046,0.0039932537,0.00066702184],"domain_scores_gemma":[0.90719575,0.07480709,0.0050140964,0.0077823405,0.004573615,0.00062710023],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08724955,0.0016286146,0.0019803443,0.0027073273,0.0009191726,0.002143954,0.0031163797,0.0016655204,0.005266905],"category_scores_gemma":[0.21259494,0.0014547318,0.0015607621,0.003198729,0.0028248269,0.0021251512,0.004388712,0.0024176282,0.00083432475],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006967949,0.0004394559,0.0068957573,0.00089365954,0.00050517695,0.000114441,0.0017863093,0.18295872,0.0015905289,0.45619145,0.0034307998,0.34449697],"study_design_scores_gemma":[0.0007038446,0.00106299,0.0028727897,0.0004996368,0.00020248843,0.000087891385,0.00041292337,0.54873514,0.00231377,0.43234986,0.010664051,0.000094590825],"about_ca_topic_score_codex":0.0031088784,"about_ca_topic_score_gemma":0.003573467,"teacher_disagreement_score":0.08724955,"about_ca_system_score_codex":0.00247778,"about_ca_system_score_gemma":0.0066289906,"threshold_uncertainty_score":0.46142524},"labels":[],"label_agreement":null},{"id":"W4317869010","doi":"10.1002/sim.9666","title":"Two‐phase designs with current status data","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Current (fluid); Phase (matter); Computer science; Statistics; Econometrics; Mathematics; Physics; Thermodynamics","score_opus":0.5036974145021728,"score_gpt":0.6117296971940839,"score_spread":0.10803228269191112,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4317869010","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.050863907,0.0016613008,0.8945648,0.0011984573,0.002418367,0.044531904,0.0012764327,0.0004736599,0.0030110946],"genre_scores_gemma":[0.22546461,0.00092961336,0.63511837,0.0013073887,0.00052939117,0.13107659,0.0010408635,0.000082724895,0.0044503976],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9095797,0.076847605,0.0024907559,0.0060458793,0.0036900577,0.0013459357],"domain_scores_gemma":[0.8782189,0.08272252,0.010614136,0.021222105,0.0053695957,0.0018526623],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.091586284,0.0027117713,0.004806522,0.0016983907,0.0011837009,0.0027453627,0.0043581612,0.0046800785,0.013425665],"category_scores_gemma":[0.14710987,0.0019285263,0.004217674,0.0016418385,0.0036092713,0.0028503798,0.0030586934,0.0043067858,0.0019491174],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.10126368,0.011382325,0.020001784,0.006497334,0.0066041797,0.00065752835,0.0016288857,0.107213356,0.005322697,0.44707093,0.00902652,0.2833308],"study_design_scores_gemma":[0.055772614,0.05504646,0.006312516,0.0009520504,0.0033211575,0.00035980257,0.00039323198,0.32293248,0.0049077114,0.51388305,0.03555059,0.00056837115],"about_ca_topic_score_codex":0.0007615827,"about_ca_topic_score_gemma":0.0006284746,"teacher_disagreement_score":0.091586284,"about_ca_system_score_codex":0.0016662318,"about_ca_system_score_gemma":0.0042357463,"threshold_uncertainty_score":0.48436034},"labels":[],"label_agreement":null},{"id":"W4318024824","doi":"10.1002/sim.9658","title":"Propensity score matching after multiple imputation when a confounder has missing data","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada","keywords":"Propensity score matching; Causal inference; Missing data; Imputation (statistics); Confounding; Statistics; Observational study; Average treatment effect; Matching (statistics); Estimator; Computer science; Confidence interval; Inference; Covariate; Econometrics; Mathematics; Artificial intelligence","score_opus":0.4066710686130003,"score_gpt":0.46171335322110957,"score_spread":0.05504228460810928,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4318024824","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0036323944,0.000394262,0.9944389,0.0003704796,0.00009286911,0.00013104576,0.0001336582,0.00027584838,0.00053069263],"genre_scores_gemma":[0.1683509,0.0013486923,0.82534236,0.0007161813,0.00041193032,0.0008765252,0.00091632386,0.00028517272,0.00175201],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.95691305,0.030881766,0.0029298877,0.0036249084,0.004678513,0.0009719844],"domain_scores_gemma":[0.9136254,0.06365073,0.0060184225,0.013983706,0.0023154123,0.00040641692],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05754706,0.0010287468,0.003494705,0.0031785683,0.0012350526,0.0024637838,0.003337221,0.0028482855,0.0056487345],"category_scores_gemma":[0.17819262,0.00083227176,0.0026840586,0.00897083,0.0015570624,0.0035450056,0.003979283,0.0041336734,0.0011610454],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008266235,0.000260155,0.026577843,0.001283521,0.002359579,0.0013579204,0.001160214,0.09248371,0.0018393077,0.26748732,0.01456081,0.5898029],"study_design_scores_gemma":[0.00037989533,0.0003293149,0.009507656,0.00038762315,0.0006293431,0.0012671464,0.00019222732,0.36831313,0.003719857,0.5903515,0.024798598,0.00012363709],"about_ca_topic_score_codex":0.0026193822,"about_ca_topic_score_gemma":0.0019866403,"teacher_disagreement_score":0.94245297,"about_ca_system_score_codex":0.00078783074,"about_ca_system_score_gemma":0.0037175615,"threshold_uncertainty_score":0.3043416},"labels":[],"label_agreement":null},{"id":"W4318067345","doi":"10.1002/sim.9665","title":"Measuring the performance of prediction models to personalize treatment choice","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":43,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada; ZonMw; European Commission; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; National Science Foundation","keywords":"Computer science; Covariate; Calibration; Machine learning; Baseline (sea); Outcome (game theory); Range (aeronautics); Artificial intelligence; Predictive modelling; Data mining; Statistics; Mathematics","score_opus":0.2535139144086483,"score_gpt":0.4251422307621626,"score_spread":0.17162831635351433,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4318067345","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.13509981,0.0037396797,0.84670544,0.005733678,0.00024847302,0.0004245713,0.0016792167,0.0011726444,0.0051964917],"genre_scores_gemma":[0.77792025,0.0010571851,0.21680593,0.0009108179,0.00029263648,0.00040022648,0.001732338,0.0001894263,0.0006911443],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.94419914,0.044689663,0.0020824245,0.0039676153,0.0043250457,0.00073615124],"domain_scores_gemma":[0.527194,0.42897436,0.014881495,0.02306686,0.0045026424,0.0013806317],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.117213525,0.0027981023,0.002224666,0.004157509,0.00097564695,0.0039266488,0.0020512042,0.00397296,0.0020229619],"category_scores_gemma":[0.3105517,0.0010913075,0.0020517656,0.0035751518,0.0026645178,0.0053259647,0.0032485083,0.0050767893,0.0006948784],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012345838,0.00048398436,0.11322618,0.00055868045,0.0027107073,0.000094964395,0.00071655615,0.69426715,0.00079156173,0.036558762,0.005219519,0.14413738],"study_design_scores_gemma":[0.0001051701,0.00034355602,0.01152827,0.00018154667,0.00024567527,0.00011434923,0.0001169612,0.9287987,0.0013966692,0.055615634,0.0014636359,0.000089793044],"about_ca_topic_score_codex":0.0037223822,"about_ca_topic_score_gemma":0.0029091104,"teacher_disagreement_score":0.117213525,"about_ca_system_score_codex":0.0020623852,"about_ca_system_score_gemma":0.002880867,"threshold_uncertainty_score":0},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4318755449","doi":"10.1002/sim.9675","title":"Multistate models as a framework for estimand specification in clinical trials of complex processes","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Inference; Econometrics; Computer science; Estimator; Limiting; Statistics; Artificial intelligence; Mathematics","score_opus":0.7226251117965159,"score_gpt":0.6472965689702329,"score_spread":0.07532854282628298,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4318755449","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00087273517,0.0005080752,0.99685943,0.0009761408,0.000043618882,0.000080934406,0.0000659061,0.0000678844,0.00052528107],"genre_scores_gemma":[0.17414896,0.0031999033,0.8136955,0.0015526434,0.0005875218,0.003476662,0.0004668844,0.00020045375,0.0026714657],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.89692926,0.092945494,0.002418979,0.0029343772,0.003926001,0.00084594684],"domain_scores_gemma":[0.7094476,0.26556742,0.009303296,0.011645234,0.0030907248,0.0009457379],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.11208782,0.002014671,0.0035890073,0.0026505261,0.0008646473,0.004808251,0.0047101495,0.0046926565,0.004646479],"category_scores_gemma":[0.21827535,0.001727119,0.0035715695,0.0032546176,0.00483851,0.005866665,0.004561015,0.011307244,0.0008971748],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000084365216,0.000049709022,0.0009964762,0.00021428628,0.00023900777,0.00011471754,0.00032043929,0.078415826,0.00016384677,0.9010161,0.0010625223,0.017322676],"study_design_scores_gemma":[0.00007548142,0.00011245283,0.0002477564,0.00013216322,0.0000769138,0.000052793715,0.000037007827,0.20313893,0.0001736696,0.7933784,0.0025415104,0.00003283765],"about_ca_topic_score_codex":0.0028487134,"about_ca_topic_score_gemma":0.002619431,"teacher_disagreement_score":0.88791215,"about_ca_system_score_codex":0.0028835903,"about_ca_system_score_gemma":0.0051465915,"threshold_uncertainty_score":0.59278417},"labels":[],"label_agreement":null},{"id":"W4319058127","doi":"10.1002/sim.9663","title":"The Personalised Randomized Controlled Trial: Evaluation of a new trial design","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Menzies Centre for Australian Studies, King's College London, University of London; Medical Research Council Canada; King's College London; Medical Research Council; National Institute for Health and Care Research","keywords":"Randomized controlled trial; Pooling; Randomization; Computer science; Clinical trial; Ranking (information retrieval); Outcome (game theory); Medicine; Intervention (counseling); Machine learning; Artificial intelligence; Mathematics; Surgery; Psychiatry","score_opus":0.32244574998036885,"score_gpt":0.5144343715594577,"score_spread":0.19198862157908886,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4319058127","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.030775668,0.0062628817,0.9298763,0.0053962776,0.0026657104,0.01753636,0.0010117322,0.0010832002,0.0053918026],"genre_scores_gemma":[0.2530475,0.002133051,0.7097323,0.0031448433,0.00062478025,0.02923241,0.00043121737,0.00015917186,0.0014947176],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.64446354,0.33528474,0.0063283555,0.0072666267,0.006060037,0.0005966563],"domain_scores_gemma":[0.6228229,0.32622728,0.013488608,0.02924208,0.0060252747,0.0021938526],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.235893,0.0027677207,0.0051674494,0.0019199528,0.00066518673,0.0033235366,0.003711683,0.0065236394,0.011055361],"category_scores_gemma":[0.35405624,0.0013195003,0.0057036486,0.0018427093,0.0032545077,0.006694874,0.0036413288,0.0052721784,0.00091091363],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.12156052,0.0022410282,0.007757852,0.021125901,0.033511534,0.0005882883,0.0011599639,0.26104048,0.0016845455,0.23203933,0.009875065,0.30741543],"study_design_scores_gemma":[0.11175642,0.030297419,0.0027539607,0.0037765421,0.015534851,0.0006897585,0.00015680151,0.523579,0.0024160661,0.28676274,0.02185385,0.00042261273],"about_ca_topic_score_codex":0.00078186847,"about_ca_topic_score_gemma":0.00064619037,"teacher_disagreement_score":0.764107,"about_ca_system_score_codex":0.0028752484,"about_ca_system_score_gemma":0.0048553594,"threshold_uncertainty_score":0.94228},"labels":[],"label_agreement":null},{"id":"W4319063464","doi":"10.1002/sim.9677","title":"Conditional concordance‐assisted learning under matched case‐control design for combining biomarkers for population screening","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Roche (Canada)","funders":"University of Texas Health Science Center at Houston; National Center for Advancing Translational Sciences; University of Texas at Austin; National Institute of Diabetes and Digestive and Kidney Diseases; Center for Clinical and Translational Sciences, University of Texas Health Science Center at Houston; National Cancer Institute; National Institutes of Health; Cancer Prevention and Research Institute of Texas","keywords":"Concordance; Confounding; Logistic regression; Biomarker; Population; Discriminative model; Prostate cancer; Computer science; Medicine; Case-control study; Oncology; Machine learning; Statistics; Cancer; Internal medicine; Mathematics; Environmental health; Biology","score_opus":0.530148671326099,"score_gpt":0.5642905653385226,"score_spread":0.034141894012423624,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4319063464","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.013357737,0.00023367521,0.98501265,0.00013960147,0.00004570069,0.000325258,0.00014191946,0.00032320424,0.00042026074],"genre_scores_gemma":[0.49998885,0.0004624105,0.4923781,0.0005568729,0.00024696268,0.0021015722,0.0011631269,0.00012376292,0.002978407],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9778863,0.015884016,0.0006820193,0.0026841448,0.0023178882,0.0005454663],"domain_scores_gemma":[0.90410364,0.077980876,0.004811972,0.0065572252,0.005478362,0.0010679803],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.051649623,0.0014966825,0.0029103889,0.0021118652,0.00076115824,0.0016125997,0.0037319297,0.00204808,0.004020549],"category_scores_gemma":[0.1065256,0.0010393062,0.0017963765,0.0019518111,0.00293836,0.0019414121,0.0030174933,0.0031772282,0.00065618486],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0037394445,0.0006988857,0.021504471,0.0007731133,0.0012439925,0.0005977245,0.00041654977,0.5796946,0.0031141255,0.09831539,0.0028589184,0.2870427],"study_design_scores_gemma":[0.00015619176,0.00028546964,0.001149346,0.000027747952,0.00011025038,0.00007577831,0.000010021323,0.97706443,0.0009733237,0.019555302,0.0005602325,0.00003192713],"about_ca_topic_score_codex":0.0032204308,"about_ca_topic_score_gemma":0.0018010066,"teacher_disagreement_score":0.051649623,"about_ca_system_score_codex":0.0011902562,"about_ca_system_score_gemma":0.002774193,"threshold_uncertainty_score":0.27315253},"labels":[],"label_agreement":null},{"id":"W4321351204","doi":"10.1002/sim.9685","title":"Impute‐then‐exclude versus exclude‐then‐impute: Lessons when imputing a variable used both in cohort creation and as an independent variable in the analysis model","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Sunnybrook Hospital; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Heart and Stroke Foundation of Canada","keywords":"Missing data; Imputation (statistics); Statistics; Sample size determination; Variable (mathematics); Random variable; Computer science; Mathematics; Medicine; Econometrics","score_opus":0.09486507976885547,"score_gpt":0.4341099087394667,"score_spread":0.33924482897061126,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4321351204","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02355164,0.0014659115,0.9598555,0.010834203,0.0003800208,0.00060861715,0.0003377505,0.0005308304,0.002435588],"genre_scores_gemma":[0.22770236,0.0011330406,0.7617737,0.005136517,0.0005627738,0.0010850031,0.00046918372,0.0008238064,0.0013135187],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.6377845,0.33990732,0.005363295,0.006837366,0.008649953,0.001457675],"domain_scores_gemma":[0.27265966,0.6755845,0.00822499,0.033610437,0.0085193375,0.0014010995],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.38047433,0.0021302516,0.0046919635,0.0024656293,0.002169766,0.0064564385,0.0077979234,0.0048490968,0.0034505036],"category_scores_gemma":[0.63441414,0.0024472042,0.0055680606,0.0038851537,0.0065036165,0.007279988,0.0061604492,0.014458095,0.0008525943],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024644542,0.0005356178,0.07610411,0.0017073705,0.005961111,0.0014324836,0.010387472,0.12085008,0.00072883,0.29715756,0.023181556,0.45948946],"study_design_scores_gemma":[0.0008143933,0.0011420569,0.0116552785,0.0014691114,0.0007872205,0.0010699343,0.0021833165,0.47461182,0.0037761186,0.47572103,0.026305338,0.00046444059],"about_ca_topic_score_codex":0.008383493,"about_ca_topic_score_gemma":0.010068768,"teacher_disagreement_score":0.6195257,"about_ca_system_score_codex":0.0022787787,"about_ca_system_score_gemma":0.0067448583,"threshold_uncertainty_score":0.76398546},"labels":[],"label_agreement":null},{"id":"W4321605231","doi":"10.1002/sim.9691","title":"Identifying important gene signatures of BMI using network structure‐aided nonparametric quantile regression","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bioinformatics and Genomic Networks","field":"Biochemistry, Genetics and Molecular Biology","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"National Institute of Arthritis and Musculoskeletal and Skin Diseases; National Institute of Diabetes and Digestive and Kidney Diseases; National Heart, Lung, and Blood Institute; National Institutes of Health","keywords":"Quantile regression; Computer science; Nonparametric statistics; Penalty method; Regression; Data mining; Machine learning; Mathematics; Statistics; Mathematical optimization","score_opus":0.021451876655548905,"score_gpt":0.32355374604990617,"score_spread":0.30210186939435724,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4321605231","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09869918,0.00029894526,0.89923227,0.00029457646,0.000017911792,0.000041253872,0.00038772542,0.0002742248,0.0007538252],"genre_scores_gemma":[0.87431604,0.00040157244,0.122249484,0.00013828179,0.00005653509,0.00012897076,0.0010607728,0.0000655369,0.0015827868],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99963045,0.00016359442,0.000010731275,0.00010155978,0.000056000747,0.00003763931],"domain_scores_gemma":[0.9986117,0.0008803144,0.00022840488,0.000116368676,0.000118871234,0.000044194607],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0012543441,0.00052693393,0.0005656384,0.0008604312,0.0002502693,0.00048992323,0.0006967939,0.00041676493,0.0011294247],"category_scores_gemma":[0.0042079687,0.00020852449,0.00059730007,0.0008301123,0.00044442381,0.00048579994,0.00067034346,0.0008660157,0.0002197903],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002777541,0.00016843868,0.04208027,0.00018145434,0.00022131344,0.000252084,0.00012296283,0.7707848,0.015995571,0.023192132,0.0020434242,0.14467987],"study_design_scores_gemma":[0.000011349096,0.000017627968,0.004650477,0.0000063632738,0.000017604985,0.000033366363,0.000014110245,0.9828531,0.0008242604,0.011191526,0.0003720264,0.00000822538],"about_ca_topic_score_codex":0.002655963,"about_ca_topic_score_gemma":0.003429609,"teacher_disagreement_score":0.002655963,"about_ca_system_score_codex":0.000418481,"about_ca_system_score_gemma":0.0006515833,"threshold_uncertainty_score":0.006633699},"labels":[],"label_agreement":null},{"id":"W4323350148","doi":"10.1002/sim.9704","title":"A two‐level copula joint model for joint analysis of longitudinal and competing risks data","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary; Alberta Health Services","funders":"Natural Sciences and Engineering Research Council of Canada; KU Leuven","keywords":"Copula (linguistics); Econometrics; Covariate; Quantile; Computer science; Statistics; Bayesian probability; Marginal model; Mathematics; Regression analysis","score_opus":0.6336928598209846,"score_gpt":0.5331565992678886,"score_spread":0.10053626055309595,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4323350148","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0020662085,0.00027166895,0.99674267,0.00022737811,0.000033322245,0.000035480545,0.00015776254,0.00012804347,0.00033748028],"genre_scores_gemma":[0.31004646,0.0026087863,0.6743409,0.0010692712,0.000531102,0.0014760849,0.0021649122,0.00056566176,0.0071968245],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9927227,0.004068957,0.00030566982,0.0014741973,0.0008927995,0.0005356547],"domain_scores_gemma":[0.9849555,0.0104787825,0.0013648673,0.0014576067,0.0013411107,0.0004021157],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011583716,0.001808517,0.0028725418,0.0021992798,0.0009307647,0.0028171383,0.0040922826,0.002353946,0.0040054587],"category_scores_gemma":[0.030282304,0.0012627424,0.0029794925,0.0033577892,0.00178531,0.0038533933,0.0028338241,0.004814107,0.0013601888],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013376957,0.00015181718,0.008919245,0.00030147308,0.0007934891,0.00065077026,0.0004432107,0.56948453,0.001262361,0.35465017,0.006873861,0.056335326],"study_design_scores_gemma":[0.000019858282,0.000038601615,0.0008979284,0.00002658754,0.00007509269,0.00011351781,0.000029031347,0.9306988,0.00016614297,0.06565637,0.0022427088,0.000035396522],"about_ca_topic_score_codex":0.010362605,"about_ca_topic_score_gemma":0.0060169874,"teacher_disagreement_score":0.011583716,"about_ca_system_score_codex":0.0017519481,"about_ca_system_score_gemma":0.0031159148,"threshold_uncertainty_score":0.061261296},"labels":[],"label_agreement":null},{"id":"W4360938449","doi":"10.1002/sim.9715","title":"Penalized maximum likelihood inference under the mixture cure model in sparse data","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Lunenfeld-Tanenbaum Research Institute; University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Ontario Institute for Cancer Research","keywords":"Inference; Maximum likelihood; Expectation–maximization algorithm; Computer science; Mixture model; Statistics; Econometrics; Mathematics; Artificial intelligence","score_opus":0.22343168482760778,"score_gpt":0.4621887924704388,"score_spread":0.23875710764283103,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4360938449","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.016267302,0.00020674267,0.9824938,0.00028292267,0.000016586253,0.000055312034,0.00012167854,0.00022365214,0.00033190261],"genre_scores_gemma":[0.53619313,0.00058708247,0.45878798,0.00046593678,0.00020243645,0.0005431277,0.0010401449,0.00020993446,0.0019702513],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9917801,0.006072872,0.00026067422,0.0010643108,0.0006208608,0.0002012301],"domain_scores_gemma":[0.91093946,0.080364324,0.0034118195,0.0031300732,0.001738394,0.0004160071],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.021205263,0.0010829804,0.0020056865,0.0018096665,0.0005578705,0.0018266767,0.0026803303,0.0018893274,0.0024853565],"category_scores_gemma":[0.11126916,0.0011423173,0.0016322,0.0014555487,0.0024013666,0.0022465396,0.0023174942,0.0025429802,0.00038620253],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00054632674,0.00009071114,0.012080123,0.0004071314,0.00047044773,0.00034216518,0.0003383093,0.809541,0.0010932356,0.090212315,0.0022759882,0.08260223],"study_design_scores_gemma":[0.000045435125,0.000032369793,0.0009163162,0.00003805541,0.000038504433,0.00006516355,0.00001649045,0.9531086,0.00032958545,0.044935577,0.0004556198,0.000018182294],"about_ca_topic_score_codex":0.003942642,"about_ca_topic_score_gemma":0.0027637517,"teacher_disagreement_score":0.021205263,"about_ca_system_score_codex":0.0010040193,"about_ca_system_score_gemma":0.0013650131,"threshold_uncertainty_score":0.11214548},"labels":[],"label_agreement":null},{"id":"W4362455928","doi":"10.1002/sim.9709","title":"New late‐emphasis and combination tests based on infimum and supremum logrank statistics with application in oncology trials","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Ligue Contre le Cancer","keywords":"Infimum and supremum; Statistics; Log-rank test; Medicine; Clinical trial; Hazard; Survival analysis; Statistical hypothesis testing; Rank (graph theory); Computer science; Mathematics; Internal medicine","score_opus":0.36581877912489835,"score_gpt":0.575420074643495,"score_spread":0.20960129551859663,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4362455928","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.027908534,0.0019028193,0.96577317,0.0011215489,0.00018572422,0.00017986886,0.00021051634,0.00043476306,0.0022829804],"genre_scores_gemma":[0.44650978,0.0011838826,0.5480916,0.00064669125,0.0007384924,0.0008942153,0.00026015518,0.00019915777,0.0014759985],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.94688046,0.044400163,0.0015908313,0.0018814835,0.004774741,0.00047226905],"domain_scores_gemma":[0.6373926,0.33383054,0.012053681,0.009198167,0.005529008,0.001995994],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.058782317,0.0012006973,0.002687441,0.0037331814,0.0005099912,0.002733091,0.0020789544,0.0017028667,0.0030746097],"category_scores_gemma":[0.21507879,0.0005752863,0.0017239813,0.0029944372,0.004210629,0.0050537726,0.0023995806,0.0045147873,0.00042808478],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.003449056,0.0005162825,0.022463182,0.0010122593,0.0008839833,0.0006240287,0.0005905858,0.110610805,0.0034777257,0.4873138,0.0048827324,0.36417565],"study_design_scores_gemma":[0.0005746014,0.0032556178,0.006863067,0.00023219781,0.00029785087,0.00065863953,0.00014996286,0.49466377,0.0032732666,0.4828664,0.006992431,0.00017213242],"about_ca_topic_score_codex":0.00037595668,"about_ca_topic_score_gemma":0.00038360542,"teacher_disagreement_score":0.058782317,"about_ca_system_score_codex":0.0011327405,"about_ca_system_score_gemma":0.001980827,"threshold_uncertainty_score":0.31087434},"labels":[],"label_agreement":null},{"id":"W4362458453","doi":"10.1002/sim.9721","title":"A Bayesian approach for two‐stage multivariate Mendelian randomization with mixed outcomes","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic and phenotypic traits in livestock","field":"Biochemistry, Genetics and Molecular Biology","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Ottawa; Princess Margaret Cancer Centre; Ottawa Hospital; Queen's University; University of Toronto; University Health Network","funders":"Princess Margaret Cancer Foundation","keywords":"Mendelian randomization; Multivariate statistics; Bayesian probability; Randomization; Multivariate analysis; Statistics; Computer science; Econometrics; Mathematics; Randomized controlled trial; Internal medicine; Medicine; Biology; Genetic variants; Genetics","score_opus":0.01779488646499171,"score_gpt":0.3049085384656339,"score_spread":0.2871136520006422,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4362458453","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0009247254,0.00019530265,0.99778444,0.00020098058,0.000046856232,0.00018546692,0.00013219526,0.000250546,0.00027962663],"genre_scores_gemma":[0.037992973,0.00050397305,0.9565303,0.00041132842,0.00022087444,0.0016133751,0.00058083463,0.00026271574,0.0018836147],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9783176,0.016803859,0.0007127576,0.001791353,0.0017741406,0.0006002209],"domain_scores_gemma":[0.9654944,0.028928153,0.0014659703,0.0017570426,0.0017726625,0.00058172306],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.031371865,0.0021132866,0.004389485,0.0031456773,0.0013982793,0.0025796143,0.0052123666,0.002636478,0.013426094],"category_scores_gemma":[0.054431386,0.0018872388,0.004645613,0.0035672737,0.0023602545,0.0028728957,0.0040025394,0.005184988,0.0017044498],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013176975,0.00030079222,0.00465294,0.00086126773,0.0015322192,0.0007595538,0.0006010754,0.23720199,0.0015592154,0.46452814,0.009034291,0.27765086],"study_design_scores_gemma":[0.00038637454,0.00015320054,0.0005896595,0.00011131343,0.0002730236,0.00019318514,0.000038258193,0.762731,0.00046533384,0.22742206,0.00754979,0.00008683413],"about_ca_topic_score_codex":0.012419866,"about_ca_topic_score_gemma":0.013656779,"teacher_disagreement_score":0.031371865,"about_ca_system_score_codex":0.002224174,"about_ca_system_score_gemma":0.0066860677,"threshold_uncertainty_score":0.16591227},"labels":[],"label_agreement":null},{"id":"W4362507535","doi":"10.1002/sim.9733","title":"Online error rate control for platform trials","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"NIHR Cambridge Biomedical Research Centre; Innovative Medicines Initiative; Medical Research Council Canada; European Commission; Department of Health and Social Care; National Institute for Health and Care Research; Springworks Therapeutics; Medical Research Council; European Federation of Pharmaceutical Industries and Associations","keywords":"False discovery rate; Bonferroni correction; Word error rate; Computer science; Multiple comparisons problem; Type I and type II errors; Null hypothesis; Statistical hypothesis testing; Protocol (science); Control (management); Error detection and correction; Statistics; Artificial intelligence; Algorithm; Mathematics; Medicine","score_opus":0.7550225175797985,"score_gpt":0.6523072582200737,"score_spread":0.10271525935972481,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4362507535","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0014127614,0.00072871643,0.99163824,0.0007906499,0.00047833892,0.0017847979,0.00014800932,0.001325579,0.0016928588],"genre_scores_gemma":[0.124516696,0.0007917133,0.85307986,0.002062203,0.00073246413,0.01470889,0.0003428266,0.0010598963,0.0027054355],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.70782435,0.2459153,0.010600588,0.011768287,0.022253342,0.0016382267],"domain_scores_gemma":[0.40744945,0.47708857,0.039782465,0.054584034,0.018481398,0.0026141768],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.26264194,0.003352945,0.003844837,0.0031072383,0.0013900633,0.0050694174,0.006534072,0.005151513,0.010704326],"category_scores_gemma":[0.6086341,0.0015775603,0.0028666665,0.0040767514,0.005698524,0.0056029465,0.0057613226,0.010458767,0.0031981007],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0066580386,0.000841064,0.006539587,0.003269346,0.0019072554,0.0005109593,0.001959804,0.07603005,0.004076088,0.3203757,0.021994486,0.55583763],"study_design_scores_gemma":[0.004461215,0.0031512647,0.0054147104,0.0020361885,0.00092512846,0.00060467795,0.00023689232,0.46414676,0.013256045,0.4403691,0.06508154,0.0003164819],"about_ca_topic_score_codex":0.0013902368,"about_ca_topic_score_gemma":0.0010283084,"teacher_disagreement_score":0.26264194,"about_ca_system_score_codex":0.0034213997,"about_ca_system_score_gemma":0.009386048,"threshold_uncertainty_score":0.9092938},"labels":[],"label_agreement":null},{"id":"W4362658038","doi":"10.1002/sim.9734","title":"Point estimation for adaptive trial designs <scp>II</scp>: Practical considerations and guidance","year":2023,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"NIHR Cambridge Biomedical Research Centre; Medical Research Council; National Institute for Health and Care Research; Medical Research Council Canada; Department of Health and Social Care; Cancer Research UK; Health and Care Research Wales","keywords":"Estimation; Computer science; Point estimation; Point (geometry); Econometrics; Statistics; Mathematical optimization; Mathematics; Economics","score_opus":0.8358348090652867,"score_gpt":0.6690181020821543,"score_spread":0.16681670698313245,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4362658038","genre_codex":"methods","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00046354055,0.016020214,0.91208786,0.045571692,0.004070856,0.0035604856,0.0011308743,0.0017725524,0.015321886],"genre_scores_gemma":[0.0077807875,0.010106318,0.9464024,0.01778984,0.0036444091,0.010318211,0.00056188984,0.00072801654,0.002668129],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.62960196,0.31961998,0.02104149,0.005237798,0.023588426,0.0009103673],"domain_scores_gemma":[0.29597315,0.63191664,0.015573361,0.02411288,0.030935638,0.0014884131],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2512707,0.0028387264,0.004692654,0.0045326017,0.0014126517,0.007906327,0.0073160953,0.017460661,0.015636308],"category_scores_gemma":[0.6518013,0.0029469389,0.004798477,0.0058392994,0.008351244,0.0077721477,0.0046237605,0.02438579,0.01459033],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007189919,0.00014287439,0.00068888674,0.0078098862,0.00034918654,0.00038740993,0.0015386057,0.014031613,0.0005107025,0.33674806,0.29012528,0.3469485],"study_design_scores_gemma":[0.0012338292,0.00063886784,0.0011419323,0.015596004,0.00024307356,0.0009647739,0.0002054653,0.04646418,0.001346409,0.4974116,0.43439308,0.00036084553],"about_ca_topic_score_codex":0.0057536634,"about_ca_topic_score_gemma":0.004508335,"teacher_disagreement_score":0.7487293,"about_ca_system_score_codex":0.0046565486,"about_ca_system_score_gemma":0.014439781,"threshold_uncertainty_score":0.92331654},"labels":[],"label_agreement":null},{"id":"W4365459023","doi":"10.1002/sim.9727","title":"Causal inference with longitudinal data subject to irregular assessment times","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital; Institute for Clinical Evaluative Sciences; University of Toronto; SickKids Foundation; Hospital for Sick Children; Public Health Ontario","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Causal inference; Inference; Subject (documents); Computer science; Longitudinal data; Statistics; Econometrics; Statistical inference; Artificial intelligence; Mathematics; Data mining","score_opus":0.21944373833936742,"score_gpt":0.5103172274469393,"score_spread":0.2908734891075719,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4365459023","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029566465,0.00033755696,0.96820974,0.0006636418,0.000046798126,0.00008815279,0.0001888025,0.00015996711,0.0007388608],"genre_scores_gemma":[0.5357679,0.0009439046,0.45820135,0.00030861387,0.00020626566,0.0007215538,0.0007843283,0.000120394725,0.0029456343],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9787709,0.016322711,0.0008027936,0.0022054857,0.0013722782,0.0005259359],"domain_scores_gemma":[0.7797603,0.19591095,0.009590052,0.011669123,0.002282175,0.0007875099],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.067806885,0.00083703216,0.0019693784,0.002473008,0.0011027273,0.0021265622,0.0030941954,0.0020697666,0.0030282352],"category_scores_gemma":[0.19418086,0.0012182692,0.0026164379,0.0033772113,0.0027695824,0.004694643,0.0035931293,0.0041376525,0.0003130008],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00047941055,0.0001996864,0.030563364,0.0003437915,0.00067904766,0.0007955087,0.0014816143,0.35107532,0.00062219356,0.49458537,0.001515408,0.11765921],"study_design_scores_gemma":[0.00010212562,0.00008131735,0.00235659,0.000057928522,0.00008956278,0.0001420513,0.0001411075,0.58699536,0.000386141,0.40784582,0.001767133,0.000034866986],"about_ca_topic_score_codex":0.010186693,"about_ca_topic_score_gemma":0.007196274,"teacher_disagreement_score":0.067806885,"about_ca_system_score_codex":0.001824857,"about_ca_system_score_gemma":0.002636122,"threshold_uncertainty_score":0.3586014},"labels":[],"label_agreement":null},{"id":"W4376121573","doi":"10.1002/sim.9765","title":"Incorporating biological knowledge in analyses of environmental mixtures and health","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"National Institute of Environmental Health Sciences; Natural Sciences and Engineering Research Council of Canada","keywords":"Interpretability; Prior probability; Computer science; Bayesian probability; Prior information; Nonparametric statistics; Dirichlet distribution; Set (abstract data type); Index (typography); Data mining; Statistics; Econometrics; Machine learning; Mathematics; Artificial intelligence","score_opus":0.23700524302746534,"score_gpt":0.5039231798181436,"score_spread":0.26691793679067827,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4376121573","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00509062,0.00039822562,0.99315083,0.0003824439,0.000015277445,0.000024867462,0.00007677863,0.00007878557,0.0007820963],"genre_scores_gemma":[0.30314502,0.0018426604,0.6912934,0.0005084947,0.00018148808,0.00039382148,0.00040982934,0.00010935809,0.0021159134],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9902508,0.006988132,0.00024421583,0.0010105856,0.0013333231,0.0001728407],"domain_scores_gemma":[0.9551125,0.039254688,0.002386176,0.0023521162,0.0006827872,0.00021180314],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01703657,0.0011004001,0.00118674,0.003432459,0.00068744906,0.0020884313,0.0018099869,0.0018212844,0.0018152309],"category_scores_gemma":[0.061249856,0.0010233715,0.0015629141,0.0022910207,0.0045951046,0.0037750294,0.003728049,0.0028229,0.00038452147],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000116154595,0.00011471063,0.006338337,0.00032726387,0.0002723093,0.00013850511,0.00031226844,0.3399983,0.002260907,0.5270074,0.0009128872,0.1222009],"study_design_scores_gemma":[0.000018098208,0.000053177464,0.0015427781,0.00005059018,0.000038313614,0.000050608713,0.000040204166,0.31703955,0.0010629875,0.67737454,0.002696709,0.000032378703],"about_ca_topic_score_codex":0.0031455099,"about_ca_topic_score_gemma":0.0032220325,"teacher_disagreement_score":0.01703657,"about_ca_system_score_codex":0.0016014712,"about_ca_system_score_gemma":0.0018192199,"threshold_uncertainty_score":0.0900991},"labels":[],"label_agreement":null},{"id":"W4376598528","doi":"10.1002/sim.9764","title":"A Bayesian joint model for compositional mediation effect selection in microbiome data","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Metabolomics and Mass Spectrometry Studies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"School of Medicine, New York University; York University; Washington University in St. Louis","keywords":"Computer science; Bayesian probability; Microbiome; Multivariate statistics; Mediation; Identification (biology); Selection (genetic algorithm); Benchmark (surveying); Model selection; Data mining; Econometrics; Machine learning; Artificial intelligence; Bioinformatics; Biology; Mathematics; Ecology","score_opus":0.028528566420889294,"score_gpt":0.33345932926112437,"score_spread":0.30493076284023507,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4376598528","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009659739,0.00045140632,0.9872885,0.0006879377,0.00006418779,0.00023946726,0.0005791531,0.00031561966,0.0007140263],"genre_scores_gemma":[0.34454304,0.0020642471,0.6349492,0.0016047339,0.0006128324,0.0043752925,0.003439379,0.0003392008,0.008072056],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9797085,0.01505964,0.0005972995,0.0024499963,0.0014873655,0.00069726235],"domain_scores_gemma":[0.9284456,0.063415825,0.0023722888,0.0027885935,0.0022872516,0.0006903632],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.040722653,0.0020032676,0.0040428117,0.0022904768,0.0015234171,0.0029447484,0.006094875,0.0032845123,0.007610691],"category_scores_gemma":[0.077237144,0.0017057599,0.0040124324,0.0035887754,0.0034096343,0.0034620052,0.0035172228,0.005114166,0.0013552906],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012091812,0.0004191221,0.01623412,0.00083666685,0.0015284944,0.00094216084,0.0010562519,0.5187441,0.002580214,0.3136483,0.0064672213,0.13633426],"study_design_scores_gemma":[0.00022741589,0.00015564647,0.0017749324,0.000105361374,0.0002967037,0.00012901209,0.00008749029,0.8744135,0.0005277837,0.11931551,0.0028945864,0.00007200408],"about_ca_topic_score_codex":0.011881834,"about_ca_topic_score_gemma":0.010926841,"teacher_disagreement_score":0.040722653,"about_ca_system_score_codex":0.0019526993,"about_ca_system_score_gemma":0.0048746774,"threshold_uncertainty_score":0.21536458},"labels":[],"label_agreement":null},{"id":"W4377563942","doi":"10.1002/sim.9771","title":"Developing prediction models to estimate the risk of two survival outcomes both occurring: A comparison of techniques","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Chronic Disease Management Strategies","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council Canada; University of Manchester; Medical Research Council; National Institute for Health and Care Research","keywords":"Statistics; Computer science; Econometrics; Survival analysis; Proportional hazards model; Mathematics","score_opus":0.11919425273443744,"score_gpt":0.46984651115818477,"score_spread":0.3506522584237473,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4377563942","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029503435,0.001568466,0.9664515,0.00053594616,0.00005712279,0.00012201417,0.0001524514,0.0005120337,0.0010970129],"genre_scores_gemma":[0.31862667,0.0029119323,0.6756018,0.0003181416,0.00023231284,0.0004734032,0.00052162627,0.0001883163,0.0011258251],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99291575,0.0043146885,0.0003715555,0.00088482746,0.0013078455,0.00020534072],"domain_scores_gemma":[0.936971,0.05417344,0.0026932415,0.0021243645,0.0036854611,0.00035247696],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.023854561,0.001576254,0.0015695586,0.0030737035,0.00045205312,0.001734592,0.00270427,0.0015897849,0.0015804373],"category_scores_gemma":[0.04241727,0.0009144817,0.0025689956,0.0016371974,0.00062529783,0.002397975,0.0022588572,0.0027419548,0.0006340895],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005912099,0.000388382,0.045665585,0.0005610335,0.0014859034,0.00020419231,0.00071243953,0.37942454,0.0012238518,0.022857985,0.0021545659,0.5447303],"study_design_scores_gemma":[0.0000661972,0.00017354183,0.003860264,0.00013586866,0.00014420557,0.00015379308,0.00008317136,0.9793443,0.0007316201,0.013891512,0.0013524825,0.00006309231],"about_ca_topic_score_codex":0.0063004643,"about_ca_topic_score_gemma":0.0029231969,"teacher_disagreement_score":0.023854561,"about_ca_system_score_codex":0.0010657791,"about_ca_system_score_gemma":0.0018755274,"threshold_uncertainty_score":0.12615651},"labels":[],"label_agreement":null},{"id":"W4377940739","doi":"10.1002/sim.9805","title":"The person‐time ratio distribution for the exact monitoring of adverse events: Historical vs surveillance Poisson data","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Pharmacovigilance and Adverse Drug Reactions","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Manitoba Health","funders":"National Institute of General Medical Sciences; Fundação de Amparo à Pesquisa do Estado de Minas Gerais; Conselho Nacional de Desenvolvimento Científico e Tecnológico","keywords":"Poisson distribution; Statistics; Estimator; Relative risk; Confidence interval; Interval (graph theory); Medicine; Mathematics; Econometrics; Computer science","score_opus":0.18267581425771495,"score_gpt":0.4783700553137205,"score_spread":0.29569424105600556,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4377940739","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.18904689,0.0025966777,0.7989428,0.0011424959,0.00021931075,0.0005187307,0.0022376974,0.0007922812,0.0045031006],"genre_scores_gemma":[0.91631746,0.0013827147,0.07640986,0.0004288439,0.00018832492,0.0005429992,0.0021891487,0.00015762307,0.0023830035],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98431087,0.008288438,0.001031874,0.0024243062,0.00347569,0.00046888803],"domain_scores_gemma":[0.9335365,0.04499047,0.0094013885,0.008245284,0.0033778187,0.0004484698],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.033672586,0.0004043853,0.001140602,0.00194347,0.0003358534,0.0015349056,0.0025610388,0.0014276352,0.0024240285],"category_scores_gemma":[0.12686689,0.0004161369,0.0014868652,0.0017888373,0.0016142221,0.002300629,0.0010713934,0.0017207962,0.000636274],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001210382,0.00021818854,0.47018498,0.0010429225,0.0012015596,0.0020403857,0.0025037075,0.07810409,0.0039407895,0.19665866,0.008059234,0.2348351],"study_design_scores_gemma":[0.00016668517,0.0014309791,0.27642286,0.0005230025,0.00075009017,0.009134962,0.0016418402,0.54806936,0.0055645066,0.1321312,0.02381426,0.00035012077],"about_ca_topic_score_codex":0.0051245843,"about_ca_topic_score_gemma":0.0021192753,"teacher_disagreement_score":0.9663274,"about_ca_system_score_codex":0.0011709583,"about_ca_system_score_gemma":0.00089435186,"threshold_uncertainty_score":0.17807978},"labels":[],"label_agreement":null},{"id":"W4379967395","doi":"10.1002/sim.9802","title":"Causal inference for recurrent events via aggregated marginal odds ratio","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Inference; Causal inference; Odds ratio; Odds; Econometrics; Statistics; Marginal structural model; Computer science; Mathematics; Artificial intelligence; Logistic regression","score_opus":0.14131302433752374,"score_gpt":0.4749063944263714,"score_spread":0.33359337008884765,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4379967395","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017609252,0.00047784238,0.98015267,0.00035500465,0.00005908835,0.000116132534,0.00023781741,0.0003006675,0.00069155474],"genre_scores_gemma":[0.5997719,0.0014087127,0.39306426,0.00036329892,0.0003719771,0.0009386434,0.0010887536,0.00017370238,0.0028186764],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98433995,0.0110251345,0.0006233988,0.0025906102,0.0010211017,0.00039980726],"domain_scores_gemma":[0.898275,0.08870178,0.004794291,0.006078235,0.0016485186,0.0005021237],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.033291444,0.0011427287,0.002295355,0.0038843306,0.0006735268,0.0020805132,0.0028376577,0.0014390197,0.00727923],"category_scores_gemma":[0.14748776,0.0008227375,0.0032174995,0.0035544136,0.0021540893,0.005147558,0.0025274267,0.003128722,0.0004820854],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00050924934,0.00026659956,0.034986958,0.00060119596,0.0022719873,0.000520068,0.0010756798,0.123982,0.0007350303,0.6211749,0.0031235015,0.21075301],"study_design_scores_gemma":[0.00011745764,0.00015815167,0.0070206327,0.000099212506,0.0005689349,0.0001703138,0.0001409208,0.43863732,0.0006509249,0.5495216,0.0028536697,0.000060861188],"about_ca_topic_score_codex":0.004442332,"about_ca_topic_score_gemma":0.0031767646,"teacher_disagreement_score":0.033291444,"about_ca_system_score_codex":0.0012567107,"about_ca_system_score_gemma":0.0015055148,"threshold_uncertainty_score":0.17606407},"labels":[],"label_agreement":null},{"id":"W4380537590","doi":"10.1002/sim.9823","title":"Statistical inference for the two‐sample problem under likelihood ratio ordering, with application to the ROC curve estimation","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Receiver operating characteristic; Estimator; Statistics; Statistical inference; Consistency (knowledge bases); Mathematics; Inference; Asymptotic distribution; Sample size determination; Maximum likelihood; Computer science; Applied mathematics; Artificial intelligence","score_opus":0.0702520978945821,"score_gpt":0.4293884576788077,"score_spread":0.3591363597842256,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4380537590","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0021713541,0.0004832466,0.99614954,0.00057463575,0.000041299616,0.00009019548,0.00005492672,0.00009057828,0.0003442157],"genre_scores_gemma":[0.16125327,0.001729907,0.83127934,0.0005737418,0.00075780594,0.0012222638,0.0005501968,0.00021636844,0.002417085],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9700282,0.023351584,0.00094568543,0.0025009313,0.0026989474,0.000474613],"domain_scores_gemma":[0.75631934,0.22871919,0.0052550277,0.0043390235,0.00443982,0.0009276007],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05242128,0.001716882,0.003453796,0.0037332396,0.0012111727,0.0029168662,0.0035786412,0.0034721803,0.0033826032],"category_scores_gemma":[0.23607224,0.0011431919,0.0022802795,0.0038410765,0.0047783754,0.0035604176,0.0029986766,0.0052240817,0.00073974143],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00041659665,0.00018811268,0.0059220092,0.00078390486,0.000397535,0.0008287518,0.00067274895,0.39347455,0.0013653062,0.44498175,0.0051114606,0.14585727],"study_design_scores_gemma":[0.000071955685,0.00008582509,0.00074072904,0.00004634599,0.00003579572,0.00018993013,0.000045465098,0.8050391,0.0003273896,0.1918243,0.0015628167,0.000030416102],"about_ca_topic_score_codex":0.0062918635,"about_ca_topic_score_gemma":0.0038416078,"teacher_disagreement_score":0.05242128,"about_ca_system_score_codex":0.0029794981,"about_ca_system_score_gemma":0.004545197,"threshold_uncertainty_score":0.2772336},"labels":[],"label_agreement":null},{"id":"W4380589797","doi":"10.1002/sim.9811","title":"Sample size considerations for assessing treatment effect heterogeneity in randomized trials with heterogeneous intracluster correlations and variances","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"Patient-Centered Outcomes Research Institute","keywords":"Sample size determination; Univariate; Randomized controlled trial; Statistics; Context (archaeology); Cluster (spacecraft); Multivariate statistics; Sample (material); Outcome (game theory); Econometrics; Mathematics; Computer science; Medicine; Internal medicine","score_opus":0.48912071704197413,"score_gpt":0.5797179367516512,"score_spread":0.09059721970967705,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4380589797","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00230398,0.0024529109,0.98664993,0.0030456523,0.0006192438,0.002288142,0.00024729522,0.00032336716,0.0020694493],"genre_scores_gemma":[0.13118061,0.0018441284,0.8392249,0.0050313645,0.0010438828,0.01992896,0.00041973108,0.00034691772,0.0009796366],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.667017,0.27707306,0.01918804,0.00975805,0.025839176,0.0011247132],"domain_scores_gemma":[0.3136704,0.64284253,0.012919527,0.016859619,0.013098236,0.00060967053],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27127695,0.0021512175,0.003999448,0.004195656,0.0011395301,0.0036217112,0.004577098,0.004981563,0.005056334],"category_scores_gemma":[0.6817406,0.0014842496,0.0042912327,0.002651923,0.0047731493,0.005693965,0.0037094813,0.008380572,0.00088593364],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024076079,0.00026191823,0.008775908,0.005963836,0.0031684372,0.0008799292,0.001925447,0.07192426,0.0029762567,0.51106447,0.019556174,0.37109578],"study_design_scores_gemma":[0.003174008,0.002223225,0.005718177,0.0046023615,0.002063131,0.0011077473,0.00034445166,0.26701728,0.005841614,0.6696857,0.03794645,0.00027578507],"about_ca_topic_score_codex":0.0015160461,"about_ca_topic_score_gemma":0.0012522419,"teacher_disagreement_score":0.72872305,"about_ca_system_score_codex":0.0026510616,"about_ca_system_score_gemma":0.005750598,"threshold_uncertainty_score":0.8986453},"labels":[],"label_agreement":null},{"id":"W4381187244","doi":"10.1002/sim.9824","title":"Deep causal feature extraction and inference with neuroimaging genetic data","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic Associations and Epidemiology","field":"Biochemistry, Genetics and Molecular Biology","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Institute of Biomedical Imaging and Bioengineering; National Institute of General Medical Sciences; National Institute on Aging; Canadian Institutes of Health Research; Minnesota Supercomputing Institute, University of Minnesota; National Institutes of Health; Genentech; IXICO; H. Lundbeck A/S; Pfizer; Novartis Pharmaceuticals Corporation; Eisai; U.S. Department of Defense; Alzheimer's Disease Neuroimaging Initiative; Meso Scale Diagnostics; Servier; BioClinica; Bristol-Myers Squibb; Eli Lilly and Company; Biogen","keywords":"Neuroimaging; Computer science; Causal inference; Imaging genetics; Artificial intelligence; Covariate; Inference; Genome-wide association study; Biobank; Alzheimer's Disease Neuroimaging Initiative; Instrumental variable; Machine learning; Feature (linguistics); Regression; Psychology; Econometrics; Cognition; Bioinformatics; Statistics; Cognitive impairment; Psychiatry; Biology; Mathematics","score_opus":0.03034836853606217,"score_gpt":0.3566172819314706,"score_spread":0.3262689133954084,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4381187244","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.021617599,0.0007266659,0.9740326,0.0010614479,0.000049110866,0.00004253147,0.0008896404,0.0011789845,0.00040141516],"genre_scores_gemma":[0.6488214,0.0010708125,0.34109858,0.0010653386,0.00029222431,0.00027453768,0.0047210907,0.00020345794,0.0024525838],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99818975,0.00081421196,0.0001305322,0.00054945366,0.00018268485,0.00013338322],"domain_scores_gemma":[0.9893231,0.00877568,0.00063633407,0.00074823684,0.00036970538,0.00014690781],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0044926247,0.0010224608,0.00136107,0.0017205734,0.00048220326,0.0011341939,0.0017273749,0.0011869209,0.0018779078],"category_scores_gemma":[0.019716427,0.000664414,0.0019368855,0.001519627,0.0009694748,0.0011470611,0.0015572106,0.0027919621,0.00038663446],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005060943,0.0002787674,0.03363254,0.00040432115,0.0011033199,0.0011593738,0.00026602554,0.5406762,0.004293592,0.056603488,0.009293536,0.35178283],"study_design_scores_gemma":[0.000052410498,0.000036054887,0.0021031264,0.000035597575,0.00008200887,0.00010193378,0.000020174104,0.90560436,0.0010711581,0.089374155,0.0014992773,0.000019822843],"about_ca_topic_score_codex":0.009481919,"about_ca_topic_score_gemma":0.010209618,"teacher_disagreement_score":0.009481919,"about_ca_system_score_codex":0.001055746,"about_ca_system_score_gemma":0.0018107867,"threshold_uncertainty_score":0.023759544},"labels":[],"label_agreement":null},{"id":"W4381469694","doi":"10.1002/sim.9831","title":"Permutation‐based multiple testing corrections for P$$ P $$‐values and confidence intervals for cluster randomized trials","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council Canada; Medical Research Council; National Institute for Health and Care Research","keywords":"Confidence interval; Statistics; Permutation (music); Resampling; Bonferroni correction; Multiple comparisons problem; Mathematics; Inference; Standard error; Type I and type II errors; Confidence distribution; Statistical hypothesis testing; Computer science; Artificial intelligence","score_opus":0.7094685578629825,"score_gpt":0.6281742164474525,"score_spread":0.08129434141552994,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4381469694","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0015835834,0.0007881503,0.99497813,0.00045981962,0.00031597723,0.0004542147,0.00013445961,0.0004484736,0.00083708914],"genre_scores_gemma":[0.08876928,0.00085278135,0.900132,0.00085312827,0.00038687044,0.00675793,0.00037389144,0.0009330639,0.00094105094],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.77993476,0.18238546,0.008234064,0.008454984,0.019813336,0.0011773489],"domain_scores_gemma":[0.40048376,0.5243057,0.017160738,0.044529002,0.012731831,0.0007889863],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1558834,0.0021318204,0.0038373438,0.0051401523,0.0016362547,0.0030935907,0.005928422,0.0034182651,0.0066410783],"category_scores_gemma":[0.61342484,0.0013501476,0.0051752822,0.005938341,0.0052668024,0.003846558,0.0037604908,0.011281275,0.0014174579],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024211728,0.00035312114,0.009126748,0.005136121,0.004753938,0.0009610154,0.0016953163,0.076370105,0.0021849747,0.33166856,0.023256542,0.54207253],"study_design_scores_gemma":[0.0012510304,0.0018124771,0.007353015,0.001962218,0.0015490192,0.0013462758,0.00031486308,0.29769355,0.01065473,0.64117205,0.034586027,0.00030468998],"about_ca_topic_score_codex":0.0013253822,"about_ca_topic_score_gemma":0.00087658083,"teacher_disagreement_score":0.84411657,"about_ca_system_score_codex":0.00223011,"about_ca_system_score_gemma":0.005409237,"threshold_uncertainty_score":0.82440007},"labels":[],"label_agreement":null},{"id":"W4381716816","doi":"10.1002/sim.9768","title":"A combined multilevel factor analysis and covariance regression model with mixed effects in the mean and variance structure","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Mental Health Research Topics","field":"Psychology","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Impact","funders":"FP7 Health","keywords":"Covariance; Multilevel model; Factor regression model; Regression analysis; Statistics; Hierarchical database model; Regression; Factor analysis; Covariance matrix; Bayesian probability; Mathematics; Analysis of covariance; Computer science; Bayesian multivariate linear regression; Proper linear model; Data mining","score_opus":0.0573277386348996,"score_gpt":0.43326040256639037,"score_spread":0.3759326639314908,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4381716816","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007436038,0.00039498883,0.98777044,0.0009427402,0.00012405399,0.00025725158,0.0011639099,0.00062394806,0.0012867199],"genre_scores_gemma":[0.2146408,0.00076349534,0.77085453,0.00058155326,0.00024728745,0.0024030989,0.0030140132,0.00044208454,0.0070531075],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9846096,0.010550602,0.0005574214,0.0026256214,0.0011133772,0.00054335356],"domain_scores_gemma":[0.9854806,0.008701079,0.0010624164,0.0016786951,0.0027441436,0.00033306054],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01872271,0.002120661,0.0029886686,0.002624109,0.0011709861,0.0029119581,0.0035953324,0.0019962273,0.008805535],"category_scores_gemma":[0.03544154,0.0013689043,0.006372272,0.004590426,0.0011163466,0.003230161,0.0027532396,0.004417168,0.0031977908],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007181357,0.0005136574,0.03937679,0.00090520724,0.003998773,0.00038484737,0.002107154,0.17310426,0.0025165577,0.35370535,0.026401334,0.39626807],"study_design_scores_gemma":[0.00023679942,0.00036532147,0.00834461,0.0002221879,0.0010147837,0.00015000471,0.00022369318,0.7838469,0.0008615423,0.18508327,0.019453002,0.00019784525],"about_ca_topic_score_codex":0.021830425,"about_ca_topic_score_gemma":0.026529904,"teacher_disagreement_score":0.021830425,"about_ca_system_score_codex":0.002130202,"about_ca_system_score_gemma":0.0038312385,"threshold_uncertainty_score":0.09901637},"labels":[],"label_agreement":null},{"id":"W4382182859","doi":"10.1002/sim.9818","title":"Handling misclassified stratification variables in the analysis of randomised trials with continuous outcomes","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Health and Medical Research Council; Medical Research Council Canada; Medical Research Council; University of Adelaide","keywords":"Covariate; Stratification (seeds); Statistics; Econometrics; Medicine; Mathematics","score_opus":0.5420455474370366,"score_gpt":0.5745915147372646,"score_spread":0.03254596730022796,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4382182859","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014409115,0.012525943,0.9461335,0.0077651837,0.0041355565,0.010297402,0.000965287,0.001343416,0.0024246294],"genre_scores_gemma":[0.27324572,0.004481189,0.68018943,0.010719004,0.0017307722,0.02626111,0.0011204875,0.0007299722,0.0015222925],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.20180196,0.71946365,0.042129334,0.013018772,0.022054434,0.001531895],"domain_scores_gemma":[0.12356092,0.742989,0.061065115,0.06176397,0.009745958,0.0008751226],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6282866,0.0030896214,0.008390382,0.0048615453,0.0018985152,0.007570557,0.0047553037,0.006173291,0.0056586713],"category_scores_gemma":[0.8331282,0.0024371475,0.013299325,0.0064912857,0.007539994,0.0077825394,0.005892116,0.010476879,0.0012162136],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.017040107,0.00061773945,0.04844567,0.037601765,0.04172313,0.001120879,0.011310636,0.06584992,0.002009605,0.2627118,0.027882585,0.48368606],"study_design_scores_gemma":[0.010037348,0.005279934,0.018141944,0.020676801,0.015820658,0.0010002813,0.00074738386,0.11873832,0.007821147,0.7382819,0.06222965,0.0012246104],"about_ca_topic_score_codex":0.0024767807,"about_ca_topic_score_gemma":0.0026835585,"teacher_disagreement_score":0.3717134,"about_ca_system_score_codex":0.004567211,"about_ca_system_score_gemma":0.009907741,"threshold_uncertainty_score":0.45838886},"labels":[],"label_agreement":null},{"id":"W4384406729","doi":"10.1002/sim.9848","title":"Inference for covariate‐adjusted time‐dependent prognostic accuracy measures","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Covariate; Discriminative model; Receiver operating characteristic; Estimator; Computer science; Inference; Statistics; Econometrics; Artificial intelligence; Machine learning; Mathematics","score_opus":0.2147288485494072,"score_gpt":0.45917408079072347,"score_spread":0.24444523224131628,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4384406729","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.026613278,0.0007853205,0.97070616,0.0006407643,0.00006246726,0.00009414553,0.00037477174,0.00026109794,0.00046194487],"genre_scores_gemma":[0.6208011,0.0017581797,0.37031388,0.00073698076,0.0003772055,0.0008663663,0.0029399465,0.00022345917,0.0019830065],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9835338,0.011428312,0.0007809007,0.0022486185,0.0014841208,0.0005243044],"domain_scores_gemma":[0.73249793,0.24119659,0.008995882,0.011854277,0.004638648,0.0008167486],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05192717,0.0012908445,0.0021880693,0.0033589592,0.000534129,0.002681474,0.003363381,0.0026136418,0.0023730926],"category_scores_gemma":[0.25004095,0.0010407374,0.0023405065,0.003540311,0.0022939076,0.004017915,0.002185439,0.0044888086,0.0005481001],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00058771815,0.00030535215,0.061104648,0.00048468422,0.0015931715,0.0004084028,0.0004970894,0.5865908,0.0015574,0.15645294,0.0047746394,0.18564315],"study_design_scores_gemma":[0.00006328308,0.00010640436,0.0067569357,0.00009678503,0.00014869675,0.00014105983,0.000039071958,0.88196266,0.00080288976,0.108748294,0.0010924649,0.00004147191],"about_ca_topic_score_codex":0.0033071311,"about_ca_topic_score_gemma":0.0020660188,"teacher_disagreement_score":0.05192717,"about_ca_system_score_codex":0.0017145191,"about_ca_system_score_gemma":0.0017528597,"threshold_uncertainty_score":0.27462035},"labels":[],"label_agreement":null},{"id":"W4385235004","doi":"10.1002/sim.9855","title":"A time‐dependent Poisson‐Gamma model for recruitment forecasting in multicenter studies","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University Health Centre; McGill University","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Styrelsen för Internationellt Utvecklingssamarbete","keywords":"Poisson distribution; Gamma process; Constant (computer programming); Gamma distribution; Generalization; Econometrics; Poisson process; Bayesian probability; Computer science; Poisson regression; Statistics; Compound Poisson process; Count data; Range (aeronautics); Stochastic modelling; Mathematics; Demography; Population; Engineering","score_opus":0.4169691259152335,"score_gpt":0.5039540648960044,"score_spread":0.08698493898077087,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385235004","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.028013112,0.0015792417,0.9637822,0.0024751688,0.00017376587,0.0002771411,0.0008810781,0.0003278195,0.0024904336],"genre_scores_gemma":[0.6603038,0.005495653,0.31073663,0.0011200368,0.0007528735,0.002480774,0.0028035569,0.00020829069,0.016098443],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99562377,0.0027817343,0.00019610873,0.00063726254,0.00045000532,0.0003111352],"domain_scores_gemma":[0.9816109,0.014540516,0.0014985873,0.000801666,0.0010970922,0.00045126711],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.019577812,0.001106047,0.0022498125,0.0024162324,0.0012084852,0.0019826537,0.0054780375,0.0037384452,0.004520097],"category_scores_gemma":[0.041159924,0.0010582816,0.0020900527,0.0033671474,0.0016151798,0.0025179246,0.00166473,0.0040237755,0.0011529059],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00021393326,0.00009217368,0.010924459,0.0002757033,0.00014480518,0.0004992646,0.00072702736,0.5559424,0.0006101076,0.37484124,0.0073724953,0.04835645],"study_design_scores_gemma":[0.00006009462,0.000072687246,0.0015661848,0.00009303768,0.000057266887,0.00012150716,0.0001086547,0.8589609,0.00015006994,0.13448574,0.0042714784,0.000052477983],"about_ca_topic_score_codex":0.03058073,"about_ca_topic_score_gemma":0.01839098,"teacher_disagreement_score":0.9804222,"about_ca_system_score_codex":0.002553809,"about_ca_system_score_gemma":0.003274277,"threshold_uncertainty_score":0.10353863},"labels":[],"label_agreement":null},{"id":"W4386010298","doi":"10.1002/sim.9879","title":"A threshold longitudinal Tobit quantile regression model for identification of treatment‐sensitive subgroups based on interval‐bounded longitudinal measurements and a continuous covariate","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Natural Science Foundation of Anhui Province; Natural Sciences and Engineering Research Council of Canada; National Natural Science Foundation of China","keywords":"Covariate; Statistics; Estimator; Mathematics; Tobit model; Econometrics; Quantile; Confidence interval; Random effects model; Identification (biology); Regression analysis; Quantile regression; Computer science; Medicine","score_opus":0.27177774041003855,"score_gpt":0.443233959218317,"score_spread":0.17145621880827844,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386010298","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014610247,0.00036183783,0.98314744,0.00052094064,0.000052256702,0.00012277473,0.00035697644,0.00019067405,0.00063678325],"genre_scores_gemma":[0.65881693,0.0016491199,0.32325503,0.0007622185,0.00017931277,0.0020544895,0.002551699,0.00024603505,0.010485194],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9946214,0.003512007,0.00018765074,0.00084581104,0.0004110177,0.00042200246],"domain_scores_gemma":[0.9901995,0.007410418,0.0008107219,0.00050658116,0.0008378257,0.00023487095],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013510035,0.0011592705,0.0027949705,0.0011118656,0.00053196657,0.0020129613,0.0033020396,0.0019090389,0.005704279],"category_scores_gemma":[0.02664584,0.00082126417,0.0018013573,0.0023046131,0.0011478126,0.0022357255,0.0018189428,0.0035020104,0.0012272455],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011128274,0.00032103513,0.025053743,0.0005135548,0.00064404187,0.0007136945,0.00089851633,0.5716356,0.0017729955,0.27702317,0.006288273,0.114022575],"study_design_scores_gemma":[0.000055633856,0.00010339852,0.001460461,0.000041419815,0.00008168761,0.000085739186,0.00007274022,0.94882745,0.00020021762,0.047657523,0.0013865938,0.000027075843],"about_ca_topic_score_codex":0.0061329366,"about_ca_topic_score_gemma":0.00304634,"teacher_disagreement_score":0.013510035,"about_ca_system_score_codex":0.0011561771,"about_ca_system_score_gemma":0.0017342203,"threshold_uncertainty_score":0.07144874},"labels":[],"label_agreement":null},{"id":"W4386116424","doi":"10.1002/sim.9867","title":"A comparison of Bayesian information borrowing methods in basket trials and a novel proposal of modified exchangeability‐nonexchangeability method","year":2023,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Engineering and Physical Sciences Research Council; Medical Research Council Canada; Department of Health and Social Care; Medical Research Council; National Institute for Health and Care Research","keywords":"Bayesian probability; Type I and type II errors; Sample size determination; Computer science; Econometrics; Homogeneity (statistics); Statistics; Bayesian inference; Generalization; Contrast (vision); Bayesian hierarchical modeling; Machine learning; Artificial intelligence; Mathematics","score_opus":0.7989345094425112,"score_gpt":0.7116990724016434,"score_spread":0.08723543704086778,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386116424","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0035707005,0.0020778638,0.99135125,0.00080085226,0.00009764694,0.00033897755,0.00013039692,0.00018928452,0.001443058],"genre_scores_gemma":[0.2059893,0.0049614464,0.7779175,0.0016453604,0.0006437759,0.0024423015,0.0007880482,0.0004013379,0.0052108993],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.94958377,0.04279709,0.0014990374,0.0021748315,0.0034980369,0.0004472771],"domain_scores_gemma":[0.88467026,0.09751504,0.0045679626,0.008573601,0.0036726706,0.001000529],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0730627,0.00169118,0.0031917926,0.0019648504,0.0006233227,0.0024281286,0.0053532096,0.002725439,0.006778449],"category_scores_gemma":[0.15240982,0.0009763913,0.0029438636,0.0025241817,0.002265504,0.0062628267,0.003448963,0.004024475,0.0008140109],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018754603,0.0001818231,0.0030383489,0.0014982665,0.001474948,0.00028705463,0.00058151106,0.12184683,0.000965722,0.46986765,0.0074212267,0.3909613],"study_design_scores_gemma":[0.00069191284,0.0006763666,0.0015477071,0.00039051843,0.0005950902,0.00029668785,0.00005780271,0.6339445,0.0012560015,0.3500047,0.010409326,0.00012932158],"about_ca_topic_score_codex":0.0022091665,"about_ca_topic_score_gemma":0.0014053307,"teacher_disagreement_score":0.0730627,"about_ca_system_score_codex":0.0018337083,"about_ca_system_score_gemma":0.00413198,"threshold_uncertainty_score":0.38639712},"labels":[],"label_agreement":null},{"id":"W4386243887","doi":"10.1002/sim.9887","title":"Analysis of secondary failure time responses in studies with response‐dependent sampling schemes","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"University of Waterloo; National Natural Science Foundation of China","keywords":"Estimator; Sample size determination; Statistics; Computer science; Gaussian; Sampling (signal processing); Econometrics; Mathematics","score_opus":0.1401599664157765,"score_gpt":0.4655415873936388,"score_spread":0.32538162097786233,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386243887","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0242081,0.0007360681,0.9718534,0.00037209206,0.00011014356,0.0015365047,0.00031545485,0.00020615062,0.0006621224],"genre_scores_gemma":[0.5063848,0.00085839746,0.47816527,0.0008404721,0.0002497395,0.008704792,0.001354004,0.00018434845,0.0032581587],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8421685,0.13690236,0.0055381227,0.00864483,0.005744404,0.001001826],"domain_scores_gemma":[0.52647436,0.40382814,0.019521115,0.043188967,0.0057398397,0.001247526],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.25903496,0.0014728997,0.0027363081,0.0025365634,0.0011820957,0.0021906237,0.004168322,0.0037140239,0.005482817],"category_scores_gemma":[0.3892622,0.0010819667,0.004486429,0.001918882,0.0031890857,0.002916195,0.0041341097,0.0038434328,0.000677963],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0058175,0.0007091313,0.17475232,0.0037768902,0.006056441,0.0015535643,0.0039081606,0.15744601,0.0074448157,0.344389,0.0053165224,0.28882954],"study_design_scores_gemma":[0.0009556842,0.0024856906,0.040107362,0.0007699701,0.002045822,0.00078685646,0.00045825684,0.64283407,0.00644323,0.2883546,0.014551954,0.00020651787],"about_ca_topic_score_codex":0.001432587,"about_ca_topic_score_gemma":0.0009881896,"teacher_disagreement_score":0.740965,"about_ca_system_score_codex":0.0014286212,"about_ca_system_score_gemma":0.0019707838,"threshold_uncertainty_score":0.9137418},"labels":[],"label_agreement":null},{"id":"W4386878567","doi":"10.1002/sim.9899","title":"Addressing missing data in the estimation of time‐varying treatments in comparative effectiveness research","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada","keywords":"Missing data; Inverse probability weighting; Imputation (statistics); Weighting; Confounding; Statistics; Inverse probability; Covariate; Econometrics; Computer science; Mathematics; Medicine; Posterior probability; Propensity score matching; Bayesian probability","score_opus":0.6914773159287642,"score_gpt":0.6378653717663129,"score_spread":0.0536119441624513,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386878567","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0037572694,0.006480613,0.984278,0.002398824,0.00039040495,0.0009185566,0.00029296128,0.00017083455,0.0013124581],"genre_scores_gemma":[0.17945355,0.008794849,0.80268943,0.001966842,0.00069423416,0.004986374,0.00053538725,0.00015877628,0.0007205687],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.54254216,0.4296343,0.00823657,0.00747027,0.010988819,0.0011278228],"domain_scores_gemma":[0.4054809,0.5604142,0.016151128,0.013274547,0.0040448657,0.00063436205],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.30661985,0.0026276708,0.006989851,0.005929638,0.0015371676,0.004398294,0.006267223,0.005644788,0.006920725],"category_scores_gemma":[0.5724815,0.0022647122,0.0085889725,0.008425967,0.005238302,0.007647086,0.005390102,0.010463228,0.00079298747],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013523303,0.0004717813,0.011491155,0.008246976,0.008726136,0.00036028266,0.0018979327,0.1661641,0.0004387258,0.48253497,0.0054146764,0.312901],"study_design_scores_gemma":[0.0009218707,0.0010763893,0.0027620632,0.00397324,0.0022640105,0.0002541168,0.00029720218,0.23270631,0.0011783879,0.7417546,0.012587033,0.00022476465],"about_ca_topic_score_codex":0.00612346,"about_ca_topic_score_gemma":0.0036325452,"teacher_disagreement_score":0.6933801,"about_ca_system_score_codex":0.004543566,"about_ca_system_score_gemma":0.00933388,"threshold_uncertainty_score":0.8550612},"labels":[],"label_agreement":null},{"id":"W4387326446","doi":"10.1002/sim.9917","title":"Joint clustering multiple longitudinal features: A comparison of methods and software packages with practical guidance","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada; Queen's University","keywords":"Cluster analysis; Computer science; Data mining; Machine learning; Software; Bayesian probability; Popularity; Feature (linguistics); Frequentist inference; Artificial intelligence; Data science; Bayesian inference","score_opus":0.09040215515519481,"score_gpt":0.4441058922588974,"score_spread":0.3537037371037026,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4387326446","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015346628,0.011506733,0.94527686,0.003005976,0.0006166789,0.0010017182,0.0040191347,0.014534536,0.0046916134],"genre_scores_gemma":[0.035007518,0.005548379,0.9470047,0.00046479516,0.000168326,0.0021472364,0.0040722275,0.004655467,0.0009313734],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.96319336,0.023048187,0.0032772904,0.0025552986,0.0074295904,0.00049622403],"domain_scores_gemma":[0.804631,0.16244797,0.0055708583,0.010790315,0.015337338,0.0012226062],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.061803922,0.002125663,0.002060046,0.009317602,0.0013869149,0.0046322653,0.003633657,0.0022917534,0.009167964],"category_scores_gemma":[0.19334419,0.0017942667,0.003706441,0.011178516,0.0012726131,0.0040127453,0.004129842,0.0030958138,0.0049033836],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019725405,0.00038172657,0.013219465,0.0049424395,0.002092957,0.00017232004,0.0013760831,0.03037326,0.0017952085,0.025750207,0.07141262,0.8465111],"study_design_scores_gemma":[0.0019197444,0.0012310459,0.037518747,0.0057172035,0.0023267553,0.002080132,0.002380189,0.5326774,0.011475982,0.15792653,0.24356838,0.0011778183],"about_ca_topic_score_codex":0.009023149,"about_ca_topic_score_gemma":0.011935333,"teacher_disagreement_score":0.93819606,"about_ca_system_score_codex":0.0023087678,"about_ca_system_score_gemma":0.006557816,"threshold_uncertainty_score":0.3268543},"labels":[],"label_agreement":null},{"id":"W4387697644","doi":"10.1002/sim.9934","title":"Rank selection for non‐negative matrix factorization","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Gene expression and cancer classification","field":"Biochemistry, Genetics and Molecular Biology","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Non-negative matrix factorization; Matrix decomposition; Rank (graph theory); Computer science; Feature selection; Pattern recognition (psychology); Dimensionality reduction; Matrix (chemical analysis); Model selection; Computation; Selection (genetic algorithm); Dimension (graph theory); Algorithm; Mathematics; Artificial intelligence; Eigenvalues and eigenvectors","score_opus":0.018925110906542836,"score_gpt":0.354111484269837,"score_spread":0.33518637336329415,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4387697644","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0015110867,0.00021586627,0.9973074,0.00013287028,0.000041527863,0.000058253354,0.00006473345,0.00027729754,0.00039096252],"genre_scores_gemma":[0.08821356,0.00064008054,0.9065868,0.0003705177,0.0003201351,0.00073949393,0.00089684397,0.00029805826,0.0019345431],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9917681,0.0049926783,0.0003255795,0.0009232774,0.0017142465,0.0002759802],"domain_scores_gemma":[0.9840904,0.011227179,0.0009405729,0.0012870115,0.002192979,0.00026188302],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008919419,0.0021763025,0.002088206,0.0016004415,0.001272793,0.0016517988,0.0018484981,0.0019517351,0.0041766516],"category_scores_gemma":[0.028938612,0.00068748416,0.0018687476,0.0016556059,0.002035892,0.0018550149,0.0015859825,0.0028072405,0.0028256758],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00039502935,0.00025194988,0.0022429528,0.0008426203,0.00045864994,0.000475597,0.000316949,0.44201744,0.008456604,0.12428798,0.028653845,0.3916004],"study_design_scores_gemma":[0.000050047387,0.00007276798,0.0003584164,0.00003235289,0.00001683664,0.000096448995,0.000027182903,0.9530735,0.0017823948,0.04069885,0.003762512,0.000028663211],"about_ca_topic_score_codex":0.0029731463,"about_ca_topic_score_gemma":0.0034096842,"teacher_disagreement_score":0.008919419,"about_ca_system_score_codex":0.0011354522,"about_ca_system_score_gemma":0.0024753013,"threshold_uncertainty_score":0.047170937},"labels":[],"label_agreement":null},{"id":"W4388398918","doi":"10.1002/sim.9940","title":"Principal stratification for quantile causal effects under partial compliance","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs; Fonds de Recherche du Québec - Santé; Institut de Valorisation des Données; Compute Canada","keywords":"Quantile; Econometrics; Causal inference; Copula (linguistics); Covariate; Missing data; Statistics; Bivariate analysis; Identifiability; Inference; Mathematics; Computer science; Artificial intelligence","score_opus":0.2595680897329376,"score_gpt":0.5012533153693227,"score_spread":0.24168522563638511,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4388398918","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009640127,0.00018973305,0.98804945,0.0003324558,0.000033827815,0.00022812563,0.00017648688,0.00021732949,0.0011326037],"genre_scores_gemma":[0.46346888,0.00082171254,0.5280201,0.00067392277,0.00027172378,0.0027750486,0.0007761017,0.00020861541,0.0029838553],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96133935,0.029776398,0.0012456538,0.0037848682,0.0026951488,0.0011584861],"domain_scores_gemma":[0.7968297,0.16807334,0.008661927,0.021090407,0.0044849073,0.00085966947],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07234291,0.0012331329,0.003450699,0.0029852053,0.0016354355,0.0025359702,0.002903906,0.0026727049,0.010710736],"category_scores_gemma":[0.20440735,0.0012462218,0.0047670417,0.0031807416,0.0060237213,0.005185062,0.0054404954,0.005557362,0.00081659685],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002494219,0.00015975411,0.01660361,0.0003613902,0.00053192955,0.00044021697,0.0012970372,0.076710105,0.0005895548,0.83682245,0.0018499879,0.06438457],"study_design_scores_gemma":[0.000076314835,0.0001959315,0.004384249,0.0001336263,0.00015959151,0.00015010503,0.00015756337,0.3407289,0.0006220008,0.6503136,0.0030350375,0.00004314832],"about_ca_topic_score_codex":0.0051891664,"about_ca_topic_score_gemma":0.0025855924,"teacher_disagreement_score":0.07234291,"about_ca_system_score_codex":0.0017766157,"about_ca_system_score_gemma":0.003710673,"threshold_uncertainty_score":0.38259047},"labels":[],"label_agreement":null},{"id":"W4388458822","doi":"10.1002/sim.9954","title":"A flexible model based on piecewise linear approximation for the analysis of left truncated right censored data with covariates, and applications to Worcester Heart Attack Study data and Channing House data","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Distribution Estimation and Applications","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"Science and Engineering Research Board; Natural Sciences and Engineering Research Council of Canada","keywords":"Covariate; Piecewise; Computer science; Parametric model; Parametric statistics; Robustness (evolution); Proportional hazards model; Inference; Hazard; Applied mathematics; Statistics; Mathematics; Algorithm; Data mining; Artificial intelligence","score_opus":0.2283174352599303,"score_gpt":0.46335943499915616,"score_spread":0.23504199973922585,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4388458822","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00469921,0.0005067206,0.9931878,0.00045827753,0.00004988753,0.000053472275,0.00031022864,0.00020249763,0.00053202914],"genre_scores_gemma":[0.44307455,0.004384859,0.5330532,0.00069878355,0.00039502705,0.0010368187,0.0032883135,0.00029087727,0.013777494],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99749446,0.0012720682,0.00012301373,0.0005227593,0.00039202592,0.00019562275],"domain_scores_gemma":[0.9932001,0.00472594,0.0007560108,0.0006189972,0.0005341764,0.00016478336],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0064527877,0.00091685384,0.0010565775,0.0014445666,0.0005186242,0.0013126851,0.0031255407,0.0015911374,0.0042253793],"category_scores_gemma":[0.017393544,0.0006355028,0.0018253756,0.0034319158,0.00096000935,0.0020812033,0.0014738069,0.0036211123,0.0009431361],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001565725,0.000087725246,0.009296102,0.00042897614,0.00020286968,0.0008671595,0.0004653229,0.6382301,0.0026019774,0.21883294,0.004825952,0.12400435],"study_design_scores_gemma":[0.000019320032,0.00011535368,0.0019441827,0.00004335593,0.00006559461,0.00035593603,0.000060589067,0.9298813,0.00041849338,0.061574224,0.0054730824,0.00004863795],"about_ca_topic_score_codex":0.008904921,"about_ca_topic_score_gemma":0.0063477997,"teacher_disagreement_score":0.008904921,"about_ca_system_score_codex":0.0012558104,"about_ca_system_score_gemma":0.0019455124,"threshold_uncertainty_score":0.034125984},"labels":[],"label_agreement":null},{"id":"W4388851130","doi":"10.1002/sim.9963","title":"Two‐stage targeted maximum likelihood estimation for mixed aggregate and individual participant data analysis with an application to multidrug resistant tuberculosis","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre; Université de Montréal; McGill University; McMaster University","funders":"Canadian Institutes of Health Research","keywords":"Aggregate data; Weighting; Computer science; Missing data; Estimation; Tuberculosis; Aggregate (composite); Inverse probability weighting; Meta-analysis; Statistics; Population; Contrast (vision); Medicine; Econometrics; Machine learning; Artificial intelligence; Mathematics; Internal medicine; Estimator; Pathology","score_opus":0.1470997536980372,"score_gpt":0.4430485501490196,"score_spread":0.2959487964509824,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4388851130","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011912417,0.00051694154,0.99757725,0.00020820339,0.000039985305,0.00012619603,0.00007926829,0.0001220511,0.00013879867],"genre_scores_gemma":[0.07679463,0.0008154121,0.91878873,0.0005039438,0.00017047022,0.0014938698,0.00040562364,0.00014555747,0.00088185404],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.951552,0.042844158,0.0012302158,0.002413998,0.0016686969,0.00029096747],"domain_scores_gemma":[0.8623272,0.12713863,0.0030332657,0.0051007112,0.0019797108,0.00042052826],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05683197,0.0016659312,0.0034196426,0.0032909862,0.0007611655,0.0023814621,0.003346439,0.0022215154,0.0034721042],"category_scores_gemma":[0.13934258,0.0014592494,0.006172033,0.0034454144,0.0014267021,0.0026764597,0.003068032,0.004132041,0.0006685357],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010340235,0.00029164008,0.015177412,0.0039061708,0.010875997,0.0011000338,0.0010333782,0.3061823,0.002680245,0.26009974,0.006060535,0.39155853],"study_design_scores_gemma":[0.00036434995,0.000388761,0.0021437178,0.0003487865,0.0010615463,0.00034782404,0.000095507145,0.71116906,0.001093002,0.27411163,0.008767915,0.00010789338],"about_ca_topic_score_codex":0.0034212256,"about_ca_topic_score_gemma":0.003920983,"teacher_disagreement_score":0.05683197,"about_ca_system_score_codex":0.001249513,"about_ca_system_score_gemma":0.0031975156,"threshold_uncertainty_score":0.30055976},"labels":[],"label_agreement":null},{"id":"W4389676925","doi":"10.1002/sim.9973","title":"Accommodating misclassification effects on optimizing dynamic treatment regimes with Q‐learning","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Actua; Western University","funders":"Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs","keywords":"Covariate; Computer science; Econometrics; Estimation; Statistics; Machine learning; Artificial intelligence; Mathematics","score_opus":0.09089600466695962,"score_gpt":0.43294174638760535,"score_spread":0.3420457417206457,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389676925","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008921116,0.0007098665,0.98807484,0.0012496897,0.00006474899,0.00012537118,0.00004503238,0.00012405412,0.0006853134],"genre_scores_gemma":[0.5589623,0.0019411105,0.43292302,0.0019773652,0.0003961952,0.0008646546,0.0002862398,0.00014473537,0.0025043841],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98252213,0.0135876825,0.00063784746,0.0016732622,0.0010318451,0.0005471485],"domain_scores_gemma":[0.8560436,0.12926506,0.005153809,0.0058156187,0.0028293275,0.0008926517],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04004045,0.0013740015,0.0033313485,0.0011858784,0.0010196558,0.0022827045,0.0029980084,0.0033445803,0.002799441],"category_scores_gemma":[0.13024619,0.000900294,0.0012710795,0.0015056588,0.0038902874,0.0035329387,0.0025377786,0.00478942,0.00044993154],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00043843818,0.00028409154,0.008398351,0.00061677396,0.0003465769,0.00018318284,0.0004305218,0.60601234,0.00076972006,0.19671084,0.002946283,0.18286291],"study_design_scores_gemma":[0.00010808996,0.000107884734,0.00053554436,0.00009300137,0.000053021733,0.000053275056,0.00003059131,0.84083086,0.00054113264,0.15647419,0.0011458402,0.000026588177],"about_ca_topic_score_codex":0.0038430328,"about_ca_topic_score_gemma":0.0022247306,"teacher_disagreement_score":0.04004045,"about_ca_system_score_codex":0.002025051,"about_ca_system_score_gemma":0.004497904,"threshold_uncertainty_score":0.21175665},"labels":[],"label_agreement":null},{"id":"W4390031813","doi":"10.1002/sim.9984","title":"Modeling multiple correlated end‐organ disease trajectories: A tutorial for multistate and joint models with applications in diabetes complications","year":2023,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Diabetes Management and Research","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sinai Health System; Public Health Ontario; University of Toronto; Lunenfeld-Tanenbaum Research Institute; University Health Network","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Statistical Sciences Institute; Canadian Institutes of Health Research","keywords":"Biostatistics; Computer science; Discretization; Outcome (game theory); Contrast (vision); Multivariate statistics; Process (computing); Machine learning; Intensive care medicine; Data mining; Artificial intelligence; Medicine; Epidemiology; Mathematics; Internal medicine","score_opus":0.05080697279266112,"score_gpt":0.33082962055929593,"score_spread":0.2800226477666348,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4390031813","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0012654087,0.0904212,0.89014685,0.0046722423,0.0025712985,0.0001723675,0.0011890533,0.0010223725,0.008539209],"genre_scores_gemma":[0.029456792,0.20498805,0.7164937,0.0055859145,0.010508168,0.0014160452,0.0028330619,0.0012412901,0.027477011],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99927825,0.0003581695,0.0000676886,0.000118244025,0.00014630068,0.00003138717],"domain_scores_gemma":[0.9962825,0.0031149848,0.00016201328,0.00011192,0.0002419167,0.000086675136],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0023365673,0.0030170656,0.0012039521,0.0017976656,0.0004094958,0.0015265871,0.0014387668,0.0024447981,0.010928712],"category_scores_gemma":[0.0073078303,0.0009538116,0.0021851864,0.0022177505,0.0007761872,0.0028829582,0.0018619966,0.0050718444,0.0057493523],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007856404,0.00026345273,0.0019235171,0.0035461942,0.00029053632,0.0011884376,0.0007448689,0.116362944,0.003008646,0.3300053,0.24211514,0.30047235],"study_design_scores_gemma":[0.000029772918,0.00015242709,0.001234085,0.0015187808,0.00009019808,0.0010786647,0.00015341278,0.16610219,0.0006057706,0.3017269,0.527159,0.00014880979],"about_ca_topic_score_codex":0.0038333521,"about_ca_topic_score_gemma":0.0038138363,"teacher_disagreement_score":0.010928712,"about_ca_system_score_codex":0.0008365326,"about_ca_system_score_gemma":0.00096188753,"threshold_uncertainty_score":0.036560237},"labels":[],"label_agreement":null},{"id":"W4390883403","doi":"10.1002/sim.10009","title":"A Bayesian framework for modeling COVID‐19 case numbers through longitudinal monitoring of SARS‐CoV‐2 RNA in wastewater","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"SARS-CoV-2 detection and testing","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Alberta Health Services; University of Alberta; University of Calgary","funders":"National Institute of General Medical Sciences; Canadian Institutes of Health Research; Illinois State University; National Institutes of Health; Alberta Health; Natural Sciences and Engineering Research Council of Canada; Alberta Health Services; University of Minnesota","keywords":"Coronavirus disease 2019 (COVID-19); Wastewater; Bayesian probability; Computer science; Missing data; Pandemic; Data mining; Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); Process (computing); Statistics; Environmental science; Artificial intelligence; Machine learning; Medicine; Mathematics; Environmental engineering; Internal medicine","score_opus":0.12995340238863706,"score_gpt":0.4334511856835037,"score_spread":0.30349778329486665,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4390883403","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.041164465,0.00029254003,0.9565196,0.00050832133,0.00002729868,0.00006898926,0.0003293972,0.0001600882,0.0009293766],"genre_scores_gemma":[0.82510024,0.0009797825,0.16645712,0.00024680627,0.00018211339,0.0005011513,0.001096777,0.00007358636,0.0053623253],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99866533,0.00072287477,0.000063734886,0.00025702408,0.0001614409,0.00012962865],"domain_scores_gemma":[0.9951356,0.0035404887,0.00055708684,0.00014362702,0.00048384664,0.00013934425],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0064785513,0.0010024508,0.0011462163,0.0016286451,0.00049575703,0.001239962,0.0018640424,0.0016577234,0.0015241839],"category_scores_gemma":[0.011352227,0.0009400375,0.0013845969,0.0010040159,0.0011619333,0.0012705353,0.0010204726,0.0015862181,0.00028620064],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000043166314,0.000042986958,0.0033948794,0.000025729367,0.00005600229,0.000064807115,0.000065714026,0.9610079,0.00035575381,0.022448026,0.00039760018,0.012097442],"study_design_scores_gemma":[0.0000041823264,0.00001302774,0.0003037431,0.000004951983,0.000007696915,0.000007264194,0.00000652898,0.9947712,0.00004053312,0.0047010565,0.0001326768,0.00000716177],"about_ca_topic_score_codex":0.026339984,"about_ca_topic_score_gemma":0.023135014,"teacher_disagreement_score":0.026339984,"about_ca_system_score_codex":0.0013254471,"about_ca_system_score_gemma":0.0020451585,"threshold_uncertainty_score":0.05237335},"labels":[],"label_agreement":null},{"id":"W4391138204","doi":"10.1002/sim.10011","title":"Multiple imputation strategies for missing event times in a multi‐state model analysis","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; University of Bristol; Medical Research Council Canada; Wellcome Trust; NHS Blood and Transplant","keywords":"Imputation (statistics); Computer science; Missing data; Data mining; Statistics; Machine learning; Mathematics","score_opus":0.09443405681533493,"score_gpt":0.4594438165445142,"score_spread":0.36500975972917926,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391138204","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.002100541,0.00032394417,0.9965313,0.0003725482,0.000050773524,0.00007276988,0.00016583205,0.00017769697,0.00020468756],"genre_scores_gemma":[0.16035762,0.0009862211,0.83279073,0.0007014548,0.00024747895,0.0013844162,0.001344197,0.00030640746,0.0018814482],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.97458863,0.019923609,0.0012423916,0.0025068775,0.0012677477,0.0004707261],"domain_scores_gemma":[0.8963637,0.08968614,0.0046282173,0.0059055905,0.0026606882,0.00075568847],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05412706,0.0018216694,0.004566985,0.0030045973,0.0017174517,0.0030102232,0.0064452374,0.003270973,0.0057962034],"category_scores_gemma":[0.12266248,0.001965941,0.0046268427,0.0040264996,0.0017922915,0.004494563,0.0038954497,0.006341776,0.0010964564],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00049153826,0.00022531414,0.012698936,0.001000164,0.002357799,0.00086977927,0.0011003673,0.4910536,0.000630282,0.32908818,0.008690097,0.1517939],"study_design_scores_gemma":[0.00007752144,0.00008468102,0.00065330206,0.00015281534,0.00016352936,0.00012854431,0.00007588702,0.71561944,0.0003510245,0.27948007,0.0031508247,0.00006228387],"about_ca_topic_score_codex":0.005405979,"about_ca_topic_score_gemma":0.0061284653,"teacher_disagreement_score":0.05412706,"about_ca_system_score_codex":0.0015555745,"about_ca_system_score_gemma":0.0039484906,"threshold_uncertainty_score":0.28625464},"labels":[],"label_agreement":null},{"id":"W4391652457","doi":"10.1002/sim.10034","title":"Correction to “Using principal stratification in analysis of clinical trials”","year":2024,"lang":"en","type":"erratum","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Bayer (Canada)","funders":"","keywords":"Statistics; Principal (computer security); Stratification (seeds); Computer science; Econometrics; Mathematics; Biology","score_opus":0.7621167309528273,"score_gpt":0.699804452428747,"score_spread":0.06231227852408028,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391652457","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00009766522,0.0012941959,0.0076134074,0.10242142,0.87808025,0.000084097745,0.0043666055,0.0018792307,0.004163117],"genre_scores_gemma":[0.009463655,0.007255721,0.052485973,0.2870638,0.41180328,0.0015570583,0.009864475,0.012330641,0.2081754],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9492471,0.016418558,0.01043869,0.00378153,0.018406115,0.0017079839],"domain_scores_gemma":[0.7312327,0.1204442,0.012528844,0.018414954,0.11304387,0.004335464],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.034449186,0.0033332347,0.0036429241,0.0076038004,0.0039918115,0.0071992488,0.0054876218,0.008506226,0.08356698],"category_scores_gemma":[0.37218946,0.0024086186,0.0037084436,0.0070375567,0.0057501947,0.0050171916,0.004368243,0.023952,0.072563566],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000019532528,0.0000035698479,0.000020184774,0.0000994633,0.0000109248285,0.00004349695,0.000023891484,0.000032909615,0.000017036606,0.0013090835,0.99523383,0.003186119],"study_design_scores_gemma":[0.00007907985,0.000019206585,0.00033141804,0.0005328303,0.00004610264,0.0002769318,0.0000405858,0.00034409744,0.00022118358,0.0068205628,0.99123514,0.000052907475],"about_ca_topic_score_codex":0.014316196,"about_ca_topic_score_gemma":0.01580959,"teacher_disagreement_score":0.08356698,"about_ca_system_score_codex":0.005756316,"about_ca_system_score_gemma":0.015107275,"threshold_uncertainty_score":0.27955937},"labels":[],"label_agreement":null},{"id":"W4391841493","doi":"10.1002/sim.10012","title":"Statistical plasmode simulations–Potentials, challenges and recommendations","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Computer science; Variety (cybernetics); Key (lock); Statistical model; Set (abstract data type); Parametric statistics; Data science; Data set; Data mining; Machine learning; Theoretical computer science; Artificial intelligence; Statistics; Mathematics","score_opus":0.11468469702013184,"score_gpt":0.44976894827582375,"score_spread":0.3350842512556919,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391841493","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.006107831,0.08544374,0.60965776,0.26918727,0.0071158432,0.00031475312,0.0016781181,0.0058345846,0.014660095],"genre_scores_gemma":[0.0933444,0.15972848,0.70414585,0.021325767,0.007152439,0.0012891002,0.0028264339,0.0019755499,0.008211926],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9904355,0.0066406815,0.00049106777,0.0006853323,0.0015201512,0.00022739064],"domain_scores_gemma":[0.88196886,0.09098778,0.0016250687,0.00743971,0.015057154,0.0029214479],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.027157973,0.001336861,0.0016012074,0.002471907,0.000925959,0.004600569,0.0055283,0.0044841063,0.011729208],"category_scores_gemma":[0.14496382,0.00076129136,0.0013419585,0.0022176302,0.0028153437,0.011050316,0.0036978335,0.0071792705,0.0059877755],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019502924,0.0002702937,0.003903078,0.0025601976,0.00018466414,0.00041193137,0.0003785297,0.07856375,0.0005856191,0.39915374,0.1487325,0.36506072],"study_design_scores_gemma":[0.000091124966,0.00006909341,0.0005144507,0.0027889372,0.00004694699,0.00022225085,0.00036532918,0.11835811,0.000782723,0.70743734,0.16916943,0.00015434543],"about_ca_topic_score_codex":0.0056408574,"about_ca_topic_score_gemma":0.0053022015,"teacher_disagreement_score":0.027157973,"about_ca_system_score_codex":0.001633886,"about_ca_system_score_gemma":0.0048953374,"threshold_uncertainty_score":0.14362681},"labels":[],"label_agreement":null},{"id":"W4396723746","doi":"10.1002/sim.10104","title":"Adaptive designs in public health: Vaccine and cluster randomized trials go Bayesian","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"National Institute of Environmental Health Sciences; National Institute on Aging","keywords":"Bayes' theorem; Bayesian probability; Computer science; Randomized controlled trial; Clinical trial; Flexibility (engineering); Psychological intervention; Non-communicable disease; Cluster (spacecraft); Public health; Matching (statistics); Cluster randomised controlled trial; Medicine; Econometrics; Risk analysis (engineering); Statistics; Artificial intelligence; Economics; Mathematics; Nursing","score_opus":0.6838323276886842,"score_gpt":0.60154715796384,"score_spread":0.08228516972484423,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4396723746","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0014847023,0.0024501516,0.98695415,0.0051295236,0.00033995253,0.0005125186,0.000121617806,0.00023600274,0.0027713317],"genre_scores_gemma":[0.07990505,0.0044479845,0.9026876,0.0033917022,0.0009024241,0.004704906,0.00016773604,0.00018537106,0.0036072063],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9057463,0.085828744,0.0015715944,0.0027341503,0.0036491808,0.00047007352],"domain_scores_gemma":[0.846906,0.13550548,0.005663882,0.0078100916,0.0033431663,0.0007713854],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09010871,0.0016240483,0.0028280213,0.0026989446,0.00086845004,0.004218433,0.0030518756,0.005843295,0.008362415],"category_scores_gemma":[0.21975735,0.0015512457,0.0022704848,0.0029504546,0.0064828033,0.0058671017,0.0034755825,0.0076435665,0.0015637939],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002456374,0.000053655753,0.0006071942,0.000638977,0.0002589939,0.000073541414,0.0002825706,0.036379654,0.00016948846,0.895563,0.005100737,0.06062641],"study_design_scores_gemma":[0.0003004667,0.000112306814,0.00021772637,0.00039658614,0.000080992926,0.000055354976,0.000031889955,0.0798253,0.00017408776,0.90806633,0.010693204,0.00004573548],"about_ca_topic_score_codex":0.0028194124,"about_ca_topic_score_gemma":0.0021261333,"teacher_disagreement_score":0.9098913,"about_ca_system_score_codex":0.0026708788,"about_ca_system_score_gemma":0.00554833,"threshold_uncertainty_score":0.4765461},"labels":[],"label_agreement":null},{"id":"W4396799016","doi":"10.1002/sim.10089","title":"Novel non‐linear models for clinical trial analysis with longitudinal data: A tutorial using <scp>SAS</scp> for both frequentist and Bayesian methods","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"College of Veterinarians of British Columbia","funders":"National Institute on Aging; National Institutes of Health","keywords":"Frequentist inference; Categorical variable; Bayesian probability; Repeated measures design; Random effects model; Computer science; Inference; Mixed model; Statistics; Bayesian inference; Linear model; Generalized linear mixed model; Mathematics; Econometrics; Artificial intelligence; Medicine","score_opus":0.7321217157620442,"score_gpt":0.6629236950961048,"score_spread":0.06919802066593939,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4396799016","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000061947954,0.0009403781,0.996067,0.00036069078,0.00014029777,0.00005986683,0.00035731588,0.0014190039,0.0005935077],"genre_scores_gemma":[0.002405829,0.0029474471,0.9890265,0.0005714198,0.0004721698,0.001051813,0.0006520562,0.0010801326,0.0017926681],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99090695,0.0066335956,0.0006419211,0.00050985254,0.0011971606,0.00011056504],"domain_scores_gemma":[0.9638568,0.031371,0.0012923209,0.0015837048,0.0016029003,0.00029325197],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016970087,0.0025682282,0.0018853912,0.002061881,0.00038018308,0.002608543,0.0027530969,0.002299339,0.03068391],"category_scores_gemma":[0.041464765,0.0018541445,0.0034357097,0.0026463594,0.0010013226,0.0027131399,0.0020222727,0.008005259,0.019694394],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001425221,0.00019717508,0.0010746551,0.0034318424,0.00085861166,0.0005411972,0.0005043858,0.0692051,0.003728446,0.32874689,0.17937006,0.41219914],"study_design_scores_gemma":[0.00020109404,0.00015510397,0.00075539324,0.00092931,0.0001614188,0.0009213515,0.00007357261,0.37225565,0.0020572206,0.32557347,0.29672158,0.00019483494],"about_ca_topic_score_codex":0.002513243,"about_ca_topic_score_gemma":0.0038610261,"teacher_disagreement_score":0.03068391,"about_ca_system_score_codex":0.0011288739,"about_ca_system_score_gemma":0.0028262632,"threshold_uncertainty_score":0.1026479},"labels":[],"label_agreement":null},{"id":"W4399135167","doi":"10.1002/sim.10116","title":"Structured learning in time‐dependent Cox models","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Neural Networks and Applications","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; McGill University; Mila - Quebec Artificial Intelligence Institute","funders":"Fonds de recherche du Québec – Nature et technologies; Fonds de Recherche du Québec - Santé; Canadian Institutes of Health Research; Ministère de la Santé et des Services sociaux; Réseau Québécois de Recherche sur les Médicaments; Natural Sciences and Engineering Research Council of Canada; Heart and Stroke Foundation of Canada","keywords":"Computer science; Proportional hazards model; Econometrics; Statistics; Mathematics","score_opus":0.017194394397149235,"score_gpt":0.3004267088436853,"score_spread":0.283232314446536,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4399135167","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009459822,0.00021319481,0.9891362,0.00035086364,0.000021606354,0.000049250244,0.00016691552,0.00021776615,0.000384322],"genre_scores_gemma":[0.43344057,0.0009942556,0.55595833,0.00062671764,0.00029249967,0.00095585635,0.0020513507,0.00023977472,0.005440681],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9973628,0.0019294693,0.000084242405,0.00030484446,0.00020970903,0.00010882747],"domain_scores_gemma":[0.97706485,0.020334594,0.00086596777,0.0007438523,0.0007045555,0.00028617628],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008072981,0.0007647083,0.0011683718,0.0008754489,0.0005240561,0.00096076104,0.0017862031,0.0012450437,0.002881851],"category_scores_gemma":[0.026357288,0.00074843975,0.0011859908,0.00089170266,0.0009552495,0.0012836398,0.0015928971,0.0024792228,0.0005667161],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017342456,0.000074639574,0.004217313,0.00015794823,0.00011584163,0.00018905623,0.00015784056,0.84870625,0.00043821067,0.070354775,0.0035212226,0.07189344],"study_design_scores_gemma":[0.000021085767,0.000014410888,0.00017066911,0.000010472442,0.000007992988,0.000012956799,0.000007359652,0.9611133,0.00009983669,0.038121033,0.00041513157,0.0000057991233],"about_ca_topic_score_codex":0.005233412,"about_ca_topic_score_gemma":0.007947012,"teacher_disagreement_score":0.008072981,"about_ca_system_score_codex":0.0009985337,"about_ca_system_score_gemma":0.001698225,"threshold_uncertainty_score":0.04269457},"labels":[],"label_agreement":null},{"id":"W4399262027","doi":"10.1002/sim.10126","title":"The effect of number of clusters and magnitude of within‐cluster homogeneity in outcomes on the performance of four variance estimators for a marginal multivariable Cox regression model fit to clustered data in the context of observational research","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sunnybrook Hospital; University of Toronto","funders":"Ministry of Long-Term Care; Canadian Institutes of Health Research; Ministry of Health, Ontario","keywords":"Estimator; Statistics; Homogeneity (statistics); Mathematics; Covariate; Confidence interval; Cluster (spacecraft); Econometrics; Proportional hazards model; Regression; Standard error; Regression analysis; Hazard ratio; Computer science","score_opus":0.25730499408334184,"score_gpt":0.5027874771247031,"score_spread":0.24548248304136122,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4399262027","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.61097366,0.0036677802,0.3787345,0.0015154917,0.00023547692,0.0009817071,0.00042556165,0.0007657769,0.0027000192],"genre_scores_gemma":[0.9283232,0.00029348518,0.06985425,0.00019967566,0.000042399828,0.00050293043,0.00032699163,0.00015891469,0.0002981154],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.79176915,0.17567582,0.008968146,0.013442382,0.0078598885,0.0022846742],"domain_scores_gemma":[0.16419528,0.77987164,0.016752917,0.027335139,0.010239424,0.0016055264],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27790016,0.0017992097,0.0025045422,0.0030128474,0.0018521775,0.0035570308,0.0026314163,0.0038388239,0.0012710086],"category_scores_gemma":[0.60987854,0.0013011285,0.0043807495,0.0026159238,0.0049653444,0.0046563717,0.004469447,0.0033850232,0.0002847049],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.013057812,0.00066466565,0.5379377,0.001451137,0.013514752,0.00069915265,0.006144278,0.23056097,0.0030286016,0.022808857,0.0026968657,0.1674352],"study_design_scores_gemma":[0.002282951,0.0059200223,0.26943377,0.0012022151,0.006588152,0.00093469897,0.0028141309,0.6447001,0.011715782,0.04963664,0.0038797623,0.00089185685],"about_ca_topic_score_codex":0.0074714557,"about_ca_topic_score_gemma":0.0041558174,"teacher_disagreement_score":0.72209984,"about_ca_system_score_codex":0.0019428743,"about_ca_system_score_gemma":0.004193131,"threshold_uncertainty_score":0.89047766},"labels":[],"label_agreement":null},{"id":"W4399593952","doi":"10.1002/sim.10138","title":"Non‐parametric inference on calibration of predicted risks","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Canadian Institutes of Health Research","keywords":"Inference; Calibration; Parametric statistics; Computer science; Statistics; Econometrics; Parametric model; Mathematics; Artificial intelligence","score_opus":0.1888057989843291,"score_gpt":0.5166805173376785,"score_spread":0.3278747183533494,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4399593952","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04777624,0.00016446067,0.9501442,0.00032482078,0.000038005703,0.00009463558,0.0001285579,0.00025871964,0.0010704764],"genre_scores_gemma":[0.8707578,0.00023833779,0.12680283,0.00031596798,0.00010156566,0.0005015057,0.00049840845,0.00011651714,0.00066714076],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.93920594,0.04356159,0.002478638,0.008299084,0.005311753,0.0011430483],"domain_scores_gemma":[0.60065204,0.3367814,0.019641649,0.0368481,0.0047805216,0.0012962994],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09298708,0.00132846,0.0027252792,0.0029456124,0.0010293036,0.0034831576,0.0039722337,0.0032510755,0.003260972],"category_scores_gemma":[0.38580313,0.001143954,0.0030161047,0.0021193316,0.007921073,0.0070527634,0.006253029,0.005856731,0.00061812135],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010694753,0.00031224114,0.051500697,0.00052862655,0.00156534,0.00057665625,0.0018496315,0.2777751,0.0035661354,0.503704,0.0019797557,0.15557234],"study_design_scores_gemma":[0.000070561895,0.0003759468,0.01336824,0.00016232757,0.00013937414,0.00023514654,0.00021432656,0.564019,0.003368704,0.416634,0.0013143136,0.00009809974],"about_ca_topic_score_codex":0.0009482535,"about_ca_topic_score_gemma":0.0004873243,"teacher_disagreement_score":0.09298708,"about_ca_system_score_codex":0.0011890795,"about_ca_system_score_gemma":0.001406586,"threshold_uncertainty_score":0.4917686},"labels":[],"label_agreement":null},{"id":"W4399721808","doi":"10.1002/sim.10151","title":"A sparse factor model for clustering high‐dimensional longitudinal data","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Cluster analysis; Computer science; Categorical variable; Dirichlet process; Curse of dimensionality; Gibbs sampling; Clustering high-dimensional data; Data mining; Bayesian probability; Artificial intelligence; Machine learning","score_opus":0.12987540817975268,"score_gpt":0.39157382235535,"score_spread":0.2616984141755973,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4399721808","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0036025464,0.0001838627,0.99557245,0.00018027639,0.000021522219,0.000031045594,0.000111526555,0.000096940596,0.00019970027],"genre_scores_gemma":[0.33662656,0.001789065,0.6521302,0.0004127938,0.00032919677,0.0008566469,0.0021152834,0.00020473044,0.0055355453],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9962698,0.0020837837,0.00016246115,0.0007280715,0.00054412143,0.00021164672],"domain_scores_gemma":[0.99150956,0.0059311744,0.00072938245,0.0007829849,0.00085598783,0.000190882],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008476835,0.0010650845,0.0020782335,0.0022077458,0.0009447585,0.0016982139,0.0032955927,0.0024709336,0.0023150954],"category_scores_gemma":[0.021539273,0.0010137927,0.0019571641,0.0030210917,0.002476921,0.003205418,0.0018054062,0.0028580122,0.001058326],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019578346,0.00007386976,0.0031860988,0.00019920751,0.00017490868,0.00013393344,0.00042619993,0.6441863,0.0018382011,0.28780064,0.0033399214,0.05844494],"study_design_scores_gemma":[0.000016024462,0.000023065611,0.00033914426,0.000016030515,0.000018413148,0.000045456818,0.000021289137,0.9291872,0.00017391048,0.06896033,0.0011753655,0.000023702369],"about_ca_topic_score_codex":0.009736656,"about_ca_topic_score_gemma":0.008493644,"teacher_disagreement_score":0.009736656,"about_ca_system_score_codex":0.0015422094,"about_ca_system_score_gemma":0.0016287861,"threshold_uncertainty_score":0.044830322},"labels":[],"label_agreement":null},{"id":"W4399817401","doi":"10.1002/sim.10152","title":"deepAFT: A nonlinear accelerated failure time model with artificial neural network","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Kingston General Hospital; Queen's University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Censoring (clinical trials); Covariate; Proportional hazards model; Accelerated failure time model; Computer science; Artificial neural network; Inverse probability weighting; Imputation (statistics); Weighting; Artificial intelligence; Regression; Nonparametric statistics; Survival analysis; Statistics; Machine learning; Econometrics; Mathematics; Missing data; Estimator","score_opus":0.09290964956243886,"score_gpt":0.3986075386957393,"score_spread":0.3056978891333004,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4399817401","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.038212363,0.00065637025,0.95659333,0.0008425469,0.0001256006,0.00004978862,0.0006354039,0.00048983196,0.0023947933],"genre_scores_gemma":[0.87389326,0.0011510339,0.107336245,0.000508743,0.00017108576,0.00039230063,0.001148177,0.00009667701,0.015302581],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99962246,0.00013100597,0.000018982519,0.00008885959,0.000074949974,0.00006373971],"domain_scores_gemma":[0.9991234,0.00048548382,0.00013052578,0.000055609948,0.00014751629,0.000057371864],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0014718398,0.0007079732,0.0008344257,0.00045996904,0.0002643709,0.00074753945,0.0018151115,0.0011507499,0.0023493827],"category_scores_gemma":[0.003359079,0.00038384995,0.0007182201,0.000538308,0.0007049349,0.001035653,0.0011028675,0.0018652297,0.00034125388],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000090187576,0.00004138218,0.0019669116,0.000058066773,0.000049541308,0.0001014781,0.000040523377,0.93247485,0.0005061562,0.029839246,0.0020291652,0.032802526],"study_design_scores_gemma":[0.0000045889847,0.000010826844,0.00012231417,0.000003504057,0.0000050801777,0.000010539652,0.0000016172378,0.99405473,0.00006234342,0.005416516,0.0003045978,0.0000033522952],"about_ca_topic_score_codex":0.011349867,"about_ca_topic_score_gemma":0.010054409,"teacher_disagreement_score":0.011349867,"about_ca_system_score_codex":0.00094074704,"about_ca_system_score_gemma":0.0013454842,"threshold_uncertainty_score":0.02256757},"labels":[],"label_agreement":null},{"id":"W4399827830","doi":"10.1002/sim.10144","title":"Bayesian mixture modelling with ranked set samples","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba; Memorial University of Newfoundland","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Gibbs sampling; Bayesian probability; Ranking (information retrieval); Bayesian average; Computer science; Statistics; Sampling (signal processing); RSS; Simple random sample; Bayesian inference; Data mining; Variable-order Bayesian network; Mathematics; Artificial intelligence","score_opus":0.03274320391658409,"score_gpt":0.3164800722356023,"score_spread":0.28373686831901823,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4399827830","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0030072082,0.00015645345,0.99613047,0.00008372473,0.000021947702,0.000051631363,0.000059218124,0.000064904176,0.00042442704],"genre_scores_gemma":[0.2886782,0.0014295991,0.7004266,0.0003194526,0.00031017128,0.0011037639,0.0008140783,0.00014770092,0.0067704585],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9924765,0.004695122,0.00022958023,0.0008347325,0.0014289403,0.0003350697],"domain_scores_gemma":[0.9792517,0.017196253,0.0013047159,0.0010325956,0.0010273263,0.00018746026],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013530438,0.0015505916,0.0026949046,0.0022832984,0.0006847617,0.0027769792,0.004899232,0.002500758,0.0038698576],"category_scores_gemma":[0.034604356,0.0014804482,0.0024187888,0.0029001501,0.0023197047,0.0030766183,0.0020523611,0.0026015467,0.001003447],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001444472,0.00010981147,0.002774373,0.00032233892,0.0002646213,0.00024461382,0.00030592596,0.58322227,0.0010841509,0.3500388,0.0013557489,0.060132936],"study_design_scores_gemma":[0.000021021204,0.000045350756,0.00036868622,0.000033231416,0.000036636717,0.000040352148,0.000018321003,0.9343103,0.00034216928,0.063597545,0.001155017,0.00003139058],"about_ca_topic_score_codex":0.007112621,"about_ca_topic_score_gemma":0.006090486,"teacher_disagreement_score":0.013530438,"about_ca_system_score_codex":0.0014411062,"about_ca_system_score_gemma":0.0015266438,"threshold_uncertainty_score":0.07155663},"labels":[],"label_agreement":null},{"id":"W4400098172","doi":"10.1002/sim.10149","title":"Addressing dispersion in mis‐measured multivariate binomial outcomes: A novel statistical approach for detecting differentially methylated regions in bisulfite sequencing data","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre; McGill University; Université Laval; Université du Québec à Montréal; HEC Montréal; Jewish General Hospital; York University","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Alliance de recherche numérique du Canada; Genome Canada","keywords":"Multivariate statistics; Covariate; Expectation–maximization algorithm; Univariate; DNA methylation; Negative binomial distribution; Bioconductor; Computer science; Statistics; Multiplicative function; Mathematics; Biology; Poisson distribution; Genetics; Gene; Maximum likelihood","score_opus":0.42460416474090396,"score_gpt":0.4816868083342835,"score_spread":0.057082643593379556,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4400098172","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010534536,0.00008643938,0.98879236,0.00010339679,0.0000139157155,0.000033925804,0.00006244827,0.00028188812,0.000091073816],"genre_scores_gemma":[0.2639746,0.00035101606,0.7322163,0.00038903416,0.00013074481,0.00037723014,0.0009318233,0.000292582,0.0013367593],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9951592,0.0024867163,0.0002889348,0.001213684,0.0006586409,0.00019270019],"domain_scores_gemma":[0.9758902,0.018476129,0.0021547796,0.00209792,0.001015604,0.0003653255],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014947724,0.00090793194,0.0013357474,0.0015770167,0.0007047847,0.0011876153,0.0031701473,0.001908086,0.0012304737],"category_scores_gemma":[0.039016638,0.00078352867,0.0018917989,0.0016071154,0.0018964767,0.0019318067,0.0029413595,0.0026949956,0.00037450722],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00062557665,0.00020158704,0.04568231,0.00053130114,0.00080921425,0.00094991515,0.0014706625,0.52684844,0.024030795,0.08250017,0.0026066536,0.31374338],"study_design_scores_gemma":[0.000033752433,0.00007557127,0.0032608437,0.000033258286,0.00004793214,0.00018902576,0.00006642494,0.9502802,0.0029922384,0.041323382,0.0016568744,0.00004052231],"about_ca_topic_score_codex":0.0040619387,"about_ca_topic_score_gemma":0.0051299552,"teacher_disagreement_score":0.014947724,"about_ca_system_score_codex":0.0009017492,"about_ca_system_score_gemma":0.0017103098,"threshold_uncertainty_score":0.07905209},"labels":[],"label_agreement":null},{"id":"W4400302774","doi":"10.1002/sim.10154","title":"A Bayesian non‐stationary heteroskedastic time series model for multivariate critical care data","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Financial Risk and Volatility Modeling","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University","funders":"","keywords":"Heteroscedasticity; Computer science; Markov chain Monte Carlo; Multivariate statistics; Bayesian probability; Bayesian inference; State-space representation; State space; Gibbs sampling; Series (stratigraphy); Data mining; Missing data; Statistics; Artificial intelligence; Machine learning; Algorithm; Mathematics","score_opus":0.07632011854582142,"score_gpt":0.35087499666637695,"score_spread":0.2745548781205555,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4400302774","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.033602238,0.0005384845,0.9609404,0.0014223864,0.000110999375,0.00008543169,0.00078538223,0.00039503892,0.0021196464],"genre_scores_gemma":[0.8358694,0.0013840871,0.14980286,0.0005467284,0.0003103733,0.0004939981,0.0016690589,0.00016812811,0.009755392],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9982135,0.00078026747,0.00008437732,0.00041921175,0.00033690955,0.00016572588],"domain_scores_gemma":[0.9962608,0.0023276035,0.0006237562,0.00028395143,0.0003883263,0.00011543497],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005122108,0.0008361724,0.0012868022,0.0011687307,0.000441567,0.0017183839,0.002813541,0.0016822122,0.0030644247],"category_scores_gemma":[0.009877103,0.00077727047,0.0011759553,0.001565894,0.0010805705,0.0021337029,0.0010930601,0.0023752993,0.0006139325],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000099466735,0.0000976255,0.0064513446,0.00014144422,0.00019039573,0.00020749871,0.00026793085,0.66675854,0.0013626248,0.28443652,0.003539537,0.036447097],"study_design_scores_gemma":[0.00001790568,0.000029423829,0.0011744478,0.000015231015,0.000023759207,0.00004044149,0.000014352247,0.95941305,0.00014551586,0.0379093,0.001189968,0.00002663746],"about_ca_topic_score_codex":0.0095864525,"about_ca_topic_score_gemma":0.008714107,"teacher_disagreement_score":0.0095864525,"about_ca_system_score_codex":0.0014039401,"about_ca_system_score_gemma":0.0016835226,"threshold_uncertainty_score":0.027088642},"labels":[],"label_agreement":null},{"id":"W4400672891","doi":"10.1002/sim.10172","title":"Extending the DeLong algorithm for comparing areas under correlated receiver operating characteristic curves with missing data","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Missing data; Nonparametric statistics; Computer science; Receiver operating characteristic; Software; Multivariate statistics; Algorithm; Statistics; Data mining; Mathematics; Machine learning","score_opus":0.15550918662417176,"score_gpt":0.44116230951500207,"score_spread":0.28565312289083034,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4400672891","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0019749794,0.00021734514,0.9963911,0.00015190829,0.000048885428,0.00029141022,0.00016939257,0.00031669013,0.00043835893],"genre_scores_gemma":[0.031537108,0.00025556626,0.96439636,0.00024872637,0.000065588036,0.0017299249,0.00045195705,0.00040495498,0.00090984843],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9318371,0.055872437,0.0031456996,0.0040276325,0.004428896,0.0006882421],"domain_scores_gemma":[0.809582,0.15437475,0.010417666,0.013389254,0.0113180075,0.0009183594],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06604241,0.0014927542,0.0028321568,0.006082928,0.001139928,0.0030118693,0.0031694514,0.0015430945,0.0073099886],"category_scores_gemma":[0.2144182,0.0013621438,0.0029309206,0.006600841,0.0023907674,0.003274464,0.004280478,0.0044431435,0.0018431006],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012134351,0.00033844987,0.018973725,0.0016489988,0.0019792921,0.00047309036,0.001703368,0.057400014,0.0015572116,0.14051422,0.015055062,0.7591431],"study_design_scores_gemma":[0.0006976869,0.0009919866,0.00997146,0.0007291464,0.0005430892,0.0013806206,0.0005892912,0.53082824,0.003446192,0.39848924,0.05196308,0.0003700362],"about_ca_topic_score_codex":0.0064399573,"about_ca_topic_score_gemma":0.0056370576,"teacher_disagreement_score":0.06604241,"about_ca_system_score_codex":0.0015906685,"about_ca_system_score_gemma":0.0061995755,"threshold_uncertainty_score":0.3492698},"labels":[],"label_agreement":null},{"id":"W4401037694","doi":"10.1002/sim.10167","title":"Modern approaches for evaluating treatment effect heterogeneity from clinical trials and observational data","year":2024,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Bayer (Canada)","funders":"","keywords":"Observational study; Computer science; Identification (biology); Rule of thumb; Clinical trial; Medical physics; Machine learning; Artificial intelligence; Medicine; Statistics; Mathematics; Algorithm","score_opus":0.9530672804330427,"score_gpt":0.7219823764569416,"score_spread":0.23108490397610104,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401037694","genre_codex":"methods","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0008313991,0.26181307,0.7273455,0.005374742,0.0007459952,0.0005518164,0.0005391511,0.00023298638,0.0025653713],"genre_scores_gemma":[0.062910035,0.34903416,0.5722813,0.00480475,0.004273254,0.0042548208,0.0010663638,0.0002620183,0.0011133744],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.8919283,0.082296886,0.0068394653,0.0041135848,0.014443368,0.00037845827],"domain_scores_gemma":[0.60205597,0.37724167,0.007676662,0.0075172917,0.0050978903,0.00041058462],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.11392827,0.002804964,0.0060016382,0.011195754,0.00063819526,0.004624818,0.0045167296,0.003348429,0.0037064443],"category_scores_gemma":[0.24017584,0.0012773758,0.0064831567,0.00898669,0.005203971,0.0051012556,0.0036168105,0.0068665706,0.0010087591],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001479982,0.00008683471,0.0023789548,0.023121176,0.0053670476,0.00016419851,0.0004444127,0.015679952,0.0002736667,0.24585216,0.0067725144,0.6997111],"study_design_scores_gemma":[0.00017998998,0.00024230093,0.0022484497,0.008596461,0.0020873817,0.00042485818,0.00015611351,0.018553482,0.0005470078,0.92268145,0.044140045,0.00014246297],"about_ca_topic_score_codex":0.0020475704,"about_ca_topic_score_gemma":0.0014715027,"teacher_disagreement_score":0.88607174,"about_ca_system_score_codex":0.0030722765,"about_ca_system_score_gemma":0.0059716254,"threshold_uncertainty_score":0.6025175},"labels":[],"label_agreement":null},{"id":"W4401195388","doi":"10.1002/sim.10188","title":"Evaluation and comparison of covariate balance metrics in studies with time‐dependent confounding","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; Centre hospitalier de l'Université Laval","funders":"Canadian Institutes of Health Research","keywords":"Covariate; Confounding; Statistics; Inverse probability weighting; Censoring (clinical trials); Marginal structural model; Weighting; Inverse probability; Mahalanobis distance; Econometrics; Causal inference; Proportional hazards model; Residual; Propensity score matching; Computer science; Mathematics; Medicine; Bayesian probability; Posterior probability","score_opus":0.26558728525217706,"score_gpt":0.5410066149450267,"score_spread":0.2754193296928496,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401195388","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2305296,0.014797657,0.745679,0.0022426245,0.00027062607,0.0012731855,0.0022713614,0.0006109609,0.002324855],"genre_scores_gemma":[0.6501953,0.0017760122,0.34447908,0.00019489686,0.00010931764,0.0013281,0.0014359083,0.00013986982,0.00034153572],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8838681,0.09994693,0.0068404945,0.002440838,0.0064912187,0.00041237034],"domain_scores_gemma":[0.47244114,0.4784858,0.021637654,0.015040635,0.011117664,0.0012771707],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.22929187,0.0013586617,0.0022388594,0.0090062795,0.000887857,0.0027791364,0.0018097518,0.0023988262,0.0018437814],"category_scores_gemma":[0.4061198,0.00052496925,0.0026456849,0.0066528404,0.0017880226,0.0039959545,0.0037504148,0.001576564,0.0001366062],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0059937816,0.0003979044,0.22712405,0.0044802283,0.014034227,0.00036251242,0.002630454,0.2409045,0.002542395,0.08694503,0.0050373757,0.40954754],"study_design_scores_gemma":[0.0010391063,0.003302728,0.06905515,0.0013754978,0.0029449489,0.0006343696,0.0009940288,0.7939963,0.0036861377,0.1140434,0.008673485,0.00025480866],"about_ca_topic_score_codex":0.0022476858,"about_ca_topic_score_gemma":0.0019541087,"teacher_disagreement_score":0.77070814,"about_ca_system_score_codex":0.0022602526,"about_ca_system_score_gemma":0.0023113198,"threshold_uncertainty_score":0.9504203},"labels":[],"label_agreement":null},{"id":"W4402555558","doi":"10.1002/sim.10210","title":"Trivariate Joint Modeling for Family Data with Longitudinal Counts, Recurrent Events and a Terminal Event with Application to Lynch Syndrome","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Genetic factors in colorectal cancer","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Public Health Ontario; Mount Sinai Hospital; Lunenfeld-Tanenbaum Research Institute; University of Toronto; Memorial University of Newfoundland; Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Overdispersion; Statistics; Event (particle physics); Mixture model; Bivariate analysis; Random effects model; Colonoscopy; Bayesian probability; Count data; Computer science; Medicine; Econometrics; Mathematics; Poisson distribution; Colorectal cancer; Internal medicine; Cancer","score_opus":0.06002248682518295,"score_gpt":0.36620758516023605,"score_spread":0.3061850983350531,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402555558","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.037644196,0.0005239801,0.96014917,0.00037817453,0.000048889342,0.00007084085,0.00036268836,0.00030335522,0.00051877694],"genre_scores_gemma":[0.7165057,0.0015720974,0.27227738,0.00033206167,0.00015164157,0.0006699947,0.0018771274,0.00026424826,0.0063498076],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99387395,0.0041091745,0.00022795821,0.0010304955,0.00043394105,0.00032450233],"domain_scores_gemma":[0.97347057,0.021540454,0.002127274,0.0013103118,0.0011748936,0.0003766022],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014350042,0.0009679561,0.0019088048,0.0016670934,0.0008131713,0.0018169587,0.0031582217,0.0018704089,0.0029314095],"category_scores_gemma":[0.031633414,0.0009958963,0.0031495418,0.002570054,0.0018518434,0.0017869463,0.0019069588,0.0028974558,0.00042223593],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001871867,0.00010926972,0.019801699,0.0001435829,0.00059232017,0.000844216,0.0008794923,0.8369601,0.0008211134,0.10681415,0.0016446342,0.031202208],"study_design_scores_gemma":[0.000019258045,0.000039266142,0.0020441583,0.00001857391,0.00007462891,0.0000979935,0.00006415297,0.97599936,0.000098347016,0.020580962,0.0009344557,0.000028912233],"about_ca_topic_score_codex":0.054428894,"about_ca_topic_score_gemma":0.036553495,"teacher_disagreement_score":0.054428894,"about_ca_system_score_codex":0.0019381358,"about_ca_system_score_gemma":0.0023262866,"threshold_uncertainty_score":0.10822421},"labels":[],"label_agreement":null},{"id":"W4403497067","doi":"10.1002/sim.10247","title":"A Nonparametric Global Win Probability Approach to the Analysis and Sizing of Randomized Controlled Trials With Multiple Endpoints of Different Scales and Missing Data: Beyond O'Brien–Wei–Lachin","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Missing data; Statistics; Confidence interval; Nonparametric statistics; Coverage probability; Point estimation; Mathematics; Type I and type II errors; Clinical endpoint; Sample size determination; Computer science; Randomized controlled trial; Medicine","score_opus":0.303822623694461,"score_gpt":0.5183622358018237,"score_spread":0.21453961210736272,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4403497067","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0010833436,0.00041713833,0.9963437,0.0003428112,0.000076581666,0.0006010434,0.00018639799,0.00026872763,0.000680367],"genre_scores_gemma":[0.037893694,0.00059512473,0.95649374,0.00053913885,0.00016178549,0.0028966262,0.00035225105,0.00022188212,0.000845748],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.90311784,0.07620274,0.00496022,0.0054493155,0.009592624,0.0006773072],"domain_scores_gemma":[0.7248388,0.24309744,0.008775979,0.014403273,0.0076976186,0.0011869115],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.090652056,0.001936667,0.004863268,0.0054752487,0.00089547574,0.0049322625,0.003360642,0.0024053336,0.010043935],"category_scores_gemma":[0.28064948,0.0014504375,0.0040760622,0.004301346,0.0040761516,0.0058438517,0.0040530963,0.005773348,0.0013428663],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016682196,0.00016084804,0.004278233,0.0023088248,0.001388115,0.0004292262,0.00070375303,0.14277296,0.0015545825,0.38811213,0.010954295,0.44566885],"study_design_scores_gemma":[0.00039731228,0.00051032944,0.0010605184,0.0005777156,0.0002778378,0.00022952068,0.000082985614,0.4501718,0.0014464774,0.5308893,0.01425414,0.00010206762],"about_ca_topic_score_codex":0.0018787732,"about_ca_topic_score_gemma":0.0016104815,"teacher_disagreement_score":0.90934795,"about_ca_system_score_codex":0.002601168,"about_ca_system_score_gemma":0.006450843,"threshold_uncertainty_score":0.47941965},"labels":[],"label_agreement":null},{"id":"W4403879491","doi":"10.1002/sim.10236","title":"Estimands and Cumulative Incidence Function Regression in Clinical Trials: Some New Results on Interpretability and Robustness","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Interpretability; Econometrics; Estimator; Statistics; Multiplicative function; Regression; Robustness (evolution); Regression analysis; Censoring (clinical trials); Binomial regression; Computer science; Mathematics; Artificial intelligence","score_opus":0.31647089437465475,"score_gpt":0.5636028541672076,"score_spread":0.2471319597925528,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4403879491","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.004449044,0.0060902485,0.9810042,0.004341715,0.0002256166,0.00018661564,0.00022208909,0.0003127204,0.0031678025],"genre_scores_gemma":[0.32652053,0.015167626,0.6401462,0.0048408904,0.0039057636,0.0025385942,0.0009967617,0.0013492089,0.0045343875],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.86593145,0.115652904,0.0040800204,0.0057532596,0.0078059593,0.0007763876],"domain_scores_gemma":[0.23184027,0.727409,0.011547189,0.023649981,0.0049072085,0.00064635073],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1882564,0.0024069143,0.0045059216,0.006826861,0.0010220353,0.006474615,0.003954971,0.0045888056,0.005066681],"category_scores_gemma":[0.6187237,0.0014631477,0.005420248,0.0063769417,0.012089316,0.012003103,0.004790059,0.013965058,0.00087301864],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00036166055,0.00010133667,0.00592823,0.0010379843,0.0008399626,0.00060596125,0.001119741,0.04310445,0.0006616052,0.83141655,0.0036503596,0.11117221],"study_design_scores_gemma":[0.00012676764,0.00020242522,0.0016292734,0.00039160665,0.00025286677,0.00035462787,0.00010604325,0.13821407,0.0007323854,0.85197604,0.005942548,0.00007126506],"about_ca_topic_score_codex":0.0020593903,"about_ca_topic_score_gemma":0.0006249323,"teacher_disagreement_score":0.1882564,"about_ca_system_score_codex":0.002883157,"about_ca_system_score_gemma":0.0026063256,"threshold_uncertainty_score":0.9956069},"labels":[],"label_agreement":null},{"id":"W4403955442","doi":"10.1002/sim.10260","title":"A Comparison of Variance Estimators for Logistic Regression Models Estimated Using Generalized Estimating Equations (<scp>GEE</scp>) in the Context of Observational Health Services Research","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Sunnybrook Hospital; University of Toronto","funders":"Ministry of Long-Term Care; Canadian Institutes of Health Research; Ministry of Health, Ontario","keywords":"Estimator; Covariate; Statistics; Generalized estimating equation; Mathematics; Context (archaeology); Standard error; Estimating equations; Logistic regression; Econometrics; Variance (accounting); Generalized linear model; Cluster (spacecraft); Linear regression; Computer science","score_opus":0.5887665530378992,"score_gpt":0.6053092357592396,"score_spread":0.016542682721340407,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4403955442","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.046149403,0.014084921,0.9330342,0.0015998189,0.00064980413,0.00040750802,0.0005203368,0.000657058,0.0028969974],"genre_scores_gemma":[0.30025667,0.009312473,0.6832806,0.000851194,0.0004714626,0.0016523754,0.002035138,0.0008988103,0.0012412787],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8672487,0.11590016,0.0041389745,0.0041153017,0.007745184,0.00085160305],"domain_scores_gemma":[0.5564286,0.40980387,0.0071171685,0.012875985,0.013124689,0.0006496359],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.14054963,0.0012646465,0.0029117023,0.00332478,0.00054346345,0.0029382424,0.0030542181,0.002460809,0.0023679514],"category_scores_gemma":[0.4037508,0.00083891576,0.0037919835,0.0046879747,0.0016536319,0.0064477785,0.0029740725,0.0035756936,0.0005521431],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.003459196,0.00038310682,0.045068387,0.003675387,0.008235554,0.00033963888,0.0025556448,0.10294487,0.0011105848,0.20642999,0.012440067,0.61335754],"study_design_scores_gemma":[0.0017316003,0.0025528516,0.05189445,0.0043646935,0.0033643905,0.0009626109,0.0028723786,0.68498313,0.002700918,0.20808484,0.035819042,0.0006691524],"about_ca_topic_score_codex":0.004329064,"about_ca_topic_score_gemma":0.0029698662,"teacher_disagreement_score":0.14054963,"about_ca_system_score_codex":0.0015652335,"about_ca_system_score_gemma":0.0030635726,"threshold_uncertainty_score":0.7433064},"labels":[],"label_agreement":null},{"id":"W4404129156","doi":"10.1002/sim.10253","title":"Generalized Fused Lasso for Treatment Pooling in Network Meta‐Analysis","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Pooling; Lasso (programming language); Meta-analysis; Computer science; Statistics; Mathematics; Econometrics; Medicine; Artificial intelligence; Internal medicine","score_opus":0.6714134504704429,"score_gpt":0.6145489179625329,"score_spread":0.056864532507910015,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404129156","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0027496212,0.0019933411,0.99262553,0.0008271803,0.00010552464,0.00013355857,0.00049665425,0.00044058112,0.0006279739],"genre_scores_gemma":[0.17904462,0.0014846427,0.8100323,0.0017691093,0.0004897709,0.0021837011,0.0021477619,0.0008382889,0.0020098232],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9675129,0.027993234,0.0008609567,0.0019346252,0.0013642991,0.00033390682],"domain_scores_gemma":[0.97198784,0.02322558,0.0015121262,0.0019335361,0.0010698367,0.00027106624],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.043954425,0.0024557211,0.004726987,0.002806345,0.001064915,0.0033938838,0.0041870684,0.0026459328,0.00498868],"category_scores_gemma":[0.06536854,0.001402071,0.0059496164,0.002975061,0.0013889776,0.0022759538,0.0032234448,0.0049656155,0.0008556829],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00080441125,0.00012662161,0.0035033724,0.002365726,0.0069307596,0.00040352726,0.00029072203,0.7367842,0.0018730165,0.064765535,0.014970372,0.16718172],"study_design_scores_gemma":[0.00015073696,0.000081291524,0.00048127313,0.0002395083,0.0005091762,0.00006144627,0.000024429984,0.91941166,0.00057162385,0.07385353,0.0045700567,0.000045253917],"about_ca_topic_score_codex":0.004314532,"about_ca_topic_score_gemma":0.004571531,"teacher_disagreement_score":0.043954425,"about_ca_system_score_codex":0.0019755822,"about_ca_system_score_gemma":0.0037959216,"threshold_uncertainty_score":0.23245597},"labels":[],"label_agreement":null},{"id":"W4404345839","doi":"10.1002/sim.10263","title":"ℓ1$$ {\\ell}_1 $$‐Penalized Multinomial Regression: Estimation, Inference, and Prediction, With an Application to Risk Factor Identification for Different Dementia Subtypes","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Science Foundation of Sri Lanka; York University; National Institute on Aging; National Institutes of Health; National Science Foundation","keywords":"Multinomial logistic regression; Debiasing; Inference; Robustness (evolution); Logistic regression; Econometrics; Computer science; Statistics; Multinomial distribution; Regression; Confidence interval; Contrast (vision); Mathematics; Artificial intelligence; Psychology","score_opus":0.04742279060518179,"score_gpt":0.41376036239485287,"score_spread":0.3663375717896711,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404345839","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.031106839,0.00034108083,0.9669232,0.0006189901,0.000046574252,0.000053713487,0.00006471883,0.00028343173,0.0005615466],"genre_scores_gemma":[0.50297827,0.0004987176,0.49221176,0.00025777015,0.00011855736,0.000301417,0.00032281858,0.00016429054,0.0031464526],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99731183,0.001974195,0.000084289684,0.00032015727,0.0002173299,0.00009222288],"domain_scores_gemma":[0.96976423,0.026234526,0.0012615346,0.0013254136,0.0011784853,0.00023585603],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011435792,0.00077396,0.00081356306,0.0010674989,0.00045691067,0.000874706,0.0016824051,0.0010496441,0.0027296268],"category_scores_gemma":[0.04586518,0.00043076253,0.00074276474,0.0007667255,0.0013904241,0.0011348439,0.0014004817,0.0020507132,0.00033984293],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005527073,0.00021786414,0.022316583,0.00032581785,0.0002580224,0.00045627737,0.0002657098,0.6682465,0.0030924354,0.104829,0.0059116404,0.1935275],"study_design_scores_gemma":[0.000013937915,0.000023371042,0.0009406874,0.000019815536,0.000010303043,0.000037381775,0.000015363616,0.98974293,0.00042323503,0.008436101,0.00032737243,0.000009553835],"about_ca_topic_score_codex":0.005684697,"about_ca_topic_score_gemma":0.0039150747,"teacher_disagreement_score":0.011435792,"about_ca_system_score_codex":0.0006418001,"about_ca_system_score_gemma":0.001196713,"threshold_uncertainty_score":0.060478985},"labels":[],"label_agreement":null},{"id":"W4404567028","doi":"10.1002/sim.10223","title":"Q‐Learning in Dynamic Treatment Regimes With Misclassified Binary Outcome","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Outcome (game theory); Artificial intelligence; Machine learning; Population; Binary classification; Statistics; Data mining; Medicine; Mathematics; Support vector machine","score_opus":0.10451381065403711,"score_gpt":0.4508819754438676,"score_spread":0.34636816478983046,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404567028","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029349865,0.0010812592,0.9669311,0.0013656219,0.000061404215,0.00017780425,0.0001048514,0.00014290598,0.0007852113],"genre_scores_gemma":[0.7269988,0.0013788895,0.26714373,0.0008205979,0.0002256274,0.0007454679,0.00031481715,0.00006919336,0.002302941],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9824614,0.013178543,0.0006328222,0.0022169168,0.0009197136,0.0005906757],"domain_scores_gemma":[0.8203136,0.16812448,0.00596136,0.0024543446,0.0023439075,0.00080233894],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04634653,0.001143395,0.0037088965,0.0015747381,0.0009953694,0.0019141356,0.0027366115,0.0027603789,0.002043276],"category_scores_gemma":[0.104782805,0.00095754483,0.0014395377,0.0017405519,0.003812018,0.0028068225,0.0018206154,0.0042742006,0.00024856522],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00052682945,0.00020606774,0.006914867,0.00037259926,0.00022181483,0.0003048314,0.0003858534,0.8236507,0.0004713475,0.07643753,0.0011192538,0.08938829],"study_design_scores_gemma":[0.00008585947,0.00011391945,0.000707057,0.000042043728,0.000032817657,0.000049438477,0.000026580172,0.93455213,0.00026096747,0.06365238,0.00045745537,0.000019286343],"about_ca_topic_score_codex":0.008532523,"about_ca_topic_score_gemma":0.0036243962,"teacher_disagreement_score":0.04634653,"about_ca_system_score_codex":0.0027700618,"about_ca_system_score_gemma":0.0035213935,"threshold_uncertainty_score":0.24510682},"labels":[],"label_agreement":null},{"id":"W4404837160","doi":"10.1002/sim.10278","title":"Dynamic Treatment Regimes on Dyadic Networks","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Dyad; Computer science; Context (archaeology); Outcome (game theory); Population; Path (computing); Mathematical optimization; Econometrics; Machine learning; Mathematics; Medicine; Psychology","score_opus":0.07713181625879532,"score_gpt":0.45058179005342597,"score_spread":0.37344997379463063,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404837160","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.10225388,0.000800923,0.88937414,0.0013910716,0.0000766637,0.00029408644,0.0008012408,0.00016051673,0.004847463],"genre_scores_gemma":[0.8514774,0.0013140751,0.13519596,0.0003707272,0.00013338201,0.0010836956,0.0010104737,0.00008443411,0.009329797],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9930179,0.0047682016,0.00016788022,0.0013399592,0.00035239663,0.00035365546],"domain_scores_gemma":[0.977725,0.018432176,0.0020884234,0.0008268765,0.00053103425,0.0003964305],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.007943798,0.0009270126,0.002179682,0.0012016308,0.0009888838,0.0017757196,0.002632052,0.0018428083,0.0066444413],"category_scores_gemma":[0.031834114,0.0010193038,0.001243834,0.0014249933,0.0018503449,0.0030419424,0.0024669736,0.00242425,0.00059952383],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00029896485,0.00012355373,0.0058977786,0.00015370529,0.00022003666,0.00033097758,0.00030945856,0.844201,0.00039933532,0.11486659,0.0016898772,0.03150869],"study_design_scores_gemma":[0.000050345705,0.00006647269,0.00163675,0.000039944418,0.00005798194,0.000057388792,0.000112031594,0.90538067,0.00013832671,0.091014,0.0014224392,0.00002364332],"about_ca_topic_score_codex":0.013482488,"about_ca_topic_score_gemma":0.011371596,"teacher_disagreement_score":0.9920562,"about_ca_system_score_codex":0.0022238777,"about_ca_system_score_gemma":0.0013258234,"threshold_uncertainty_score":0.04201132},"labels":[],"label_agreement":null},{"id":"W4404898302","doi":"10.1002/sim.10277","title":"Bayesian Decision Curve Analysis With Bayesdca","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"BC Children's Hospital; University of British Columbia","funders":"Provincial Health Services Authority; BC Children's Hospital; Children's Hospital Foundation","keywords":"Frequentist inference; Computer science; Bayesian probability; Workflow; Machine learning; Decision analysis; False positives and false negatives; False positive paradox; Probabilistic logic; Bayes' theorem; Artificial intelligence; Data mining; Bayesian inference; Statistics; Mathematics","score_opus":0.17346630158868082,"score_gpt":0.4589526291861392,"score_spread":0.2854863275974584,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404898302","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.002650199,0.0023137126,0.98727554,0.0011492026,0.0001759334,0.0005991544,0.001048929,0.0009955439,0.0037918587],"genre_scores_gemma":[0.14151593,0.0033323132,0.8409639,0.0012836644,0.0005253483,0.003357847,0.0024706174,0.0008707561,0.005679614],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.97625256,0.01751179,0.0013358615,0.0021372316,0.0022396864,0.0005228406],"domain_scores_gemma":[0.91710496,0.07236367,0.002610052,0.0029192625,0.004285492,0.0007165052],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03746584,0.0022212723,0.004514522,0.0052252933,0.0013708117,0.0070972345,0.0039265705,0.0044563627,0.020398298],"category_scores_gemma":[0.13266405,0.0020793641,0.0050137583,0.0048670177,0.0019632361,0.0035637906,0.003520711,0.0064592627,0.0052345414],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000729127,0.0001705553,0.004004623,0.0010241723,0.0009650937,0.00031418705,0.00036159,0.5823365,0.00024842902,0.16982728,0.02097036,0.2190481],"study_design_scores_gemma":[0.00015321703,0.00007991261,0.00047723664,0.0003555769,0.00019681639,0.0001804195,0.00004465994,0.76751536,0.00027266963,0.21427605,0.016352843,0.00009532325],"about_ca_topic_score_codex":0.019620711,"about_ca_topic_score_gemma":0.011482223,"teacher_disagreement_score":0.9625342,"about_ca_system_score_codex":0.0035304653,"about_ca_system_score_gemma":0.0075669037,"threshold_uncertainty_score":0.19814068},"labels":[],"label_agreement":null},{"id":"W4406146487","doi":"10.1002/sim.10273","title":"Editorial","year":2025,"lang":"en","type":"editorial","venue":"Statistics in Medicine","topic":"Health and Medical Research Impacts","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Gratitude; Editor in chief; Editorial board; Library science; Sadness; History; Management; Psychology; Computer science; Economics","score_opus":0.04959678286750923,"score_gpt":0.4880930862123854,"score_spread":0.43849630334487616,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4406146487","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00015736997,0.0071408134,0.00058196054,0.058423467,0.9037395,0.00006261117,0.0005569383,0.00048502634,0.02885232],"genre_scores_gemma":[0.0035237574,0.01297045,0.00088809716,0.0644215,0.7738477,0.000117339616,0.0012644406,0.00065921724,0.1423076],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9913974,0.0010539021,0.00081432017,0.0015923633,0.0044455477,0.0006963901],"domain_scores_gemma":[0.95816904,0.0056071575,0.0024334984,0.003084566,0.025176894,0.0055288645],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0053704535,0.0017182105,0.0015879846,0.0027621042,0.002575793,0.0088195335,0.002752657,0.0052578514,0.19674341],"category_scores_gemma":[0.048359074,0.00053206773,0.0014394802,0.0011591745,0.0017176355,0.005207695,0.0022489254,0.0069527742,0.1444259],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000015985761,0.0000054611633,0.000036265,0.00010957529,0.000004769466,0.000057359295,0.00001798478,0.000008624551,0.00003674708,0.00072043686,0.9856188,0.013367991],"study_design_scores_gemma":[0.000009493347,0.000010032845,0.000074342955,0.00022448794,0.0000056012673,0.000175751,0.00003862133,0.000018906621,0.00006344879,0.0008485396,0.99852496,0.000005795241],"about_ca_topic_score_codex":0.0006851929,"about_ca_topic_score_gemma":0.0010434744,"teacher_disagreement_score":0.8032566,"about_ca_system_score_codex":0.0020592394,"about_ca_system_score_gemma":0.0044401283,"threshold_uncertainty_score":0},"labels":[],"label_agreement":null},{"id":"W4406801646","doi":"10.1002/sim.10324","title":"Supervised Functional Principal Component Analysis Under the Mixture Cure Rate Model: An Application to Alzheimer'S Disease","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"National Institute on Aging; Natural Sciences and Engineering Research Council of Canada","keywords":"Principal component analysis; Computer science; Censoring (clinical trials); Bivariate analysis; Artificial intelligence; Disease; Functional principal component analysis; Proportional hazards model; Neuroimaging; Event (particle physics); Pattern recognition (psychology); Machine learning; Econometrics; Statistics; Mathematics; Medicine; Psychiatry; Internal medicine","score_opus":0.08268806086690261,"score_gpt":0.40944814030246357,"score_spread":0.32676007943556096,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4406801646","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.030933362,0.00030060665,0.96771383,0.0003066993,0.000028973298,0.000044497865,0.000068454756,0.00030076146,0.00030292475],"genre_scores_gemma":[0.57969165,0.0007014144,0.4156328,0.00014236392,0.00014386853,0.00028457842,0.0004493725,0.00017027296,0.002783654],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99862826,0.0009033497,0.000037680944,0.00017267866,0.0001791674,0.000078795274],"domain_scores_gemma":[0.9961029,0.0027135457,0.00027999817,0.0003282799,0.00047537265,0.000099933204],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0052855616,0.0007957534,0.0011167635,0.0009807247,0.00042770564,0.000688995,0.0010769669,0.0011244151,0.0008304561],"category_scores_gemma":[0.010779075,0.00037765683,0.0015062776,0.00093394186,0.00092529715,0.00089757884,0.0012174955,0.0017820317,0.00029575403],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00036657046,0.00014835467,0.0060093445,0.00014930706,0.00020902535,0.00027112378,0.00027571185,0.75866455,0.0040214034,0.026824716,0.002551386,0.2005084],"study_design_scores_gemma":[0.000006607063,0.000024362245,0.00053466536,0.0000036763922,0.000007793045,0.000039387258,0.0000071066806,0.99403715,0.00024226027,0.004805162,0.00028162537,0.000010161779],"about_ca_topic_score_codex":0.0055201426,"about_ca_topic_score_gemma":0.003741423,"teacher_disagreement_score":0.0055201426,"about_ca_system_score_codex":0.0004746167,"about_ca_system_score_gemma":0.001128311,"threshold_uncertainty_score":0.027953029},"labels":[],"label_agreement":null},{"id":"W4406808516","doi":"10.1002/sim.10321","title":"Bioequivalence Design With Sampling Distribution Segments","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; McGill University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Bioequivalence; Sampling (signal processing); Sample size determination; Computer science; Statistics; Sampling distribution; Sampling design; Mathematics; Econometrics; Medicine","score_opus":0.22427735462934364,"score_gpt":0.5229092611377446,"score_spread":0.298631906508401,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4406808516","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0055742334,0.00015022451,0.99237084,0.00013428053,0.00002924378,0.00069181283,0.00008562475,0.00031899978,0.00064478576],"genre_scores_gemma":[0.20399776,0.00035198175,0.7862203,0.000433586,0.00009165051,0.0061839153,0.0006735022,0.00024611744,0.0018011997],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97353506,0.021472313,0.00070968195,0.0018772993,0.0020240434,0.00038171106],"domain_scores_gemma":[0.92507756,0.06447046,0.0027501709,0.005209549,0.0020503558,0.00044184772],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.037092462,0.0014848044,0.0026351218,0.0018736334,0.0007107259,0.0020074474,0.0022346408,0.0027083342,0.0075302906],"category_scores_gemma":[0.09793398,0.0011921972,0.0026545501,0.0016142619,0.0021856064,0.003627513,0.0027917165,0.004328813,0.001260071],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004760281,0.0007463434,0.007810885,0.0010187814,0.000597153,0.00037774647,0.0011140158,0.4499034,0.004385897,0.21457118,0.004848245,0.30986616],"study_design_scores_gemma":[0.00051480316,0.00095160014,0.0011567535,0.0001483459,0.00012482212,0.000114467766,0.00006134747,0.8365654,0.0029661017,0.1500424,0.0072929375,0.00006112334],"about_ca_topic_score_codex":0.0016169039,"about_ca_topic_score_gemma":0.0010522512,"teacher_disagreement_score":0.037092462,"about_ca_system_score_codex":0.001740301,"about_ca_system_score_gemma":0.0028220506,"threshold_uncertainty_score":0.19616604},"labels":[],"label_agreement":null},{"id":"W4407173849","doi":"10.1002/sim.70000","title":"Regression—A Means, Not an End","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Statistics; Regression; Regression analysis; Mathematics","score_opus":0.1246073718061269,"score_gpt":0.4895811887662412,"score_spread":0.3649738169601143,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407173849","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0048833108,0.030949203,0.61562526,0.25506115,0.011610174,0.00020209647,0.0015456998,0.0017053535,0.078417785],"genre_scores_gemma":[0.27823862,0.054923248,0.4698322,0.0822936,0.03664217,0.0013173489,0.0016792457,0.003106848,0.07196672],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97846884,0.011610426,0.0011213262,0.003879528,0.0044747563,0.00044516288],"domain_scores_gemma":[0.9503317,0.02831773,0.002903665,0.013205174,0.004318414,0.00092333247],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.027956713,0.0018568151,0.0032286488,0.0024788515,0.0027653596,0.011911838,0.0026945216,0.0044068377,0.011686993],"category_scores_gemma":[0.066279456,0.0012738112,0.0015429439,0.0019518611,0.018110774,0.021140438,0.0066259583,0.011370668,0.010327879],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015367847,0.00006413778,0.0018569605,0.00072882423,0.00026244178,0.0001013153,0.0006715777,0.0012043878,0.00055244134,0.8522466,0.052914966,0.08924272],"study_design_scores_gemma":[0.00003306888,0.00009947698,0.00078783894,0.0008146671,0.000074595184,0.00021423721,0.0003649621,0.003450495,0.0007025951,0.89803416,0.09534507,0.00007886857],"about_ca_topic_score_codex":0.0012132515,"about_ca_topic_score_gemma":0.001528043,"teacher_disagreement_score":0.027956713,"about_ca_system_score_codex":0.0021107541,"about_ca_system_score_gemma":0.003950087,"threshold_uncertainty_score":0.14785099},"labels":[],"label_agreement":null},{"id":"W4407262364","doi":"10.1002/sim.70009","title":"Inverse Probability of Treatment Weighting Using the Propensity Score With Competing Risks in Survival Analysis","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research","keywords":"Propensity score matching; Inverse probability weighting; Inverse probability; Statistics; Weighting; Survival analysis; Econometrics; Inverse; Average treatment effect; Computer science; Posterior probability; Mathematics; Medicine; Bayesian probability","score_opus":0.3133595065600282,"score_gpt":0.4585707417670098,"score_spread":0.1452112352069816,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407262364","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010381333,0.00091188424,0.9863024,0.00050577405,0.00015046699,0.00086957106,0.00014268621,0.00015447271,0.0005814027],"genre_scores_gemma":[0.2972577,0.0013887846,0.69482034,0.0005318099,0.00029119282,0.004337751,0.00038158428,0.0001295278,0.0008613161],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.87236094,0.11398224,0.0039700647,0.0036779302,0.0052146865,0.000794175],"domain_scores_gemma":[0.90004545,0.0761983,0.009461967,0.01019146,0.0035219032,0.00058091374],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09151755,0.0012332745,0.002280286,0.0037967418,0.00073613424,0.0028589405,0.0024817996,0.0024866892,0.0029655613],"category_scores_gemma":[0.19616625,0.0009584384,0.0046631615,0.005011166,0.002634172,0.0029683104,0.0039003599,0.00426157,0.000345077],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013615149,0.0004963898,0.056466352,0.0020047512,0.00480065,0.0005062932,0.0016062767,0.210506,0.0013898077,0.2803167,0.006586877,0.43395832],"study_design_scores_gemma":[0.001125556,0.0014072354,0.016260859,0.0009177076,0.0013897349,0.0005906212,0.0002247537,0.5754505,0.002612992,0.3848627,0.014813914,0.00034350785],"about_ca_topic_score_codex":0.0032428454,"about_ca_topic_score_gemma":0.002188918,"teacher_disagreement_score":0.09151755,"about_ca_system_score_codex":0.0017898984,"about_ca_system_score_gemma":0.00338511,"threshold_uncertainty_score":0.48399687},"labels":[],"label_agreement":null},{"id":"W4407295943","doi":"10.1002/sim.70013","title":"Challenges for Predictive Modeling With Neural Network Techniques Using Error‐Prone Dietary Intake Data","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Nutritional Studies and Diet","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; University of New Brunswick","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Overfitting; Computer science; Machine learning; Artificial neural network; Artificial intelligence; Replicate; Predictive modelling; Data mining; Statistics; Mathematics","score_opus":0.16158573192486647,"score_gpt":0.40385528688710676,"score_spread":0.24226955496224029,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407295943","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.013035942,0.0025392142,0.9740249,0.00771871,0.00016317943,0.0001227456,0.00027258747,0.000417722,0.0017049526],"genre_scores_gemma":[0.4587928,0.0055033728,0.5276871,0.0024967566,0.0008357848,0.0009014811,0.0008148181,0.00031690777,0.0026509552],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9810126,0.014035967,0.0010214694,0.0015847209,0.002089487,0.00025575195],"domain_scores_gemma":[0.8408537,0.143262,0.003915398,0.00671547,0.0048207864,0.00043265498],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.056306742,0.0017036031,0.0021258593,0.002152742,0.0012010874,0.0049544466,0.004427749,0.002348889,0.0014901225],"category_scores_gemma":[0.16045861,0.0014617,0.0015798195,0.0025061734,0.003306751,0.005110102,0.0041533574,0.008122899,0.0007361829],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001292481,0.00011528864,0.012098316,0.0005583039,0.0006488986,0.00022144256,0.0005601564,0.76866657,0.0006113863,0.076480016,0.0033144443,0.13659595],"study_design_scores_gemma":[0.000014004342,0.00002816029,0.000906478,0.00012738901,0.000026452732,0.000040316794,0.000074113064,0.9104692,0.0002889033,0.08601858,0.001975331,0.00003107322],"about_ca_topic_score_codex":0.019460173,"about_ca_topic_score_gemma":0.013584623,"teacher_disagreement_score":0.056306742,"about_ca_system_score_codex":0.0025167023,"about_ca_system_score_gemma":0.0034032399,"threshold_uncertainty_score":0.29778206},"labels":[],"label_agreement":null},{"id":"W4407777323","doi":"10.1002/sim.70007","title":"Identification and Estimation of the Average Causal Effects Under Dietary Substitution Strategies","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Nutritional Studies and Diet","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"National Institute of General Medical Sciences; National Institutes of Health; National Cancer Institute; Natural Sciences and Engineering Research Council of Canada; Harvard T.H. Chan School of Public Health; Brigham and Women's Hospital","keywords":"Counterfactual thinking; Substitution (logic); Estimator; Estimation; Red meat; Econometrics; Intervention (counseling); Identification (biology); Food science; Statistics; Computer science; Medicine; Environmental health; Mathematics; Economics; Psychology; Biology; Social psychology","score_opus":0.01628704538152652,"score_gpt":0.3313234211693104,"score_spread":0.3150363757877839,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407777323","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.13760369,0.00059127604,0.8581462,0.00076908787,0.000043779808,0.00049119914,0.00041058465,0.00018766311,0.0017565296],"genre_scores_gemma":[0.7659657,0.0008909177,0.22861871,0.0002732381,0.00006038542,0.0010032575,0.00053309876,0.000035540626,0.0026190975],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98022515,0.01570075,0.00076032546,0.0018986416,0.0008803623,0.0005347485],"domain_scores_gemma":[0.7746629,0.20955043,0.007086064,0.0062373243,0.0020163662,0.0004468899],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04408759,0.0012585735,0.00298626,0.0019322002,0.00061858364,0.0021243799,0.0025782248,0.0028978747,0.0030745964],"category_scores_gemma":[0.16559492,0.0011282057,0.0024534387,0.0020688903,0.0032687355,0.003591421,0.0022056585,0.0030465282,0.00033660012],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006170464,0.00038164968,0.035558596,0.0004342457,0.0010511299,0.00036843258,0.0007377525,0.6948264,0.001077185,0.168835,0.0008801863,0.09523241],"study_design_scores_gemma":[0.00015389288,0.00035657547,0.008186955,0.00012341407,0.00022511164,0.000116329065,0.0002620971,0.85822254,0.00090391125,0.13024704,0.0011351393,0.000066955305],"about_ca_topic_score_codex":0.007364685,"about_ca_topic_score_gemma":0.0047046063,"teacher_disagreement_score":0.04408759,"about_ca_system_score_codex":0.0015542908,"about_ca_system_score_gemma":0.0027129864,"threshold_uncertainty_score":0.23316026},"labels":[],"label_agreement":null},{"id":"W4407860565","doi":"10.1002/sim.70025","title":"A Double Machine Learning Approach for the Evaluation of COVID‐19 Vaccine Effectiveness Under the Test‐Negative Design: Analysis of Québec Administrative Data","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Influenza Virus Research Studies","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Institut National de Santé Publique du Québec; Université Laval; Université de Montréal","funders":"Canadian Institutes of Health Research; Canada Research Chairs; Fonds de Recherche du Québec - Santé; Centre de Recherches Mathématiques","keywords":"Estimator; Econometrics; Identifiability; Consistency (knowledge bases); Causal inference; Computer science; Nonparametric statistics; Robustness (evolution); Confounding; Asymptotic distribution; Mathematics; Coronavirus disease 2019 (COVID-19); Statistics; Artificial intelligence; Medicine; Biology","score_opus":0.36590857376146146,"score_gpt":0.5361572645866105,"score_spread":0.17024869082514904,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407860565","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.26944163,0.0015648504,0.7218562,0.001196051,0.00014813388,0.0009426241,0.0014759528,0.0003524693,0.0030222035],"genre_scores_gemma":[0.8919541,0.0001612811,0.10427139,0.0002446331,0.00006199136,0.00067636516,0.0007495374,0.000033481592,0.001847227],"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","domain_scores_codex":[0.95370704,0.039868042,0.0008311294,0.002711844,0.0019137984,0.0009682273],"domain_scores_gemma":[0.82814705,0.146069,0.009560901,0.009766092,0.0054308926,0.0010260455],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07333119,0.0011185234,0.0019485905,0.0022997514,0.0009148381,0.0013289504,0.0029597003,0.0014372373,0.0043940526],"category_scores_gemma":[0.12943351,0.00047315404,0.002264,0.0014557997,0.002326736,0.0010582352,0.0016064116,0.0016829964,0.00018387711],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0032098189,0.000676113,0.22406358,0.0009614467,0.004195212,0.00075624144,0.0005672339,0.43257293,0.0022175352,0.1307136,0.004847198,0.19521908],"study_design_scores_gemma":[0.00021358745,0.00063207815,0.03339447,0.0000989327,0.00025660626,0.00006944653,0.000119618875,0.9461524,0.00078848464,0.015832396,0.0023902555,0.000051739262],"about_ca_topic_score_codex":0.10210893,"about_ca_topic_score_gemma":0.058191635,"teacher_disagreement_score":0.10210893,"about_ca_system_score_codex":0.0046637347,"about_ca_system_score_gemma":0.0046354504,"threshold_uncertainty_score":0.38781703},"labels":[],"label_agreement":null},{"id":"W4408408683","doi":"10.1002/sim.70034","title":"Guidelines and Best Practices for the Use of Targeted Maximum Likelihood and Machine Learning When Estimating Causal Effects of Exposures on Time‐To‐Event Outcomes","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université de Montréal; Institut National de Santé Publique du Québec; Université Laval","funders":"Canadian Institutes of Health Research; Canada Research Chairs; Fonds de Recherche du Québec - Santé; International Society for Pharmacoepidemiology","keywords":"Computer science; Outcome (game theory); Machine learning; Counterfactual thinking; Event (particle physics); Causal inference; Covariate; A priori and a posteriori; Artificial intelligence; Data mining; Statistics","score_opus":0.16736663226226037,"score_gpt":0.4647206712009353,"score_spread":0.29735403893867496,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408408683","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0007541157,0.0062302537,0.9473961,0.0066751908,0.0008687826,0.00096309336,0.0072229453,0.016165946,0.01372353],"genre_scores_gemma":[0.00408799,0.0058806487,0.96662116,0.003047697,0.00041697518,0.0035970712,0.004747573,0.006188829,0.0054121013],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9712355,0.018150482,0.0041303374,0.0016255353,0.004416645,0.00044140406],"domain_scores_gemma":[0.85315865,0.11806927,0.004535336,0.010785911,0.012026663,0.001424219],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.029976755,0.0031585882,0.0021224166,0.006100195,0.0011117879,0.0046556913,0.0060041677,0.0068896445,0.062241163],"category_scores_gemma":[0.16925088,0.0029589676,0.0037541026,0.005626542,0.0019980976,0.004420984,0.0035527623,0.007838032,0.051685583],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00020045166,0.0003246097,0.0016869914,0.006498113,0.00030235,0.0011411549,0.0015419202,0.015831111,0.002497226,0.12863722,0.45901904,0.38231975],"study_design_scores_gemma":[0.00025522924,0.00012980468,0.0014238366,0.0058198003,0.00012569396,0.0017881104,0.00029685395,0.024358071,0.004138627,0.24326523,0.718121,0.00027784315],"about_ca_topic_score_codex":0.0058073425,"about_ca_topic_score_gemma":0.009070295,"teacher_disagreement_score":0.9700233,"about_ca_system_score_codex":0.0018416888,"about_ca_system_score_gemma":0.005642826,"threshold_uncertainty_score":0.20821744},"labels":[],"label_agreement":null},{"id":"W4408651332","doi":"10.1002/sim.70023","title":"Variable Selection for Progressive Multistate Processes Under Intermittent Observation","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Computer science; Maximization; Variable (mathematics); Selection (genetic algorithm); Feature selection; Expectation–maximization algorithm; Regression; Exploit; Poisson regression; Machine learning; Artificial intelligence; Statistics; Mathematical optimization; Maximum likelihood; Mathematics; Medicine","score_opus":0.10407953292784239,"score_gpt":0.4503327739983338,"score_spread":0.34625324107049144,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408651332","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.023072349,0.00019783451,0.97570884,0.00038882234,0.000029078581,0.000075292795,0.00008221589,0.00014780666,0.00029768713],"genre_scores_gemma":[0.6774804,0.0008558915,0.313493,0.00041014596,0.0003011674,0.0009421441,0.0010761402,0.000185057,0.00525599],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9941362,0.0038149673,0.00021062806,0.0010542,0.000462204,0.00032186636],"domain_scores_gemma":[0.9459086,0.048212998,0.0025625466,0.0013265228,0.0012736112,0.0007157193],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.018950982,0.001115604,0.0024525824,0.0015250345,0.0011349315,0.0018801965,0.0030949763,0.0020054118,0.0034798277],"category_scores_gemma":[0.046378784,0.0010747034,0.0023399934,0.0017686355,0.0031747175,0.0024756077,0.0036860765,0.0035444533,0.0004499176],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003994993,0.00015413377,0.011431852,0.00024592856,0.00036028048,0.00073328626,0.0004507094,0.79700905,0.0012632636,0.14049469,0.001305422,0.04615184],"study_design_scores_gemma":[0.000031045074,0.000041520205,0.00055181433,0.000017612749,0.000022774922,0.000030568168,0.0000244196,0.9567895,0.00024396332,0.04193102,0.00030084656,0.000014934779],"about_ca_topic_score_codex":0.004649345,"about_ca_topic_score_gemma":0.004485094,"teacher_disagreement_score":0.018950982,"about_ca_system_score_codex":0.0011128098,"about_ca_system_score_gemma":0.0019847334,"threshold_uncertainty_score":0.1002236},"labels":[],"label_agreement":null},{"id":"W4409368810","doi":"10.1002/sim.70049","title":"A Generalized Phase I/II Dose Optimization Trial Design With Multi‐Categorical and Multi‐Graded Outcomes","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University; Simon Fraser University","funders":"Canadian Statistical Sciences Institute; Natural Sciences and Engineering Research Council of Canada; National Natural Science Foundation of China","keywords":"Categorical variable; Monotonic function; Benchmarking; Robustness (evolution); Computer science; Maximum tolerated dose; Optimal design; Clinical trial; Continuous variable; Reliability (semiconductor); Reliability engineering; Mathematical optimization; Medicine; Mathematics; Machine learning","score_opus":0.5347114705751628,"score_gpt":0.5995985752183284,"score_spread":0.0648871046431656,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409368810","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.042119645,0.00074507814,0.9418533,0.000747292,0.0002185897,0.011870596,0.00045986313,0.00039879658,0.0015868054],"genre_scores_gemma":[0.36330718,0.00044024838,0.599354,0.0014740084,0.00011481248,0.03260459,0.000414608,0.00008112189,0.002209417],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9719297,0.023870563,0.0007843833,0.0018255782,0.0011797919,0.00040994305],"domain_scores_gemma":[0.98816746,0.007073447,0.0016051542,0.0019693675,0.00074302277,0.000441513],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.027825333,0.0016545628,0.002802215,0.0007516405,0.00042437803,0.0011322573,0.0017916373,0.0017258782,0.0044055767],"category_scores_gemma":[0.029371709,0.00072119874,0.0021546604,0.0008361079,0.0018815849,0.0009944419,0.0014661856,0.0026831182,0.0005793574],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.026940074,0.0026070296,0.011338707,0.0037520125,0.002866404,0.00057641655,0.0006008149,0.494186,0.021440249,0.13167454,0.0065193614,0.29749835],"study_design_scores_gemma":[0.018497445,0.025292113,0.00670362,0.00040354466,0.0018986039,0.00057770906,0.00011919512,0.79192215,0.009773413,0.13098778,0.01354928,0.00027519016],"about_ca_topic_score_codex":0.000347725,"about_ca_topic_score_gemma":0.00045557946,"teacher_disagreement_score":0.027825333,"about_ca_system_score_codex":0.0007761798,"about_ca_system_score_gemma":0.0025882204,"threshold_uncertainty_score":0.14715624},"labels":[],"label_agreement":null},{"id":"W4409800482","doi":"10.1002/sim.70074","title":"A Variance Estimator for Marginal Cox Regression Models Fit to Non‐Nested Multilevel Data","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sunnybrook Hospital; University of Toronto","funders":"Canadian Institutes of Health Research","keywords":"Marginal model; Statistics; Estimator; Generalized linear model; Multilevel model; Generalized estimating equation; Mathematics; Regression analysis; Proportional hazards model; Covariate; Censored regression model; Variance (accounting); Econometrics","score_opus":0.18844865111631784,"score_gpt":0.4953526763331343,"score_spread":0.3069040252168165,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409800482","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011028851,0.00015374135,0.997971,0.0001515112,0.0000552499,0.00008819854,0.00009023174,0.00018853246,0.00019873516],"genre_scores_gemma":[0.0895064,0.00058705354,0.90450925,0.0004639306,0.00026088857,0.0015968077,0.0009059046,0.00046096538,0.0017087191],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97625345,0.018095085,0.0007683353,0.002027913,0.0023079866,0.0005472218],"domain_scores_gemma":[0.94392854,0.0441467,0.0025000584,0.0047763847,0.004311077,0.0003372709],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03705213,0.0012157358,0.0020930744,0.002071707,0.0006370315,0.0015411635,0.003507368,0.002222305,0.004377301],"category_scores_gemma":[0.1252724,0.0009950788,0.0034851877,0.0024054593,0.0012989047,0.002830513,0.0025529794,0.004529393,0.001159413],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00034719618,0.00020709286,0.02196972,0.00089542137,0.001858471,0.00040730325,0.00067291997,0.16888548,0.0021733437,0.43721735,0.021485237,0.34388044],"study_design_scores_gemma":[0.00018114185,0.0003351414,0.0044795796,0.00029055413,0.00038316587,0.00038929007,0.00017036806,0.7324506,0.0014021507,0.23998149,0.019757012,0.00017958453],"about_ca_topic_score_codex":0.005078225,"about_ca_topic_score_gemma":0.0049360036,"teacher_disagreement_score":0.03705213,"about_ca_system_score_codex":0.0011658698,"about_ca_system_score_gemma":0.0031492782,"threshold_uncertainty_score":0.19595271},"labels":[],"label_agreement":null},{"id":"W4410429608","doi":"10.1002/sim.70077","title":"A Personalized Predictive Model That Jointly Optimizes Discrimination and Calibration","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Weighting; Calibration; Computer science; Flexibility (engineering); Function (biology); Measure (data warehouse); Population; Field (mathematics); A-weighting; Machine learning; Artificial intelligence; Similarity (geometry); Statistics; Econometrics; Data mining; Mathematics; Medicine","score_opus":0.028596481618226307,"score_gpt":0.34266400478740733,"score_spread":0.314067523169181,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410429608","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04538825,0.0002255725,0.9508155,0.001250359,0.00004303862,0.00009792523,0.00025676913,0.00051825325,0.0014043553],"genre_scores_gemma":[0.7285581,0.0002953968,0.2634145,0.001067072,0.00016603606,0.00047092902,0.00089324877,0.00020310207,0.00493165],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9978969,0.00091840076,0.00009298901,0.0006534918,0.0002574445,0.00018071264],"domain_scores_gemma":[0.9941736,0.0040313536,0.0005652959,0.0005303403,0.000499455,0.00020001964],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005534981,0.0011178273,0.002419668,0.0009235752,0.00062456314,0.0018803457,0.002631201,0.0028396242,0.0017325999],"category_scores_gemma":[0.018925631,0.0008007815,0.0013613435,0.0012507519,0.0011057125,0.0028251265,0.002157883,0.0034482367,0.0006670392],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015339616,0.00008189294,0.0043454724,0.000047541223,0.00008696972,0.000095955984,0.00010495154,0.92620313,0.0010032771,0.014279108,0.002032608,0.051565737],"study_design_scores_gemma":[0.000018533168,0.000050935818,0.0004814761,0.000014500159,0.000027568527,0.0000623325,0.000014774802,0.98365545,0.00041118357,0.014706419,0.0005405728,0.000016242613],"about_ca_topic_score_codex":0.00468949,"about_ca_topic_score_gemma":0.003230233,"teacher_disagreement_score":0.005534981,"about_ca_system_score_codex":0.001492229,"about_ca_system_score_gemma":0.0019459882,"threshold_uncertainty_score":0.02927208},"labels":[],"label_agreement":null},{"id":"W4410480739","doi":"10.1002/sim.70102","title":"Continuous‐Time Causal Inference With Marked Point Process Weights: An Example on Sodium‐Glucose Co‐Transporters 2 Inhibitor Medications and Urinary Tract Infection","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba; Public Health Ontario; University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Marginal structural model; Covariate; Confounding; Medicine; Causal inference; Time point; Clinical endpoint; Internal medicine; Statistics; Clinical trial; Mathematics","score_opus":0.043079783093944224,"score_gpt":0.39116520241493435,"score_spread":0.34808541932099013,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410480739","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.059592836,0.000649789,0.9365047,0.0013438875,0.00010344335,0.0001792472,0.00027250263,0.00014682033,0.0012067548],"genre_scores_gemma":[0.7415609,0.0009530227,0.2531124,0.00046382193,0.00023009787,0.000649848,0.00045648625,0.000095821975,0.0024775155],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.97727394,0.018765785,0.0005058614,0.0021522846,0.00087170437,0.0004304407],"domain_scores_gemma":[0.8112254,0.17470136,0.0043131253,0.0064090146,0.0025468876,0.0008041224],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0471253,0.0011469962,0.0018537862,0.0017228042,0.0009999283,0.0017413857,0.0024532017,0.0024935838,0.0046405024],"category_scores_gemma":[0.13955776,0.0008499026,0.0041689733,0.0023683142,0.0028858176,0.0029448927,0.0028714645,0.004008111,0.00025809646],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000759637,0.00039887326,0.033278257,0.00044810245,0.0014999659,0.0011335799,0.0014097664,0.41960317,0.0009320573,0.45831734,0.0020915195,0.0801277],"study_design_scores_gemma":[0.00015818638,0.00015523599,0.003998718,0.00007985326,0.00024337374,0.00010954544,0.00013440849,0.6386197,0.00040428838,0.35425824,0.0017683711,0.000069984766],"about_ca_topic_score_codex":0.013946644,"about_ca_topic_score_gemma":0.009773958,"teacher_disagreement_score":0.0471253,"about_ca_system_score_codex":0.0016723712,"about_ca_system_score_gemma":0.0022822255,"threshold_uncertainty_score":0.24922538},"labels":[],"label_agreement":null},{"id":"W4410503910","doi":"10.1002/sim.70094","title":"Why Recommended Visit Intervals Should Be Extracted When Conducting Longitudinal Analyses Using Electronic Health Record Data: Examining Visit Mechanism and Sensitivity to Assessment Not at Random","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; Public Health Ontario; University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Medicine; Cohort; Health records; Sensitivity (control systems); Population; Longitudinal data; Statistics; Computer science; Data mining; Health care; Mathematics; Environmental health; Internal medicine","score_opus":0.5415146117263445,"score_gpt":0.5572399168944151,"score_spread":0.01572530516807058,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410503910","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1701344,0.00631562,0.8029453,0.00909975,0.001265273,0.002393927,0.0038046911,0.0008764512,0.0031645356],"genre_scores_gemma":[0.73985076,0.00082394225,0.2511161,0.0028502683,0.00047852856,0.0023994548,0.0017629856,0.00026943354,0.00044835775],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.51067215,0.41127136,0.035734,0.021220705,0.019081794,0.0020200245],"domain_scores_gemma":[0.1570743,0.69354224,0.050747596,0.08144065,0.016265804,0.00092938833],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.45066792,0.0016409118,0.002479618,0.0035302173,0.0017447916,0.005719474,0.005169636,0.003958288,0.001045449],"category_scores_gemma":[0.7925964,0.0015887416,0.005201817,0.005700456,0.003502876,0.006481977,0.0034642122,0.004960432,0.00041648338],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024642756,0.00031802972,0.6832936,0.0026036997,0.010773421,0.0007162441,0.010239447,0.03522774,0.0015057579,0.028704355,0.007365149,0.21678832],"study_design_scores_gemma":[0.0008400089,0.0023924746,0.45807317,0.008029595,0.007944088,0.0021889864,0.003920028,0.28446093,0.012965067,0.17743258,0.040728647,0.0010243593],"about_ca_topic_score_codex":0.009125991,"about_ca_topic_score_gemma":0.0048197876,"teacher_disagreement_score":0.5493321,"about_ca_system_score_codex":0.0018144222,"about_ca_system_score_gemma":0.0037132897,"threshold_uncertainty_score":0.6774243},"labels":[],"label_agreement":null},{"id":"W4410608924","doi":"10.1002/sim.70064","title":"Dynamic Single‐Index Scalar‐On‐Function Model","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Air Quality and Health Impacts","field":"Environmental Science","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Institute of Environmental Health Sciences; National Institutes of Health; York University","keywords":"Estimator; Bivariate analysis; Univariate; Spline (mechanical); Parametric statistics; Applied mathematics; Scalar (mathematics); Nonparametric statistics; Computer science; Model selection; Generalized additive model; Parametric model; Tensor product; Econometrics; Mathematics; Statistics; Mathematical optimization; Multivariate statistics","score_opus":0.03100083973489925,"score_gpt":0.3523898610175001,"score_spread":0.32138902128260083,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410608924","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.027167311,0.00035084898,0.966865,0.0005200581,0.00009224938,0.000095524694,0.0005622837,0.00027917832,0.004067556],"genre_scores_gemma":[0.82010514,0.0014720291,0.14817561,0.00045236433,0.00020866696,0.0007656465,0.0013298136,0.00018547705,0.027305229],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9981395,0.00069862645,0.00007080547,0.0005124514,0.00037241494,0.0002061755],"domain_scores_gemma":[0.9955688,0.0027896115,0.00049855484,0.00039722645,0.00058936875,0.00015635957],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004977759,0.001189229,0.0018395659,0.0010871927,0.0004082828,0.0016134076,0.0034764428,0.002041417,0.0050691934],"category_scores_gemma":[0.011104647,0.0006857602,0.0016928723,0.001599877,0.0015165204,0.00213951,0.0016416197,0.0023800523,0.00096909015],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019716453,0.00013436866,0.0059386245,0.00015836423,0.00011321268,0.00030494152,0.0001866581,0.7824228,0.0025544302,0.17014779,0.0018838922,0.035957802],"study_design_scores_gemma":[0.000013486699,0.000048792488,0.0008149383,0.00001061282,0.000028703955,0.000051129162,0.000018115585,0.97493976,0.00026143042,0.022659332,0.0011344377,0.000019304365],"about_ca_topic_score_codex":0.0072174408,"about_ca_topic_score_gemma":0.003933191,"teacher_disagreement_score":0.0072174408,"about_ca_system_score_codex":0.001260404,"about_ca_system_score_gemma":0.0014102249,"threshold_uncertainty_score":0.026325226},"labels":[],"label_agreement":null},{"id":"W4410783369","doi":"10.1002/sim.70121","title":"Integrating Complex Selection Rules Into the Latent Overlapping Group Lasso for the Construction of Coherent Prediction Models","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Université de Montréal","funders":"Fonds de Recherche du Québec - Santé; Canadian Institutes of Health Research; Réseau Québécois de Recherche sur les Médicaments; Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs; Heart and Stroke Foundation of Canada","keywords":"Interpretability; Feature selection; Lasso (programming language); Latent variable; Computer science; Variable (mathematics); Machine learning; Predictive modelling; Selection (genetic algorithm); Elastic net regularization; Proxy (statistics); Artificial intelligence; Data mining; Mathematics","score_opus":0.3549779274483894,"score_gpt":0.5195250836602353,"score_spread":0.1645471562118459,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410783369","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0012461644,0.00003848452,0.99821645,0.00012400902,0.000010481523,0.000048148926,0.00004392539,0.000080325204,0.00019197902],"genre_scores_gemma":[0.07067007,0.00021173916,0.92624444,0.00033494507,0.00011723393,0.00091083074,0.00053527084,0.00022213512,0.00075325323],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.981237,0.012420615,0.0013439143,0.0017255716,0.0028943797,0.00037853306],"domain_scores_gemma":[0.959426,0.032344554,0.0032630619,0.0021908854,0.0023350697,0.0004404304],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.028408881,0.0018135158,0.0018748965,0.0015384209,0.00087285193,0.002825164,0.0019696222,0.0016316413,0.0027824065],"category_scores_gemma":[0.058301393,0.000944931,0.0020672814,0.0015440042,0.0022235976,0.00257873,0.0037219299,0.006071899,0.0010633431],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00014472862,0.00016200058,0.004040856,0.00032845786,0.00026333347,0.00048049158,0.00055561744,0.6704191,0.0037642028,0.2017985,0.0046285056,0.11341422],"study_design_scores_gemma":[0.000026610165,0.000041221312,0.00018854027,0.00004879547,0.000021324604,0.000037885864,0.00002238067,0.921975,0.00078589056,0.075261705,0.0015694902,0.000021065454],"about_ca_topic_score_codex":0.001598025,"about_ca_topic_score_gemma":0.0023709498,"teacher_disagreement_score":0.028408881,"about_ca_system_score_codex":0.00084414874,"about_ca_system_score_gemma":0.002678976,"threshold_uncertainty_score":0.15024233},"labels":[],"label_agreement":null},{"id":"W4411070250","doi":"10.1002/sim.70140","title":"High‐Dimensional Multiresponse Partially Functional Linear Regression","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Sensory Analysis and Statistical Methods","field":"Agricultural and Biological Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"National Key Research and Development Program of China; Natural Sciences and Engineering Research Council of Canada; National Natural Science Foundation of China","keywords":"Functional principal component analysis; Estimator; Curse of dimensionality; Scalar (mathematics); Covariate; Mathematics; Principal component analysis; Applied mathematics; Linear regression; Regression analysis; Regression; Partial least squares regression; Consistency (knowledge bases); Computer science; Statistics; Artificial intelligence","score_opus":0.04955572462124787,"score_gpt":0.3533515474465572,"score_spread":0.30379582282530937,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411070250","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008588723,0.00021001254,0.9901247,0.00023466203,0.00002301879,0.000047881163,0.00019154638,0.00016647251,0.00041296915],"genre_scores_gemma":[0.51237,0.0009634916,0.47839332,0.0006560183,0.00028853165,0.0010390652,0.0015187644,0.00018982138,0.004581032],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99079144,0.006781223,0.00023456871,0.0012319236,0.0006782598,0.00028255503],"domain_scores_gemma":[0.98733366,0.00822283,0.0014680638,0.0016613709,0.0011304498,0.00018355917],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010605181,0.0018010908,0.0018112651,0.0012764983,0.0005504206,0.0013732138,0.0029304996,0.0016318245,0.0021105171],"category_scores_gemma":[0.023256598,0.0008111788,0.0020231658,0.0015238973,0.0016793924,0.0023173885,0.0017964083,0.0017890773,0.00061491865],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00025739704,0.00025469638,0.012311638,0.000531106,0.0007539211,0.00036090036,0.00037450014,0.593987,0.0046869675,0.20819528,0.004291291,0.17399527],"study_design_scores_gemma":[0.000014421557,0.00006933963,0.0017985023,0.000023348923,0.000029977326,0.00006591707,0.00002776095,0.95806223,0.0004297919,0.037845913,0.0015950975,0.00003765205],"about_ca_topic_score_codex":0.0028057725,"about_ca_topic_score_gemma":0.0022740674,"teacher_disagreement_score":0.010605181,"about_ca_system_score_codex":0.00077525433,"about_ca_system_score_gemma":0.0012681541,"threshold_uncertainty_score":0.056086242},"labels":[],"label_agreement":null},{"id":"W4411101265","doi":"10.1002/sim.70135","title":"A Comparison Between Markov Switching Zero‐Inflated and Hurdle Models for Spatio‐Temporal Infectious Disease Counts","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"COVID-19 epidemiological studies","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Natural Sciences and Engineering Research Council of Canada; Institut de Valorisation des Données","keywords":"Negative binomial distribution; Markov chain; Markov model; Econometrics; Zero (linguistics); Statistics; Binomial distribution; Covariate; Count data; Mathematics; Overdispersion; Computer science; Poisson distribution","score_opus":0.1652249539145567,"score_gpt":0.46877656738093065,"score_spread":0.30355161346637394,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411101265","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3026773,0.002391287,0.6849608,0.0023488826,0.00026726682,0.00029936718,0.0011554954,0.0007734544,0.0051261643],"genre_scores_gemma":[0.89188534,0.0016622572,0.097649895,0.00055601966,0.00018565718,0.00048716576,0.0024799623,0.0001892868,0.0049044634],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9904169,0.007061714,0.00034670992,0.00088879454,0.0007819382,0.0005039546],"domain_scores_gemma":[0.91852313,0.07088104,0.0036040593,0.003474426,0.0026013402,0.00091599266],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02244394,0.0010591822,0.0018765328,0.002001797,0.0006094429,0.0019361296,0.0035950986,0.00182694,0.0031456042],"category_scores_gemma":[0.06354185,0.00076221477,0.0029253769,0.0015009629,0.0012799803,0.003075402,0.0014997314,0.0024373217,0.00061943266],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00095677876,0.00016938585,0.016404897,0.00022605908,0.0004341947,0.00022897785,0.0008071475,0.82171065,0.0005642664,0.11763603,0.0024691832,0.038392488],"study_design_scores_gemma":[0.00004775637,0.00016494821,0.0020370726,0.000052603646,0.000055284676,0.000052701216,0.00009321308,0.96257335,0.000111756744,0.03400928,0.00076281757,0.00003918753],"about_ca_topic_score_codex":0.01082148,"about_ca_topic_score_gemma":0.006416072,"teacher_disagreement_score":0.02244394,"about_ca_system_score_codex":0.0017634555,"about_ca_system_score_gemma":0.0019816095,"threshold_uncertainty_score":0.11869627},"labels":[],"label_agreement":null},{"id":"W4411180042","doi":"10.1002/sim.70161","title":"The Impact of Violation of the Proportional Hazards Assumption on the Calibration of the Cox Proportional Hazards Model","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Insurance and Financial Risk Management","field":"Economics, Econometrics and Finance","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Sunnybrook Hospital; University of Toronto","funders":"Canadian Institutes of Health Research","keywords":"Proportional hazards model; Calibration; Statistics; Econometrics; Mathematics","score_opus":0.022430769567123875,"score_gpt":0.290078751311249,"score_spread":0.26764798174412513,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411180042","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2644531,0.0021131292,0.71949494,0.0030817313,0.00027665248,0.00054518256,0.0006215231,0.00048300796,0.008930729],"genre_scores_gemma":[0.95287126,0.00035178228,0.045060385,0.00045307586,0.000046446134,0.00029312394,0.00031166483,0.00011423226,0.00049794535],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.92520845,0.057497464,0.0027726062,0.0046257325,0.008761024,0.0011346921],"domain_scores_gemma":[0.49002922,0.4601096,0.018217811,0.021558015,0.009475396,0.0006100387],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.082047895,0.00080457225,0.0012723184,0.0010404721,0.0011142816,0.0023733939,0.0019356866,0.0017052542,0.0024804908],"category_scores_gemma":[0.35907727,0.00079478644,0.0011501285,0.0016882198,0.0028183064,0.0032515961,0.0023567472,0.004824073,0.00030633682],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011494353,0.00018972538,0.059176087,0.00039266114,0.0006883535,0.00092229055,0.0011132981,0.7903134,0.002167875,0.07223081,0.0027904627,0.06886564],"study_design_scores_gemma":[0.00017939867,0.00047743978,0.018884128,0.00025484274,0.00020489632,0.0010813142,0.00035046026,0.84933865,0.004651147,0.11995291,0.004405058,0.00021976946],"about_ca_topic_score_codex":0.004924118,"about_ca_topic_score_gemma":0.0021950586,"teacher_disagreement_score":0.082047895,"about_ca_system_score_codex":0.0025218537,"about_ca_system_score_gemma":0.0029113975,"threshold_uncertainty_score":0.4339159},"labels":[],"label_agreement":null},{"id":"W4411515984","doi":"10.1002/sim.70176","title":"Precision of Treatment Hierarchy: A Metric for Quantifying Certainty in Treatment Hierarchies From Network Meta‐Analysis","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Actua; University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Deutsche Forschungsgemeinschaft; Hellenic Foundation for Research and Innovation; European Commission","keywords":"Metric (unit); Ranking (information retrieval); Hierarchy; Pairwise comparison; Statistics; Mathematics; Frequentist inference; Rank (graph theory); Certainty; Variance (accounting); Similarity (geometry); Computer science; Bayesian probability; Econometrics; Artificial intelligence; Bayesian inference","score_opus":0.7096301178491243,"score_gpt":0.5685468593222872,"score_spread":0.14108325852683712,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411515984","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014807966,0.028200256,0.94193643,0.002644197,0.0004028456,0.0014199272,0.0062700305,0.0013461619,0.0029721016],"genre_scores_gemma":[0.4304458,0.0076013375,0.5481261,0.0020295833,0.00092836865,0.004379377,0.005322501,0.0006672862,0.0004996928],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7194709,0.21712027,0.023259182,0.018421764,0.020905426,0.00082241284],"domain_scores_gemma":[0.28475326,0.64044076,0.03224007,0.032941338,0.008797179,0.0008274579],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.20882314,0.0036337401,0.008324383,0.020255882,0.0019414529,0.0071260934,0.003993788,0.004323836,0.003773572],"category_scores_gemma":[0.5735497,0.0017855699,0.012936337,0.0151095735,0.0038141732,0.007576298,0.0064675575,0.0066500367,0.0004835526],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002743427,0.00017483144,0.06694666,0.042210065,0.13441056,0.00073244405,0.0023900212,0.21214429,0.0025109306,0.10577599,0.015973011,0.4139877],"study_design_scores_gemma":[0.0015381944,0.0015202812,0.024236957,0.0082569625,0.048965704,0.0012363099,0.00041440348,0.2400609,0.005365844,0.63335633,0.034167286,0.0008808535],"about_ca_topic_score_codex":0.003484739,"about_ca_topic_score_gemma":0.0034687782,"teacher_disagreement_score":0.79117686,"about_ca_system_score_codex":0.004137916,"about_ca_system_score_gemma":0.0039615985,"threshold_uncertainty_score":0.97566193},"labels":[],"label_agreement":null},{"id":"W4412424131","doi":"10.1002/sim.70038","title":"Natural Effects in the Presence of an Intermediate Confounder: Evaluation of Pragmatic Estimation Strategies With an Emphasis on the Relationship Between Natural and Interventional Effects","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre; Université du Québec à Montréal","funders":"Fonds de recherche du Québec – Nature et technologies; Natural Sciences and Engineering Research Council of Canada; Fonds Québécois de la Recherche sur la Nature et les Technologies","keywords":"Estimator; Confounding; Outcome (game theory); Natural (archaeology); Econometrics; Statistics; Mediation; Computer science; Mathematics; Biology","score_opus":0.09314129423726326,"score_gpt":0.47612402867371983,"score_spread":0.3829827344364566,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412424131","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.08856714,0.00092025317,0.9036657,0.0014173038,0.00007627529,0.0010142542,0.00008768062,0.00014934404,0.004102],"genre_scores_gemma":[0.5708397,0.0003974284,0.4263647,0.00028917863,0.000039526763,0.0013123184,0.0000925236,0.000044595803,0.00062003214],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9042717,0.08960688,0.0015508552,0.0014797375,0.0027201956,0.00037056237],"domain_scores_gemma":[0.4376694,0.54525876,0.006186807,0.0066434084,0.0035245263,0.0007170975],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.106420554,0.001148662,0.0016255403,0.0013286973,0.0006980128,0.0024662907,0.0024819295,0.0019230986,0.003534022],"category_scores_gemma":[0.38370517,0.00080096617,0.0012067282,0.0009664301,0.002105513,0.0038389282,0.0039621363,0.002231217,0.00022133977],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0030963763,0.0011759275,0.016124854,0.0027143501,0.0016233029,0.0005484196,0.0040988396,0.19906056,0.0019051977,0.53843564,0.0019037166,0.2293129],"study_design_scores_gemma":[0.0010237728,0.0015677301,0.0040234025,0.0006218353,0.0005427218,0.0002694625,0.00071988837,0.7581848,0.0012234158,0.22800565,0.0036881824,0.00012915174],"about_ca_topic_score_codex":0.002090742,"about_ca_topic_score_gemma":0.0018640184,"teacher_disagreement_score":0.8935794,"about_ca_system_score_codex":0.0012956841,"about_ca_system_score_gemma":0.0028571414,"threshold_uncertainty_score":0.56281245},"labels":[],"label_agreement":null},{"id":"W4412453904","doi":"10.1002/sim.70181","title":"The Design of Large‐Scale Plasma Donation Trials","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Blood donation and transfusion practices","field":"Business, Management and Accounting","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University Medical Centre; Actua; McMaster University; University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; University of Waterloo; McMaster University","keywords":"Outcome (game theory); Variance (accounting); Donation; Poisson distribution; Adverse effect; Statistics; Medicine; Audit; Scale (ratio); Accrual; Econometrics; Computer science; Mathematics; Internal medicine; Economics","score_opus":0.04330349544810271,"score_gpt":0.342392463399069,"score_spread":0.2990889679509663,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412453904","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.028983954,0.0046333694,0.6289432,0.003445913,0.0025703777,0.32411766,0.0013067209,0.00078428804,0.005214516],"genre_scores_gemma":[0.14784403,0.0018315805,0.32038322,0.001720884,0.0005123585,0.52555454,0.0005250524,0.00008718699,0.0015411979],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7186759,0.25663736,0.010222775,0.0062479996,0.007052353,0.0011637192],"domain_scores_gemma":[0.8546479,0.109550074,0.014034981,0.015652964,0.0039111353,0.0022029548],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.21048735,0.002606462,0.00616271,0.0019390797,0.0011003242,0.0039627235,0.0028197637,0.0042206533,0.0071403915],"category_scores_gemma":[0.27488595,0.0017681387,0.002885236,0.002253338,0.002753813,0.0030036801,0.0020846375,0.004803961,0.0016933784],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.21682152,0.0074793943,0.013434046,0.022305967,0.014802303,0.0010426904,0.0012885612,0.11111245,0.00814942,0.16517706,0.014851752,0.42353478],"study_design_scores_gemma":[0.21650559,0.08850117,0.011664583,0.003144245,0.008100728,0.0006767159,0.0004655995,0.24987884,0.008577404,0.3528077,0.0590947,0.0005827017],"about_ca_topic_score_codex":0.00028520235,"about_ca_topic_score_gemma":0.00043094534,"teacher_disagreement_score":0.21048735,"about_ca_system_score_codex":0.0016195894,"about_ca_system_score_gemma":0.005055852,"threshold_uncertainty_score":0.9736096},"labels":[],"label_agreement":null},{"id":"W4412453910","doi":"10.1002/sim.70186","title":"A Latent‐Class Model for Time‐To‐Event Outcomes and High‐Dimensional Imaging Data","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Bivariate analysis; Covariate; Computer science; Logistic regression; Neuroimaging; Expectation–maximization algorithm; Class (philosophy); Latent class model; Artificial intelligence; Machine learning; Proportional hazards model; Identification (biology); Disease; Data mining; Econometrics; Statistics; Medicine; Mathematics; Maximum likelihood; Internal medicine","score_opus":0.1053803383116654,"score_gpt":0.4526441691017764,"score_spread":0.347263830790111,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412453910","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02010593,0.00061025337,0.97269905,0.0023655267,0.00020142375,0.0002626913,0.0019441367,0.0003916281,0.0014194087],"genre_scores_gemma":[0.6774606,0.00252271,0.28517509,0.001184317,0.00095172535,0.0044571473,0.0061015775,0.0002973501,0.021849446],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9903955,0.0056818295,0.00037224716,0.0019604752,0.0008933105,0.0006966697],"domain_scores_gemma":[0.9690194,0.024779772,0.0022367716,0.0019240461,0.0014594467,0.0005805937],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020367749,0.0016524603,0.0026619227,0.0026327067,0.0012767646,0.0034370094,0.006860987,0.0038874296,0.008700092],"category_scores_gemma":[0.033862982,0.0010192153,0.0028707394,0.0032167807,0.003166322,0.004846679,0.0028641908,0.005386003,0.0022591194],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005997234,0.00039798292,0.015188613,0.00043184572,0.00051972887,0.0006946366,0.0010721183,0.37277874,0.0008951205,0.53669786,0.008510754,0.062212925],"study_design_scores_gemma":[0.00014271078,0.0001393464,0.0021565594,0.000079618214,0.00010027732,0.00020769029,0.00009773139,0.8216314,0.00016603802,0.17182119,0.0033895378,0.000067901456],"about_ca_topic_score_codex":0.008448295,"about_ca_topic_score_gemma":0.00755948,"teacher_disagreement_score":0.020367749,"about_ca_system_score_codex":0.002107366,"about_ca_system_score_gemma":0.0027948963,"threshold_uncertainty_score":0.10771626},"labels":[],"label_agreement":null},{"id":"W4412527389","doi":"10.1002/sim.70194","title":"A Simple Diagnostic for the Positivity Assumption for Continuous Exposures","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"HEC Montréal; McGill University","funders":"Fonds de Recherche du Québec - Santé; Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Simple (philosophy); Statistics; Computer science; Econometrics; Mathematics","score_opus":0.0848985461080552,"score_gpt":0.46413627148375325,"score_spread":0.37923772537569805,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412527389","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.022857497,0.0005918221,0.9625274,0.0057846056,0.00046522054,0.0004978661,0.0008957758,0.0004535706,0.0059262663],"genre_scores_gemma":[0.60213,0.00052645395,0.38665465,0.0039347843,0.0010211895,0.0018997565,0.0011027737,0.00014182388,0.0025885394],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96773595,0.019558685,0.001753852,0.0054277345,0.004670186,0.00085359596],"domain_scores_gemma":[0.6065036,0.34535494,0.019755943,0.017365534,0.0087665785,0.0022534016],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.055581428,0.0012574603,0.0022364182,0.004457298,0.0017244009,0.0036548108,0.004021943,0.0067104427,0.016414968],"category_scores_gemma":[0.4067241,0.0007217863,0.0020845404,0.0029171337,0.007955331,0.007727457,0.0050481344,0.008399821,0.0018409936],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011959146,0.000308594,0.026652467,0.00094501564,0.0002915812,0.0009760086,0.0013354928,0.0116038,0.0028433255,0.8188689,0.010777776,0.124201275],"study_design_scores_gemma":[0.0003073546,0.0004711087,0.0054967264,0.00036603629,0.00012254472,0.0015758113,0.00023825544,0.09461856,0.0020524329,0.8871074,0.0075383764,0.00010536337],"about_ca_topic_score_codex":0.0007477104,"about_ca_topic_score_gemma":0.00039649638,"teacher_disagreement_score":0.055581428,"about_ca_system_score_codex":0.0015142484,"about_ca_system_score_gemma":0.0035572234,"threshold_uncertainty_score":0.29394627},"labels":[],"label_agreement":null},{"id":"W4412582858","doi":"10.1002/sim.70188","title":"The Mathematics of Serocatalytic Models With Applications to Public Health Data","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"COVID-19 epidemiological studies","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Institute for Health Research Health Protection Research Unit; National Institute for Health and Care Research; Natural Environment Research Council; International Development Research Centre; University of Oxford; Merton College, University of Oxford","keywords":"Intuition; Computer science; Data science; Epidemiology; Infectious disease (medical specialty); Context (archaeology); Public health; Disease; Management science; Medicine; Biology; Epistemology; Pathology","score_opus":0.4394107515544346,"score_gpt":0.5164224802020916,"score_spread":0.07701172864765699,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412582858","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00724656,0.0030474975,0.9725903,0.00875501,0.00024973977,0.00007727647,0.0008433506,0.00024308945,0.0069472003],"genre_scores_gemma":[0.46381125,0.01858381,0.4843473,0.0037744031,0.002420239,0.0012226185,0.0029020854,0.00059013476,0.022348199],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99566287,0.0026833855,0.00027083376,0.0004902184,0.0007037678,0.00018896413],"domain_scores_gemma":[0.9500028,0.042231202,0.0031132963,0.002083511,0.0017911426,0.0007780209],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014087768,0.0014365879,0.0013849977,0.0031414006,0.00088547164,0.0038563618,0.002341873,0.0023456507,0.003934654],"category_scores_gemma":[0.059257336,0.0009724198,0.0026160893,0.0025813389,0.0035763902,0.005492501,0.004106989,0.007449515,0.0010272998],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000017099037,0.00002537136,0.0013439939,0.00013494362,0.0000535278,0.0001242385,0.00024307906,0.11978754,0.00015582464,0.8611178,0.006220637,0.01077607],"study_design_scores_gemma":[0.000008211891,0.00001140814,0.00018708546,0.00005593013,0.00001120828,0.00006180377,0.000036306963,0.2995565,0.00008409641,0.69500583,0.004966598,0.0000149246525],"about_ca_topic_score_codex":0.0059180446,"about_ca_topic_score_gemma":0.0045286682,"teacher_disagreement_score":0.014087768,"about_ca_system_score_codex":0.0036339746,"about_ca_system_score_gemma":0.002359187,"threshold_uncertainty_score":0.07450414},"labels":[],"label_agreement":null},{"id":"W4412654453","doi":"10.1002/sim.70198","title":"Fast Variational Bayesian Inference for Correlated Survival Data: An Application to Invasive Mechanical Ventilation Duration Analysis","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Hemodynamic Monitoring and Therapy","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"The King's University; Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Mechanical ventilation; Duration (music); Bayesian probability; Bayesian inference; Inference; Survival analysis; Statistics; Computer science; Econometrics; Mathematics; Artificial intelligence; Medicine; Internal medicine","score_opus":0.027712925675312767,"score_gpt":0.3847432451131363,"score_spread":0.35703031943782354,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412654453","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0075699557,0.00025161554,0.9912918,0.00022618557,0.000021393394,0.00004602264,0.00009241916,0.00022066802,0.000279963],"genre_scores_gemma":[0.2582131,0.0004697402,0.73738843,0.00032169026,0.00009182941,0.0002923156,0.0008298194,0.00028132784,0.002111854],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99812764,0.0012332818,0.00007566462,0.00020968399,0.00025284034,0.000100843405],"domain_scores_gemma":[0.98006254,0.017568989,0.00050473446,0.00061656476,0.00096835376,0.00027891807],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007529995,0.00072703994,0.0012118507,0.0010186101,0.00077588175,0.0011171008,0.002681789,0.0016630857,0.0022659127],"category_scores_gemma":[0.031052597,0.0009800129,0.0011702758,0.0013456971,0.0010740669,0.0011427657,0.0018751518,0.0022774497,0.00032530402],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00011978892,0.000045118,0.0038606783,0.00011381282,0.00017421957,0.00013482649,0.00016781269,0.8923234,0.0009866267,0.03599334,0.0016325711,0.06444782],"study_design_scores_gemma":[0.000018655412,0.000006531196,0.00023433554,0.000009439607,0.0000055155547,0.000017415077,0.000009771647,0.9886783,0.00011654642,0.010514232,0.0003827094,0.000006567572],"about_ca_topic_score_codex":0.05627276,"about_ca_topic_score_gemma":0.05650409,"teacher_disagreement_score":0.05627276,"about_ca_system_score_codex":0.0018303667,"about_ca_system_score_gemma":0.004128438,"threshold_uncertainty_score":0.111890376},"labels":[],"label_agreement":null},{"id":"W4413048133","doi":"10.1002/sim.70215","title":"Health Utility Survival for Randomized Clinical Trials: Extensions and Statistical Properties","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Princess Margaret Cancer Centre; Public Health Ontario; University of Toronto; University Health Network","funders":"Fundamental Research Funds for the Central Universities; National Natural Science Foundation of China; Princess Margaret Cancer Foundation","keywords":"Clinical endpoint; Sample size determination; Statistics; Survival analysis; Randomized controlled trial; Medicine; Clinical trial; Statistical hypothesis testing; Statistical power; Econometrics; Computer science; Mathematics; Surgery; Internal medicine","score_opus":0.6863747557470252,"score_gpt":0.6045958885801116,"score_spread":0.08177886716691363,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413048133","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0045719203,0.0055484613,0.9795299,0.0027551576,0.00042621852,0.003727717,0.0005251312,0.0005972788,0.0023182053],"genre_scores_gemma":[0.17106615,0.0059933155,0.7918284,0.003060899,0.0009707546,0.023303,0.0012031605,0.0004544046,0.002119973],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5703445,0.40147936,0.010174239,0.0069528073,0.00991396,0.0011350979],"domain_scores_gemma":[0.268223,0.6730747,0.020574272,0.027998645,0.008929223,0.0012002166],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.4925276,0.004056591,0.006760479,0.0066134506,0.0008837289,0.0066531496,0.0043701613,0.0059877927,0.008738364],"category_scores_gemma":[0.6532767,0.0022224835,0.007849243,0.008614448,0.0070349607,0.010206217,0.006351323,0.012001079,0.0018225835],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0042843376,0.00029871357,0.013668815,0.00798706,0.006838558,0.0008020029,0.0015903104,0.12102861,0.0005405668,0.5897191,0.010303546,0.24293849],"study_design_scores_gemma":[0.001642656,0.0015973119,0.0025667148,0.002505896,0.0015579483,0.0007363499,0.00015762467,0.38049346,0.00058714295,0.5919716,0.016011946,0.00017123087],"about_ca_topic_score_codex":0.0011491616,"about_ca_topic_score_gemma":0.00073599146,"teacher_disagreement_score":0.4925276,"about_ca_system_score_codex":0.0035736219,"about_ca_system_score_gemma":0.0054251445,"threshold_uncertainty_score":0.6258038},"labels":[],"label_agreement":null},{"id":"W4413111174","doi":"10.1002/sim.70227","title":"A Framework for Generating Realistic Synthetic Tabular Data in a Randomized Controlled Trial Setting","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Imbalanced Data Classification Techniques","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Randomized controlled trial; Synthetic data; Copula (linguistics); Artificial intelligence; Machine learning; Data mining; Econometrics; Medicine; Mathematics","score_opus":0.03481874690734359,"score_gpt":0.3819513484906666,"score_spread":0.347132601583323,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413111174","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0024495767,0.00018740252,0.9951432,0.00050730444,0.00006561792,0.00046414844,0.0003357224,0.0003061946,0.00054083136],"genre_scores_gemma":[0.11133204,0.00036155383,0.8830027,0.0005683344,0.00012116612,0.0029135724,0.0007223546,0.00013078538,0.000847471],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9707568,0.02538317,0.000865755,0.0012079527,0.0014631271,0.00032326076],"domain_scores_gemma":[0.84279895,0.13898267,0.0061513293,0.007285508,0.0036527538,0.0011287556],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.056126133,0.0010975532,0.001425763,0.0018290675,0.00048139886,0.0019408236,0.0029301404,0.0023573304,0.007589698],"category_scores_gemma":[0.12641408,0.00087327574,0.0021589212,0.0015003406,0.0017475238,0.001797923,0.0021452755,0.0032678328,0.0012730047],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000817917,0.00023297366,0.0031654981,0.0007356134,0.00027996764,0.00030175885,0.00057450624,0.58659214,0.0011842182,0.29386336,0.0070358426,0.105216175],"study_design_scores_gemma":[0.0003863539,0.00031637977,0.00030751806,0.00015450236,0.000055896082,0.00011073241,0.000042920983,0.84968495,0.00063970836,0.1432136,0.005047573,0.00003986884],"about_ca_topic_score_codex":0.0017178499,"about_ca_topic_score_gemma":0.0017037101,"teacher_disagreement_score":0.056126133,"about_ca_system_score_codex":0.0012876249,"about_ca_system_score_gemma":0.0037027732,"threshold_uncertainty_score":0.2968269},"labels":[],"label_agreement":null},{"id":"W4413140657","doi":"10.1002/sim.70222","title":"An Efficient Two‐Dimensional Functional Mixed‐Effect Model Framework for Repeatedly Measured Functional Data","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Neural Networks and Applications","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"City University of Hong Kong","keywords":"Computer science; Mixed model; Functional data analysis; Econometrics; Mathematics; Machine learning","score_opus":0.054495888216838366,"score_gpt":0.35451119008282744,"score_spread":0.30001530186598907,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413140657","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0019983796,0.00030639247,0.9964646,0.0002668114,0.00004661764,0.00006446887,0.00042004042,0.00018213916,0.00025063634],"genre_scores_gemma":[0.15040714,0.0014044467,0.83606976,0.0004885535,0.0003277818,0.0024366542,0.0029974803,0.00029471246,0.0055735265],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99258375,0.0054467996,0.00030325382,0.0009222398,0.0004635593,0.00028032865],"domain_scores_gemma":[0.98737514,0.010172973,0.0005904401,0.0007993479,0.0008582484,0.00020383146],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016613856,0.0023297288,0.0026496085,0.0020141446,0.00072836864,0.0022779384,0.005237736,0.0028088787,0.006542246],"category_scores_gemma":[0.026816403,0.0017179316,0.0042230305,0.0027724192,0.0014632771,0.0022116988,0.0025884903,0.0040513757,0.0016022689],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002572959,0.000157885,0.0051433058,0.00047320512,0.0008860242,0.00046134755,0.00031420513,0.7002499,0.0012146985,0.2025516,0.004707107,0.08358341],"study_design_scores_gemma":[0.00003303413,0.00006835783,0.0005472119,0.000037377817,0.00008351608,0.00005880541,0.000027722477,0.9479403,0.0001630197,0.048322994,0.0026834696,0.00003420169],"about_ca_topic_score_codex":0.016390942,"about_ca_topic_score_gemma":0.019628398,"teacher_disagreement_score":0.016613856,"about_ca_system_score_codex":0.0016039326,"about_ca_system_score_gemma":0.0036356205,"threshold_uncertainty_score":0.087863505},"labels":[],"label_agreement":null},{"id":"W4414168825","doi":"10.1002/sim.70201","title":"Weighted Trigonometric Regression for Suboptimal Designs in Circadian Transcriptome Studies","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Circadian rhythm and melatonin","field":"Neuroscience","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Estimator; Sample size determination; Statistical hypothesis testing; Regression; Regression analysis; Kernel (algebra); Trigonometry; Statistical power; Kernel regression","score_opus":0.08897076563611843,"score_gpt":0.381796386422942,"score_spread":0.29282562078682356,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414168825","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0055612708,0.00019642076,0.9933468,0.00010036259,0.000031211464,0.00013106,0.00004265279,0.00021554032,0.0003747296],"genre_scores_gemma":[0.14900072,0.0003347884,0.8468222,0.00027136112,0.00006028847,0.0011483016,0.00033375315,0.000311272,0.0017173882],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96700704,0.02600323,0.0009956555,0.0029995798,0.0024529428,0.00054154074],"domain_scores_gemma":[0.911676,0.07274287,0.004747051,0.0054438133,0.00470842,0.0006817651],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.050919592,0.002448229,0.0026360843,0.0016280775,0.00088330335,0.0016325224,0.0028130717,0.0020449024,0.0044095963],"category_scores_gemma":[0.11666818,0.0010600726,0.0017819072,0.0020020467,0.002634643,0.0028045573,0.0027747417,0.003240023,0.0013772832],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013055007,0.00036471436,0.0054575424,0.0010055029,0.00038564848,0.00037101886,0.00046018595,0.59764063,0.010146986,0.16377954,0.0031289407,0.21595375],"study_design_scores_gemma":[0.00012679349,0.0005578238,0.001287641,0.00010718524,0.000074114876,0.000066495224,0.000064137406,0.9284643,0.0036915676,0.06267721,0.0028218944,0.00006084076],"about_ca_topic_score_codex":0.0027672986,"about_ca_topic_score_gemma":0.0023205457,"teacher_disagreement_score":0.050919592,"about_ca_system_score_codex":0.0019877572,"about_ca_system_score_gemma":0.0040212157,"threshold_uncertainty_score":0.26929176},"labels":[],"label_agreement":null},{"id":"W4414196550","doi":"10.1002/sim.70258","title":"Collapsible Kernel Machine Regression for Exposomic Analyses","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Health, Environment, Cognitive Aging","field":"Environmental Science","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"National Institute of Environmental Health Sciences; Natural Sciences and Engineering Research Council of Canada; National Institutes of Health","keywords":"Additive model; Kernel (algebra); Causal inference; Bayesian probability; Flexibility (engineering); Covariate; Kernel method; Inference; Bayesian inference; Regression","score_opus":0.04023429796767516,"score_gpt":0.4032869001057561,"score_spread":0.36305260213808094,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414196550","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0030948631,0.00015967882,0.99603564,0.00009113518,0.0000157151,0.000033195494,0.00008980855,0.000280699,0.00019933791],"genre_scores_gemma":[0.1929936,0.0005361098,0.80004066,0.00029026865,0.0001799113,0.00075644423,0.0010702208,0.0004577616,0.0036749833],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99001884,0.00644094,0.00040806117,0.0015425172,0.0012876783,0.00030185413],"domain_scores_gemma":[0.96880776,0.021252623,0.002258161,0.0058389003,0.0014981721,0.00034433106],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.015408441,0.0014485616,0.0022808588,0.0022030252,0.00084241066,0.0017067597,0.0029280763,0.0016161251,0.0037147808],"category_scores_gemma":[0.05271569,0.0009475386,0.0021795202,0.0035774168,0.0016999155,0.002165222,0.004159055,0.0032952183,0.0013588552],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00037852267,0.00026082183,0.009579847,0.0004681425,0.00082440674,0.00059986796,0.00040815494,0.4626169,0.0035577726,0.24651253,0.0054613343,0.26933166],"study_design_scores_gemma":[0.000024373694,0.00005565351,0.0010422874,0.000029787501,0.000037857648,0.00006755392,0.00002648183,0.858668,0.0006182957,0.13656664,0.0028285214,0.000034451048],"about_ca_topic_score_codex":0.0045033493,"about_ca_topic_score_gemma":0.003750887,"teacher_disagreement_score":0.015408441,"about_ca_system_score_codex":0.0010548241,"about_ca_system_score_gemma":0.0021740412,"threshold_uncertainty_score":0.08148861},"labels":[],"label_agreement":null},{"id":"W4414264132","doi":"10.1002/sim.70234","title":"What Is Fair? Defining Fairness in Machine Learning for Health","year":2025,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Women's and Gender Studies et Recherches Féministes; University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada; University of Toronto","keywords":"Operationalization; Fairness measure; Capability approach; MEDLINE; Health care","score_opus":0.2301952076018682,"score_gpt":0.5545321101943922,"score_spread":0.324336902592524,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414264132","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00022131817,0.98187363,0.0035381662,0.010434391,0.0006226545,0.000013752164,0.000022180468,0.000008113104,0.003265834],"genre_scores_gemma":[0.018975236,0.96784323,0.0036072822,0.006461752,0.0022637867,0.00009217853,0.000042860705,0.000017396807,0.0006962807],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9908915,0.005774614,0.00071112195,0.00072026654,0.0016278474,0.00027462328],"domain_scores_gemma":[0.9726041,0.024433054,0.0009820245,0.00062127237,0.0011101497,0.00024941293],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.015281214,0.00094205746,0.0020576515,0.0028361494,0.0011764028,0.0044810167,0.0016976189,0.0044855834,0.0027045682],"category_scores_gemma":[0.031531915,0.00042250264,0.0009709691,0.0030952678,0.008373649,0.007166913,0.0024506138,0.005661352,0.0006442708],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00005613499,0.00004207349,0.00038461198,0.010831526,0.00019152713,0.000068131085,0.00068343256,0.0018523012,0.00013837517,0.53423285,0.015245473,0.43627355],"study_design_scores_gemma":[0.000026714017,0.00008028942,0.0009179494,0.027924608,0.00014583085,0.00034578503,0.0005246382,0.0011235963,0.00032139395,0.58472174,0.38379294,0.00007449433],"about_ca_topic_score_codex":0.0033552323,"about_ca_topic_score_gemma":0.0030352008,"teacher_disagreement_score":0.9847188,"about_ca_system_score_codex":0.0040196558,"about_ca_system_score_gemma":0.006993595,"threshold_uncertainty_score":0.08081573},"labels":[],"label_agreement":null},{"id":"W4414283105","doi":"10.1002/sim.70248","title":"The Estimand Framework in Diagnostic Accuracy Studies","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Deutsche Forschungsgemeinschaft","keywords":"Diagnostic accuracy; Test (biology); Diagnostic test; Bridge (graph theory); Interpretation (philosophy)","score_opus":0.6242570362789946,"score_gpt":0.6233312332906149,"score_spread":0.0009258029883797425,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414283105","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0016309967,0.01394566,0.9742231,0.004858778,0.0003790328,0.0006117469,0.00060993317,0.00025554976,0.0034852116],"genre_scores_gemma":[0.13769898,0.01382248,0.8306328,0.0039490284,0.0017432041,0.008505345,0.0010246312,0.00028198486,0.002341561],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.580865,0.3886494,0.00850451,0.010052578,0.010437567,0.0014907847],"domain_scores_gemma":[0.32987985,0.64051056,0.010538401,0.013615943,0.0047579133,0.00069728476],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27742422,0.0050786324,0.008925077,0.010454356,0.0014859232,0.007999614,0.007074516,0.008290183,0.006447696],"category_scores_gemma":[0.43636647,0.0027965673,0.00804563,0.009412363,0.00964957,0.009861521,0.006450403,0.009957385,0.0013781511],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00042638826,0.000096918804,0.004088061,0.0038325535,0.003285457,0.00057592505,0.0007133166,0.031952176,0.00018808046,0.8881668,0.0032509263,0.063423485],"study_design_scores_gemma":[0.00031356778,0.00026374427,0.00067663373,0.0011598471,0.0010476476,0.00021468634,0.00011708337,0.054038513,0.0003441201,0.93111056,0.010626194,0.00008736526],"about_ca_topic_score_codex":0.0067355977,"about_ca_topic_score_gemma":0.0027895481,"teacher_disagreement_score":0.7225758,"about_ca_system_score_codex":0.005982104,"about_ca_system_score_gemma":0.0077994694,"threshold_uncertainty_score":0.8910646},"labels":[],"label_agreement":null},{"id":"W4414484180","doi":"10.1002/sim.70273","title":"Designing Stepped Wedge Cluster Randomized Trials With a Baseline Measurement of the Outcome","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"National Center for Advancing Translational Sciences; National Institute on Aging; Patient-Centered Outcomes Research Institute","keywords":"Baseline (sea); Sample size determination; Outcome (game theory); Estimator; Randomized controlled trial; Context (archaeology); Cluster randomised controlled trial; Cluster (spacecraft)","score_opus":0.2478030119876625,"score_gpt":0.4705801716423711,"score_spread":0.2227771596547086,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414484180","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011705442,0.00042822873,0.97915524,0.0005805959,0.00014519965,0.006178532,0.00024644658,0.00027452913,0.0012858954],"genre_scores_gemma":[0.1121521,0.00037186808,0.8727492,0.0005407425,0.00005466412,0.013375664,0.00018330461,0.00005244408,0.00051997683],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9231418,0.06766854,0.0022663618,0.003366899,0.0030473468,0.00050911657],"domain_scores_gemma":[0.84852636,0.12908404,0.0079436945,0.009674435,0.00406111,0.0007104487],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10018456,0.0013186949,0.002590712,0.0015991483,0.00052577374,0.0015813516,0.002975403,0.0019426008,0.004615194],"category_scores_gemma":[0.2103621,0.0010297553,0.002267074,0.0017826409,0.0019873003,0.002575792,0.0020294634,0.002689127,0.00070742593],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008114084,0.0010447359,0.010434551,0.0071568163,0.0027829576,0.00053408096,0.0011884357,0.17724885,0.0030048375,0.46633452,0.0075617977,0.3145943],"study_design_scores_gemma":[0.008510923,0.0050173374,0.0022041404,0.0012130077,0.0013394413,0.00028161402,0.0003334232,0.44147682,0.00598157,0.5212519,0.012218419,0.00017140812],"about_ca_topic_score_codex":0.00059270236,"about_ca_topic_score_gemma":0.00064557913,"teacher_disagreement_score":0.89981544,"about_ca_system_score_codex":0.0010409475,"about_ca_system_score_gemma":0.004340224,"threshold_uncertainty_score":0.52983296},"labels":[],"label_agreement":null},{"id":"W4415153955","doi":"10.1002/sim.70269","title":"Specification of Estimands for Complex Disease Processes Using Multistate Models and Utility Functions","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Interpretability; Jackknife resampling; Ranking (information retrieval); Disease; Simplicity; Variance (accounting); Resampling; Clinical trial","score_opus":0.7557390693740194,"score_gpt":0.6269280900231367,"score_spread":0.1288109793508827,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4415153955","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009505335,0.00047550752,0.9879972,0.00080974074,0.000033039385,0.00019271878,0.0002557457,0.000108678534,0.00062212674],"genre_scores_gemma":[0.56118244,0.0025082887,0.4266909,0.0007732066,0.00021559822,0.0032689255,0.0014120082,0.00016362571,0.0037850197],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9710675,0.024934394,0.0008380593,0.0012954178,0.0011778371,0.0006868174],"domain_scores_gemma":[0.83010614,0.15841472,0.005359966,0.003778886,0.001760086,0.00058025063],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06095469,0.002016053,0.003377179,0.0028180105,0.0005175565,0.0039904793,0.0022982857,0.0024957708,0.003046621],"category_scores_gemma":[0.12483749,0.0010931732,0.0031501723,0.002290722,0.002645534,0.0034771452,0.0026818516,0.005292518,0.000468523],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00027992463,0.00013590867,0.0062237424,0.0003816402,0.0004949355,0.00027782872,0.00031608637,0.54133856,0.0003361795,0.41716295,0.0015172103,0.031534947],"study_design_scores_gemma":[0.00006471457,0.00014161652,0.00091612077,0.000110824665,0.00008396801,0.00006490995,0.000062490784,0.7212766,0.00028869387,0.2758564,0.0010950621,0.000038633738],"about_ca_topic_score_codex":0.003338509,"about_ca_topic_score_gemma":0.0024778305,"teacher_disagreement_score":0.06095469,"about_ca_system_score_codex":0.0019337586,"about_ca_system_score_gemma":0.0031517043,"threshold_uncertainty_score":0.32236308},"labels":[],"label_agreement":null},{"id":"W4415996186","doi":"10.1002/sim.70309","title":"Identification of Regions of Interest in Neuroimaging Data With Irregular Boundary Based on Semiparametric Transformation Models and Interval‐Censored Outcomes","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Bivariate analysis; Covariate; Identification (biology); Neuroimaging; Nonparametric statistics; Boundary (topology); Domain (mathematical analysis); Transformation (genetics); Pattern recognition (psychology)","score_opus":0.18901742399715157,"score_gpt":0.4306194415401282,"score_spread":0.24160201754297664,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4415996186","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09851433,0.0006396035,0.89871985,0.0006130928,0.000024561428,0.00012781973,0.00057423016,0.0002994113,0.00048710516],"genre_scores_gemma":[0.76374286,0.0006253979,0.23160444,0.00016056937,0.000070491784,0.0004000268,0.0020688905,0.000109173474,0.0012181301],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9975823,0.0017052725,0.00008639745,0.00032512468,0.00017775124,0.00012300827],"domain_scores_gemma":[0.9740381,0.02186499,0.00198964,0.0012341967,0.0005247587,0.00034839186],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009353345,0.0006722183,0.0013239603,0.0016683224,0.00041115025,0.0015357703,0.0017173265,0.0012869819,0.0009860843],"category_scores_gemma":[0.028164236,0.00043506033,0.001797539,0.0013950571,0.0015635466,0.001317658,0.0023875977,0.0020491818,0.00025591615],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00070826826,0.00016140708,0.018871134,0.00027908516,0.00015685326,0.00065776065,0.000401629,0.86527175,0.0023916196,0.039389517,0.002290971,0.06941994],"study_design_scores_gemma":[0.000024477373,0.000036929712,0.0017981686,0.000020468138,0.000012993329,0.000098286895,0.000042305735,0.9768903,0.00043914907,0.020180985,0.00043949723,0.000016468284],"about_ca_topic_score_codex":0.0032522485,"about_ca_topic_score_gemma":0.0033903278,"teacher_disagreement_score":0.009353345,"about_ca_system_score_codex":0.00079486327,"about_ca_system_score_gemma":0.001127057,"threshold_uncertainty_score":0.049465775},"labels":[],"label_agreement":null},{"id":"W4416943753","doi":"10.1002/sim.70336","title":"System of Linear Equations to Derive Unreported Test Accuracy Counts for Meta‐Analysis","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto; Cancer Care Ontario","funders":"","keywords":"Rounding; Sensitivity (control systems); Test (biology); Linear system; Linear model; Accuracy and precision; Linear equation; Limit (mathematics)","score_opus":0.544600576239449,"score_gpt":0.5683193730780832,"score_spread":0.023718796838634204,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4416943753","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0017135531,0.0022365523,0.9890198,0.0009247607,0.00024426982,0.0010742205,0.0025019967,0.00078285177,0.0015019713],"genre_scores_gemma":[0.03553532,0.004051357,0.9412824,0.0009226462,0.00026995063,0.010019334,0.0029493342,0.0002803003,0.004689364],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9708981,0.021325395,0.00219499,0.0023652047,0.002888779,0.0003274992],"domain_scores_gemma":[0.92577064,0.067110635,0.0024217835,0.0017452952,0.0028282024,0.00012345801],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04329957,0.003244051,0.004512645,0.0052628214,0.00059765595,0.0030775797,0.003332207,0.0021740547,0.020889292],"category_scores_gemma":[0.111292325,0.0021617583,0.008919671,0.0058807484,0.00091782055,0.0022335346,0.003098994,0.005160857,0.005255657],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000553384,0.00023289052,0.008943725,0.010850673,0.0139497975,0.00087376975,0.0009655709,0.3506899,0.0019637952,0.27557987,0.04234483,0.2930518],"study_design_scores_gemma":[0.0006199672,0.00046267724,0.0032159134,0.0019699577,0.004762624,0.0006284146,0.00016441068,0.70429635,0.0015733943,0.22983603,0.052238274,0.00023198518],"about_ca_topic_score_codex":0.010275714,"about_ca_topic_score_gemma":0.010974602,"teacher_disagreement_score":0.95670044,"about_ca_system_score_codex":0.0027497737,"about_ca_system_score_gemma":0.0056354147,"threshold_uncertainty_score":0.22899276},"labels":[],"label_agreement":null},{"id":"W4417070771","doi":"10.1002/sim.70327","title":"Likelihood‐Based Non‐Parametric Receiver Operating Characteristic Curve Analysis in the Presence of Imperfect Reference Standard","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Clinical Laboratory Practices and Quality Control","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Receiver operating characteristic; Imperfect; Gold standard (test); Index (typography); Standard error; Mean squared error","score_opus":0.042036275745086636,"score_gpt":0.41713028637182276,"score_spread":0.37509401062673614,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4417070771","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0069213896,0.00052426563,0.99155736,0.000247313,0.000015112139,0.000052687923,0.000046633355,0.0003638757,0.00027123402],"genre_scores_gemma":[0.3891897,0.0006188683,0.6079014,0.00038964572,0.00008729452,0.0004837734,0.0003407431,0.00022230316,0.0007662656],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9758798,0.01805838,0.0010618338,0.002185131,0.002508025,0.00030688226],"domain_scores_gemma":[0.91844106,0.06700391,0.006126631,0.0050753397,0.0030467594,0.00030620184],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03559111,0.0011050022,0.002482149,0.0031657517,0.00060548447,0.00244852,0.003207085,0.0021965746,0.0008645514],"category_scores_gemma":[0.12606336,0.000807267,0.0017044287,0.0029756608,0.0023932243,0.002656644,0.0024263875,0.0022826153,0.00057861983],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00097515725,0.00029733198,0.047356106,0.0016787439,0.0013937239,0.0015096344,0.0015601445,0.26717198,0.009854395,0.06660129,0.0048744124,0.59672713],"study_design_scores_gemma":[0.000099123565,0.00024918639,0.008377582,0.00012515404,0.0001830936,0.0011507388,0.0000985565,0.9297495,0.00537648,0.050114214,0.004340499,0.00013591239],"about_ca_topic_score_codex":0.0021909352,"about_ca_topic_score_gemma":0.0015228188,"teacher_disagreement_score":0.03559111,"about_ca_system_score_codex":0.0012163689,"about_ca_system_score_gemma":0.0021808343,"threshold_uncertainty_score":0.18822604},"labels":[],"label_agreement":null}]}