{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":708,"total_is_capped":false,"direct_labels_cover":5,"predictions_cover":708,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12","author_layer_release":"2026-06-26"},"query_hash":"e941b4fc72db","filters":{"venue":"Statistics in Medicine"}},"results":[{"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2114497428934903,"gpt":0.4202548380162484,"spread":0.208805095122758,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09158804,0.001541054,0.002624626,0.01080528,0.00119458,0.002779097,0.003842496,0.003430333,0.01373318],"category_scores_gemma":[0.3592003,0.001031592,0.002396284,0.007544484,0.003957623,0.003772966,0.003041576,0.003488846,0.00205509],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00145139,"about_ca_system_score_gemma":0.002464339,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00191029,"about_ca_topic_score_gemma":0.001033397,"domain_scores_codex":[0.880947,0.09332233,0.006248797,0.006851892,0.0116394,0.0009905659],"domain_scores_gemma":[0.6266996,0.3223576,0.0240376,0.01971458,0.006229301,0.0009612389],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.003509816,0.0005444972,0.1619432,0.003480217,0.005085194,0.001175025,0.001895633,0.02503305,0.004060078,0.2669976,0.03370348,0.4925721],"study_design_scores_gemma":[0.001395891,0.002601458,0.1281327,0.001571769,0.001691646,0.002987967,0.001397963,0.3086506,0.01230779,0.4764229,0.06239864,0.0004406149],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01982108,0.001158412,0.9687081,0.0006563682,0.000320631,0.001308316,0.002772326,0.001411181,0.003843636],"genre_scores_gemma":[0.3184685,0.0007240424,0.6667436,0.000943055,0.0003946361,0.005830506,0.004215179,0.0007186478,0.001961944],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.09158804,"threshold_uncertainty_score":0.4843696,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Elizabeth A. Stuart","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6765608631638398,"gpt":0.5702584500397571,"spread":0.1063024131240827,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3219493,0.001957789,0.00456967,0.009172669,0.001790489,0.01072376,0.007313782,0.007535535,0.002848967],"category_scores_gemma":[0.5747454,0.002086388,0.003907751,0.01045066,0.02269435,0.01663251,0.01022789,0.01856037,0.001250197],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00563736,"about_ca_system_score_gemma":0.01103963,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004402351,"about_ca_topic_score_gemma":0.003131632,"domain_scores_codex":[0.5464783,0.3953557,0.02092759,0.01266301,0.02350643,0.00106892],"domain_scores_gemma":[0.4113832,0.5122638,0.02444915,0.03574902,0.01497808,0.001176717],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00009943253,0.00008329565,0.002905237,0.006163369,0.001226492,0.0001766439,0.003135765,0.009738759,0.0003221479,0.7230092,0.01097896,0.2421608],"study_design_scores_gemma":[0.00007939994,0.00007738317,0.0005319855,0.003203576,0.0001395571,0.0001094208,0.0002324469,0.008052526,0.0003939082,0.9689416,0.01816838,0.00006975682],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.001151097,0.01156664,0.9593448,0.02429986,0.0008635535,0.0003328921,0.0001452179,0.0001886445,0.002107228],"genre_scores_gemma":[0.03870384,0.01061026,0.9379324,0.008251778,0.001469575,0.00225951,0.0001742692,0.0001940757,0.000404227],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6780507,"threshold_uncertainty_score":0.8361572,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4544700176042719,"gpt":0.5290019207742545,"spread":0.0745319031699826,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2174702,0.004157696,0.004999245,0.01026048,0.001575541,0.008248786,0.006004136,0.008134787,0.01001388],"category_scores_gemma":[0.5252035,0.001889699,0.008818037,0.01824831,0.006105338,0.01237269,0.008749089,0.01232288,0.003443172],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003038803,"about_ca_system_score_gemma":0.008339749,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002994108,"about_ca_topic_score_gemma":0.001900269,"domain_scores_codex":[0.6394306,0.2924809,0.02367798,0.0108429,0.03267649,0.0008911327],"domain_scores_gemma":[0.429904,0.4315079,0.05637686,0.06826769,0.01300956,0.0009340929],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004632436,0.0002206709,0.007671661,0.01134305,0.006081732,0.0002169762,0.001714931,0.01972799,0.001002432,0.5535533,0.05548134,0.3425226],"study_design_scores_gemma":[0.0006025329,0.0006282846,0.004716466,0.005001329,0.001816547,0.0007356432,0.0003969387,0.03477466,0.003419492,0.7842777,0.1631424,0.0004880102],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0007856977,0.007410959,0.9824929,0.002335476,0.001091803,0.001315961,0.001651181,0.000509948,0.002405997],"genre_scores_gemma":[0.03131308,0.01598787,0.9277407,0.004357662,0.002733326,0.01266883,0.002271846,0.0008063878,0.00212038],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7825298,"threshold_uncertainty_score":0.9649985,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Petra Macaskill","is_ca":false},{"name":"Stephen D. Walter","is_ca":true},{"name":"Les Irwig","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8953613622891847,"gpt":0.6824856965509237,"spread":0.212875665738261,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4238874,0.005533711,0.01993433,0.02512137,0.001432013,0.007888622,0.006231738,0.007294845,0.005970607],"category_scores_gemma":[0.7209188,0.003647395,0.03434489,0.01822765,0.003330619,0.008440174,0.005306141,0.006502141,0.001049679],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004764911,"about_ca_system_score_gemma":0.005311054,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001538392,"about_ca_topic_score_gemma":0.002095102,"domain_scores_codex":[0.3105698,0.6002141,0.04400907,0.008195844,0.03623142,0.0007797424],"domain_scores_gemma":[0.1989238,0.7502161,0.01937294,0.0164789,0.01411888,0.0008894161],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"meta_analysis","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.02219209,0.0003218859,0.01206903,0.1062808,0.4883809,0.0003563853,0.001498509,0.01195437,0.000909183,0.0150688,0.008627956,0.33234],"study_design_scores_gemma":[0.0713966,0.01801563,0.05866822,0.09118789,0.4542598,0.002865035,0.001673552,0.08521409,0.00586105,0.1578561,0.04946071,0.003541309],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02213099,0.3918507,0.5524138,0.00420674,0.004916867,0.0147445,0.003970131,0.001972024,0.003794176],"genre_scores_gemma":[0.1696862,0.1060431,0.663268,0.002111973,0.001494926,0.0521397,0.002604633,0.001134366,0.001517112],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5761126,"threshold_uncertainty_score":0.7104495,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2610013184506083,"gpt":0.5385518179316583,"spread":0.27755049948105,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05833513,0.001329439,0.004062122,0.02584374,0.000761193,0.003283175,0.001759189,0.002228014,0.002259823],"category_scores_gemma":[0.2239352,0.001473965,0.003369766,0.02596256,0.001959588,0.004014051,0.001824863,0.001456791,0.0005482028],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008626286,"about_ca_system_score_gemma":0.01494324,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003820794,"about_ca_topic_score_gemma":0.006543594,"domain_scores_codex":[0.9621062,0.01544194,0.01266201,0.001348254,0.008068171,0.0003734498],"domain_scores_gemma":[0.7730316,0.1399897,0.03943323,0.003207063,0.04342225,0.0009161732],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0007162528,0.0000383889,0.002134661,0.2777888,0.00306015,0.0003837084,0.0006293024,0.0004497331,0.0003200697,0.004256745,0.028777,0.6814451],"study_design_scores_gemma":[0.0004692006,0.0005288425,0.01708895,0.5946848,0.01622293,0.001766398,0.0008723705,0.000550853,0.001195812,0.006874116,0.3596042,0.0001415402],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0004594776,0.9956713,0.0004080421,0.00229579,0.0005183068,0.0001847842,0.00007661323,0.00000605359,0.0003796209],"genre_scores_gemma":[0.006608132,0.9885823,0.001570967,0.00185631,0.0006772203,0.0003861758,0.0001445081,0.000005954185,0.0001685134],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9416649,"threshold_uncertainty_score":0.3085093,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Jason P. Fine","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7862327183202007,"gpt":0.6125248362078489,"spread":0.1737078821123518,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3093101,0.003983113,0.007289951,0.02095473,0.003750229,0.01715911,0.01551752,0.02079948,0.09798504],"category_scores_gemma":[0.758992,0.005966932,0.01366755,0.02637561,0.006287429,0.01728817,0.00960404,0.02122797,0.07530195],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007714467,"about_ca_system_score_gemma":0.04384423,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01513258,"about_ca_topic_score_gemma":0.01392906,"domain_scores_codex":[0.5829651,0.2693856,0.0887128,0.00601859,0.04878734,0.004130531],"domain_scores_gemma":[0.1328578,0.4924123,0.06120536,0.06263549,0.2440237,0.006865413],"domain_codex":null,"domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002083949,0.000172651,0.001023936,0.006452402,0.000286545,0.0002082638,0.0004874475,0.001441153,0.0002937878,0.008188965,0.9224579,0.05877852],"study_design_scores_gemma":[0.0007848372,0.0002107025,0.002484531,0.02937883,0.0004307036,0.0004834707,0.001115845,0.004822148,0.001419075,0.03671402,0.9217334,0.0004225417],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002105903,0.02497889,0.3831983,0.3080322,0.1057601,0.02624376,0.0606298,0.0219775,0.06707347],"genre_scores_gemma":[0.009412916,0.01841504,0.733041,0.1138551,0.01395861,0.05357594,0.02396997,0.008519219,0.0252523],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6906899,"threshold_uncertainty_score":0.8517436,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Paul Grootendorst","is_ca":true},{"name":"Geoffrey M. Anderson","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2043349314626578,"gpt":0.4063998310840797,"spread":0.2020648996214219,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04845132,0.0008524176,0.001680212,0.001426422,0.0007175,0.001347724,0.001399815,0.001813798,0.001168846],"category_scores_gemma":[0.1220498,0.0007098514,0.002424733,0.001186239,0.001682484,0.001906955,0.001131405,0.001636814,0.0001568893],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001986346,"about_ca_system_score_gemma":0.00167404,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007454513,"about_ca_topic_score_gemma":0.003993871,"domain_scores_codex":[0.9815192,0.01597861,0.0005199485,0.0008710177,0.0007402697,0.0003708512],"domain_scores_gemma":[0.740361,0.2402902,0.005693611,0.009473369,0.003088922,0.001092955],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.006049864,0.001247361,0.04852712,0.0002950386,0.002514031,0.0002295064,0.0006461812,0.8653165,0.001168928,0.03875001,0.001361311,0.03389415],"study_design_scores_gemma":[0.0004035115,0.00111595,0.007384622,0.00008313108,0.0002840767,0.0001026954,0.000104505,0.9747498,0.0009780137,0.01404984,0.0006708329,0.00007308325],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8463221,0.001526185,0.147948,0.000655163,0.00006792765,0.0006625701,0.0003037346,0.0001557298,0.002358643],"genre_scores_gemma":[0.9491639,0.0005614044,0.04845314,0.0002611997,0.00003691712,0.0004652184,0.000405913,0.00005329924,0.0005989801],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9515487,"threshold_uncertainty_score":0.2562381,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.220207531187187,"gpt":0.4904911441960751,"spread":0.2702836130088881,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08822343,0.002135994,0.00260438,0.006001812,0.001082461,0.00325052,0.003723077,0.001975625,0.004914273],"category_scores_gemma":[0.2533046,0.001278077,0.005449877,0.007219727,0.001711088,0.004071855,0.002929022,0.003688971,0.001151221],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001936039,"about_ca_system_score_gemma":0.003483006,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008023952,"about_ca_topic_score_gemma":0.006251296,"domain_scores_codex":[0.9445329,0.0427882,0.003535114,0.003798114,0.004837677,0.0005080369],"domain_scores_gemma":[0.8561339,0.1137137,0.00880006,0.01264468,0.008081779,0.0006259437],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001250341,0.0003003823,0.06174538,0.002009457,0.006442712,0.0002785465,0.001276182,0.2026966,0.001411559,0.1096856,0.01044673,0.6024566],"study_design_scores_gemma":[0.000765769,0.0005923841,0.02605526,0.001010339,0.001514827,0.000585917,0.0003417928,0.788509,0.003236327,0.1551221,0.02180606,0.0004602984],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01159006,0.002482056,0.982255,0.0005612064,0.0001496168,0.0007394147,0.0005586121,0.0005693061,0.001094631],"genre_scores_gemma":[0.149355,0.003587149,0.8403804,0.0004836649,0.0002539513,0.002448234,0.001760233,0.0005919473,0.001139475],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9117765,"threshold_uncertainty_score":0.4665757,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Allan Donner","is_ca":true},{"name":"Neil Klar","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8482658612698941,"gpt":0.6110797934992076,"spread":0.2371860677706865,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7382926,0.005827221,0.02295667,0.01863314,0.004805279,0.0181993,0.01470004,0.01551318,0.002453592],"category_scores_gemma":[0.8853183,0.006735492,0.02704738,0.02485002,0.01604725,0.01511536,0.01035261,0.02578525,0.001018125],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01227026,"about_ca_system_score_gemma":0.01993515,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006496266,"about_ca_topic_score_gemma":0.008145034,"domain_scores_codex":[0.11926,0.7811422,0.05986691,0.009463499,0.02927176,0.0009955857],"domain_scores_gemma":[0.06565815,0.8848082,0.01784725,0.01821662,0.01258157,0.0008882469],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.005495173,0.0001657897,0.005945275,0.1118563,0.1488085,0.001467317,0.006941953,0.01720129,0.0005770616,0.2374998,0.0843548,0.3796868],"study_design_scores_gemma":[0.00384643,0.001252869,0.004179916,0.04977133,0.05714638,0.001521877,0.001359452,0.0277568,0.002071753,0.7814651,0.06872618,0.0009018634],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002579002,0.3635793,0.4822847,0.1203297,0.02318634,0.002663543,0.0007576947,0.001217509,0.003402191],"genre_scores_gemma":[0.1375712,0.0982587,0.6579384,0.06265447,0.02499293,0.01534706,0.0004742661,0.001386301,0.00137674],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2617074,"threshold_uncertainty_score":0.3227319,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Stephen D. Walter","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08867909890496116,"gpt":0.319686173518747,"spread":0.2310070746137858,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2137865,0.002295628,0.005944086,0.01171437,0.0007488153,0.004607825,0.002781387,0.004419207,0.002128841],"category_scores_gemma":[0.5627906,0.0009767555,0.008485937,0.009762902,0.00381711,0.0055694,0.003205864,0.003877359,0.001020533],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00236818,"about_ca_system_score_gemma":0.001848366,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001923364,"about_ca_topic_score_gemma":0.0006943571,"domain_scores_codex":[0.8294262,0.1336196,0.01151037,0.01157804,0.01268207,0.001183782],"domain_scores_gemma":[0.2789591,0.6408747,0.04211726,0.02456966,0.01253742,0.0009417826],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.004695915,0.0001785374,0.238609,0.01615573,0.0406846,0.001629629,0.001980521,0.2185784,0.002173209,0.06749182,0.01889123,0.3889314],"study_design_scores_gemma":[0.0008034006,0.00416278,0.125264,0.006758908,0.01830628,0.008458802,0.001149706,0.5105479,0.003642619,0.2740712,0.04584658,0.0009878373],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0495377,0.08183751,0.8478735,0.005077437,0.0007023217,0.001363431,0.004823257,0.002173863,0.006610891],"genre_scores_gemma":[0.7465599,0.01548986,0.2214232,0.004059997,0.001274848,0.003166283,0.005854897,0.001100035,0.001070944],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9940559,"threshold_uncertainty_score":0.9695413,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Juan Merlo","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1141231773850907,"gpt":0.4782803494377693,"spread":0.3641571720526786,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02834336,0.001470548,0.002011997,0.005637617,0.00108866,0.00432164,0.002497461,0.002526956,0.01719668],"category_scores_gemma":[0.1332437,0.0009230289,0.003603144,0.009889791,0.00301197,0.004496648,0.005159437,0.008467041,0.005828047],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00169445,"about_ca_system_score_gemma":0.003567186,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00193021,"about_ca_topic_score_gemma":0.001764611,"domain_scores_codex":[0.964048,0.02592913,0.002121622,0.002672496,0.004683586,0.0005451192],"domain_scores_gemma":[0.8925722,0.09010526,0.005454675,0.006095084,0.004720509,0.001052383],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001689502,0.0001549022,0.01434522,0.004088396,0.001157697,0.0006751585,0.001448972,0.01138842,0.0008708251,0.4339261,0.1257166,0.4060588],"study_design_scores_gemma":[0.00004978274,0.0001830379,0.006036942,0.002181758,0.0002528743,0.0006592761,0.0003318209,0.04634482,0.0006501324,0.7539292,0.1892503,0.00013016],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.004759708,0.04278202,0.9099298,0.02242147,0.002956426,0.0003262149,0.001668298,0.001454522,0.01370161],"genre_scores_gemma":[0.1237067,0.04893828,0.7853301,0.007233446,0.01444516,0.002380565,0.00446634,0.002266329,0.01123322],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.02834336,"threshold_uncertainty_score":0.1498958,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Eleanor Sanderson","is_ca":false},{"name":"Wes Spiller","is_ca":false},{"name":"Jack Bowden","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02197166481314274,"gpt":0.3276798526833534,"spread":0.3057081878702106,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1206779,0.001273392,0.003229533,0.002436074,0.001253884,0.002217139,0.003607195,0.002433836,0.003295881],"category_scores_gemma":[0.3966193,0.0008981576,0.003311777,0.00339949,0.004877315,0.00308692,0.00407042,0.003161253,0.0003915182],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009800218,"about_ca_system_score_gemma":0.004863716,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002755305,"about_ca_topic_score_gemma":0.002271947,"domain_scores_codex":[0.8769611,0.1007902,0.00478314,0.00789833,0.007575898,0.001991394],"domain_scores_gemma":[0.5954171,0.3590164,0.01346634,0.02603747,0.005090317,0.0009723837],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001941186,0.0004048036,0.1319482,0.001010042,0.005247696,0.002114333,0.001767853,0.09645776,0.004030628,0.4319166,0.004249178,0.3189118],"study_design_scores_gemma":[0.0005798668,0.001126585,0.02722033,0.0002660335,0.0007739,0.0008649697,0.0003197328,0.4541307,0.005287883,0.5027009,0.00651935,0.0002097649],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02917955,0.0001804271,0.9689665,0.0004209923,0.00008459651,0.0001993354,0.0001604513,0.0003286327,0.0004795105],"genre_scores_gemma":[0.3849293,0.0002542191,0.6117235,0.0005328511,0.0001244308,0.001104413,0.000413315,0.0001621268,0.0007558428],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8793221,"threshold_uncertainty_score":0.6382133,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Muhammad Mamdani","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3195657084475746,"gpt":0.5520056064708857,"spread":0.2324398980233111,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1602731,0.001386997,0.002962042,0.003939331,0.0009784806,0.002534657,0.002575693,0.004596793,0.004196114],"category_scores_gemma":[0.3171187,0.001156537,0.007926463,0.002899309,0.002142516,0.0027392,0.002530872,0.00206121,0.0004343371],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00171994,"about_ca_system_score_gemma":0.001738637,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002435663,"about_ca_topic_score_gemma":0.001369295,"domain_scores_codex":[0.7997295,0.183722,0.005987247,0.0035064,0.006288556,0.0007662192],"domain_scores_gemma":[0.5786208,0.381482,0.01477414,0.01982217,0.004505003,0.0007958965],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.12085,0.006861855,0.314453,0.006946105,0.07888829,0.002213478,0.002451003,0.06738662,0.00223261,0.07985225,0.005979885,0.311885],"study_design_scores_gemma":[0.04950507,0.05221142,0.1831572,0.003986435,0.06093164,0.006454949,0.003474533,0.5229113,0.005480115,0.09008022,0.0206837,0.001123513],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.6975621,0.0377024,0.2486651,0.005035453,0.001027204,0.003872346,0.001404724,0.0001706066,0.004559962],"genre_scores_gemma":[0.9365861,0.004597637,0.05500231,0.0003704702,0.0002762365,0.002023075,0.0004936794,0.00004849537,0.0006019714],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.1602731,"threshold_uncertainty_score":0.8476155,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Guangyong Zou","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0854428886232476,"gpt":0.3958068923931848,"spread":0.3103640037699372,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07010303,0.001945607,0.002043521,0.006580195,0.0006737425,0.001839532,0.003850206,0.002883459,0.005361066],"category_scores_gemma":[0.438004,0.001025973,0.001452131,0.003597216,0.002146259,0.004309688,0.002959295,0.00473822,0.002070335],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001470158,"about_ca_system_score_gemma":0.001892213,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001341249,"about_ca_topic_score_gemma":0.001593563,"domain_scores_codex":[0.945505,0.03618129,0.003141245,0.002406095,0.01235633,0.0004100172],"domain_scores_gemma":[0.6935443,0.2722119,0.008609248,0.0114983,0.01373681,0.0003995289],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002959715,0.0002159738,0.004739175,0.001086075,0.0004376358,0.0003353592,0.001241872,0.05546988,0.002294069,0.2932191,0.01682123,0.6238437],"study_design_scores_gemma":[0.0005529001,0.0007522691,0.006288769,0.001625516,0.0004590926,0.001145403,0.0002471306,0.4374853,0.006104487,0.5141327,0.03098891,0.0002173885],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0008068777,0.0004975127,0.9973032,0.000145091,0.00006766779,0.0002056953,0.00004702564,0.0001767709,0.0007500206],"genre_scores_gemma":[0.03011104,0.0009770368,0.9642659,0.0003458276,0.0002481178,0.002939021,0.0001731038,0.0002034385,0.0007365174],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.07010303,"threshold_uncertainty_score":0.3707446,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4011751616883879,"gpt":0.5200551633034011,"spread":0.1188800016150132,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4396123,0.002058965,0.003944528,0.00764646,0.00156653,0.005256041,0.004350678,0.003243097,0.002818059],"category_scores_gemma":[0.7092298,0.001779044,0.01046743,0.01240259,0.003932175,0.005454705,0.004308516,0.004812075,0.0005815061],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003350895,"about_ca_system_score_gemma":0.004824952,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005815591,"about_ca_topic_score_gemma":0.003693299,"domain_scores_codex":[0.508193,0.413272,0.03620483,0.01625932,0.02482633,0.001244452],"domain_scores_gemma":[0.2754267,0.6212667,0.04339208,0.04115813,0.01791665,0.0008398208],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.004236432,0.0004164733,0.1859894,0.00841173,0.03437318,0.0004708153,0.003481008,0.1572051,0.001278455,0.09154662,0.009445312,0.5031455],"study_design_scores_gemma":[0.003740691,0.003445243,0.1456593,0.007521451,0.01458931,0.001705226,0.0009385151,0.5086161,0.007220493,0.2627421,0.04248298,0.001338544],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03475623,0.01307819,0.9416749,0.001761582,0.0005445868,0.004067523,0.00152141,0.0004699106,0.002125594],"genre_scores_gemma":[0.3292992,0.006368985,0.6510851,0.001250981,0.000412748,0.008857602,0.001723367,0.0004349096,0.000567083],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5603877,"threshold_uncertainty_score":0.6910578,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Ewout W. Steyerberg","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4241021929435487,"gpt":0.5779164790022724,"spread":0.1538142860587237,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07874872,0.001715407,0.001688042,0.004584369,0.0008355074,0.003093741,0.002132752,0.002363902,0.003445942],"category_scores_gemma":[0.3413173,0.0008258577,0.002036219,0.002491859,0.002252361,0.003270011,0.002967285,0.00340351,0.0007052567],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001084529,"about_ca_system_score_gemma":0.001235826,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001843078,"about_ca_topic_score_gemma":0.001545215,"domain_scores_codex":[0.9562913,0.03630672,0.001905734,0.002001901,0.002995259,0.0004992372],"domain_scores_gemma":[0.6342739,0.3166513,0.01714059,0.02058585,0.01048043,0.0008679633],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002475566,0.0003367301,0.0852329,0.001680568,0.002119622,0.0007577693,0.003053861,0.501369,0.008543511,0.07525796,0.00686803,0.3123046],"study_design_scores_gemma":[0.0001717481,0.0006038301,0.01588172,0.0002858764,0.0002785015,0.000429949,0.0002596489,0.9132946,0.004890879,0.06033158,0.003342779,0.0002289165],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04190331,0.0005035912,0.9541317,0.000409401,0.0000654274,0.0001816116,0.0002090454,0.001883068,0.0007128027],"genre_scores_gemma":[0.6047517,0.0004517274,0.3913553,0.0003482339,0.00007691968,0.0006768335,0.0007867817,0.0007790966,0.0007734009],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.07874872,"threshold_uncertainty_score":0.416468,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6053372113795614,"gpt":0.5275261212026646,"spread":0.0778110901768968,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3921677,0.001958479,0.003827057,0.003772027,0.001552199,0.005874407,0.004651726,0.003755744,0.008487406],"category_scores_gemma":[0.7571114,0.001709421,0.007897763,0.004512876,0.006821116,0.006840642,0.004921835,0.008159158,0.001184618],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002002545,"about_ca_system_score_gemma":0.003287762,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001084618,"about_ca_topic_score_gemma":0.001071493,"domain_scores_codex":[0.3320736,0.6052327,0.01939249,0.01550575,0.02658703,0.001208378],"domain_scores_gemma":[0.1440299,0.7566448,0.0317837,0.05181341,0.01491351,0.0008146727],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01301471,0.002176652,0.06633946,0.01309498,0.0349429,0.001193749,0.0100308,0.04449741,0.003884831,0.2143002,0.02376537,0.5727589],"study_design_scores_gemma":[0.006752149,0.01620483,0.07463906,0.00922901,0.01476885,0.002174336,0.003535578,0.265134,0.02951462,0.5006917,0.07604178,0.001314232],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02143456,0.002079766,0.9641409,0.001356372,0.001450131,0.004995408,0.0005793584,0.000549677,0.003413794],"genre_scores_gemma":[0.2332365,0.0009323264,0.7443196,0.001942047,0.0005003312,0.01689554,0.0005524331,0.0005322728,0.001088874],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6078323,"threshold_uncertainty_score":0.7495655,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Ewout W. Steyerberg","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2038005309243087,"gpt":0.4336053912476601,"spread":0.2298048603233514,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04402304,0.002642101,0.002261909,0.01206949,0.001093012,0.003376483,0.002287825,0.003314147,0.002312018],"category_scores_gemma":[0.2174493,0.0007763252,0.00215769,0.00910558,0.003334099,0.005532961,0.003874691,0.00495687,0.0009342631],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001955717,"about_ca_system_score_gemma":0.002099974,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002726519,"about_ca_topic_score_gemma":0.002229003,"domain_scores_codex":[0.9738236,0.01211325,0.002353657,0.00347187,0.007594933,0.0006427259],"domain_scores_gemma":[0.781864,0.1663212,0.02338548,0.01627878,0.01094029,0.001210272],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004804918,0.0002723754,0.1442982,0.0009911953,0.001784306,0.0003402567,0.0008320106,0.488201,0.003586958,0.07502525,0.01265025,0.2715378],"study_design_scores_gemma":[0.00007718655,0.0006526493,0.07085264,0.000582044,0.0003067689,0.001202337,0.0004008675,0.7384056,0.006472863,0.1664782,0.01405122,0.0005176348],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03875963,0.003948863,0.9474398,0.0007626055,0.000210991,0.0002631779,0.001508704,0.001217261,0.00588888],"genre_scores_gemma":[0.580875,0.002938396,0.4058964,0.001009627,0.0005870519,0.001281864,0.004394684,0.0008760622,0.002140875],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.04402304,"threshold_uncertainty_score":0.2328188,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Dylan S. Small","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3885413863266839,"gpt":0.4667305653599856,"spread":0.07818917903330169,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05055559,0.001215734,0.001752287,0.001828218,0.001210173,0.001597011,0.002130625,0.002479335,0.003221816],"category_scores_gemma":[0.1556311,0.0006687638,0.002725584,0.002880584,0.001300426,0.002685998,0.001876534,0.003218769,0.0003424583],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001856676,"about_ca_system_score_gemma":0.001775955,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008527655,"about_ca_topic_score_gemma":0.00665656,"domain_scores_codex":[0.9766982,0.02050215,0.0007575348,0.0006807402,0.0009554497,0.0004059921],"domain_scores_gemma":[0.7162633,0.2606851,0.005404834,0.009940499,0.006863188,0.0008431509],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.004719596,0.003925244,0.0564051,0.001437992,0.00201607,0.001223123,0.00158429,0.7029054,0.0012147,0.1248049,0.007153914,0.09260957],"study_design_scores_gemma":[0.0008707447,0.00121998,0.003987448,0.0003679151,0.000556774,0.0002852404,0.0003524396,0.9548794,0.001149899,0.03293831,0.003290683,0.0001011657],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5119101,0.004236835,0.4669323,0.002338636,0.0002870101,0.002882623,0.0009938418,0.0002960035,0.01012269],"genre_scores_gemma":[0.8178797,0.002010099,0.1752843,0.0003949073,0.0001263233,0.002420425,0.0007119243,0.00008999816,0.001082383],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9494444,"threshold_uncertainty_score":0.2673668,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Robarts Clinical Trials; Western University","funders":"","keywords":"Confidence interval; Computer science; Statistics; Log-normal distribution; Econometrics; Mathematics","authors":[{"name":"Guangyong Zou","is_ca":true},{"name":"Allan Donner","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5277622902536679,"gpt":0.527346483686242,"spread":0.0004158065674259026,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2446331,0.004117253,0.006552316,0.01641745,0.001767202,0.01207092,0.01001868,0.008598731,0.008173862],"category_scores_gemma":[0.6110272,0.003090081,0.007555156,0.01187527,0.008545684,0.01167755,0.009474022,0.01434381,0.002063339],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003263833,"about_ca_system_score_gemma":0.005303432,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001974521,"about_ca_topic_score_gemma":0.0006066632,"domain_scores_codex":[0.7788684,0.1673997,0.01568256,0.01109106,0.02536477,0.001593536],"domain_scores_gemma":[0.3755587,0.5682601,0.01184856,0.02521447,0.01813339,0.0009847729],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004251577,0.0001387782,0.001866996,0.002788126,0.001505473,0.0007204249,0.002046017,0.04013478,0.001194088,0.7192323,0.00439072,0.2255571],"study_design_scores_gemma":[0.0003245016,0.0003875048,0.0008952839,0.001689241,0.0005469972,0.0007254691,0.000262418,0.09991635,0.002273939,0.8753414,0.01745066,0.0001862237],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0005352398,0.0005484005,0.9974643,0.0002519017,0.00005931224,0.0001983747,0.0000590614,0.0001960295,0.0006873434],"genre_scores_gemma":[0.04681401,0.001561987,0.9465151,0.0004813574,0.000446223,0.003053898,0.0003218,0.0002799635,0.0005257062],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7553669,"threshold_uncertainty_score":0.9315019,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Michael J. Campbell","is_ca":false},{"name":"Allan Donner","is_ca":true},{"name":"Neil Klar","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08618382227126765,"gpt":0.4329159356107219,"spread":0.3467321133394543,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1457657,0.002188136,0.0059994,0.007049645,0.001398345,0.007007676,0.005191343,0.007548613,0.005148796],"category_scores_gemma":[0.229035,0.001901876,0.003888129,0.01297564,0.01439513,0.00731228,0.004300409,0.02217563,0.002344914],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005929059,"about_ca_system_score_gemma":0.01084121,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002570689,"about_ca_topic_score_gemma":0.001339911,"domain_scores_codex":[0.7996178,0.168229,0.008494695,0.005969251,0.01669221,0.0009970459],"domain_scores_gemma":[0.5907113,0.3693125,0.009268369,0.01601173,0.01283852,0.001857678],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000189679,0.00007970778,0.0004748394,0.00422494,0.0005055555,0.0001243827,0.0004970733,0.004075204,0.0002146068,0.7620524,0.02749569,0.200066],"study_design_scores_gemma":[0.0001777778,0.0003117391,0.000605182,0.002757224,0.0001829798,0.0004756189,0.000136426,0.009123604,0.000529565,0.8313168,0.1542733,0.0001096691],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0005791133,0.1698077,0.7673584,0.03882653,0.01022588,0.0006341304,0.0002573007,0.0004433196,0.01186766],"genre_scores_gemma":[0.01904556,0.1363267,0.798959,0.01945859,0.02003163,0.003277551,0.0002347149,0.0005358639,0.002130437],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8542343,"threshold_uncertainty_score":0.7708917,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Lawrence J. Brunner","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5624346742081207,"gpt":0.6001798666323802,"spread":0.03774519242425955,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4783034,0.002276254,0.00370262,0.004241432,0.001905786,0.004399271,0.003450413,0.005750129,0.001969605],"category_scores_gemma":[0.7635874,0.00161082,0.00482435,0.005101701,0.009339102,0.005247748,0.00556312,0.01027712,0.0007312943],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004499334,"about_ca_system_score_gemma":0.003941691,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002364938,"about_ca_topic_score_gemma":0.002308478,"domain_scores_codex":[0.3503839,0.5251655,0.04071745,0.02961154,0.0511537,0.002967902],"domain_scores_gemma":[0.08561659,0.8228095,0.03795846,0.0398222,0.0128536,0.000939805],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01489818,0.001153811,0.3242511,0.006483751,0.01559317,0.007480071,0.01484226,0.08283249,0.007979672,0.1433687,0.0150659,0.3660508],"study_design_scores_gemma":[0.001807121,0.006062985,0.144685,0.006466531,0.008517521,0.01012712,0.003422387,0.4521603,0.03348206,0.3083527,0.02316189,0.001754396],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.04300847,0.0044373,0.9408407,0.003113846,0.001976466,0.002134117,0.0003753471,0.0009323945,0.003181328],"genre_scores_gemma":[0.5759999,0.001706388,0.4078554,0.004886552,0.001013534,0.005482162,0.0005758298,0.0007114265,0.001768901],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5216966,"threshold_uncertainty_score":0.6433448,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Stephen D. Walter","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6102396939187572,"gpt":0.5384282700556492,"spread":0.07181142386310801,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06241108,0.002096521,0.004661723,0.01087462,0.0006143747,0.00653528,0.002476352,0.003323243,0.0068637],"category_scores_gemma":[0.2485188,0.0006307205,0.006036657,0.008307933,0.002796534,0.005195881,0.002824867,0.002699155,0.003069209],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001994031,"about_ca_system_score_gemma":0.002388953,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001737756,"about_ca_topic_score_gemma":0.0007760145,"domain_scores_codex":[0.9160307,0.05673062,0.007280464,0.010384,0.008655848,0.0009184812],"domain_scores_gemma":[0.7268672,0.2096012,0.02621201,0.01935622,0.01660447,0.00135893],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002392562,0.000158332,0.113518,0.01245956,0.01749619,0.0009072157,0.001129293,0.08147021,0.002296921,0.07091708,0.05128358,0.645971],"study_design_scores_gemma":[0.0007197437,0.005824748,0.1068771,0.007984846,0.01265615,0.01234331,0.001815044,0.3208249,0.006316218,0.2869674,0.2362472,0.001423379],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05993835,0.07337894,0.8056045,0.005104248,0.001825735,0.001832647,0.02195456,0.005140054,0.02522093],"genre_scores_gemma":[0.7351241,0.01814378,0.2197669,0.003309276,0.002124151,0.004729053,0.01215049,0.001074292,0.003577921],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9375889,"threshold_uncertainty_score":0.3300653,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1636371954417831,"gpt":0.4737266236584863,"spread":0.3100894282167033,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07765374,0.001842488,0.001909466,0.005909416,0.0006450707,0.002474871,0.002378148,0.001851124,0.003029704],"category_scores_gemma":[0.2601441,0.000908884,0.003394673,0.00519626,0.001498364,0.003180929,0.002310894,0.002083139,0.0007359381],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001531168,"about_ca_system_score_gemma":0.001973629,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002907734,"about_ca_topic_score_gemma":0.002390172,"domain_scores_codex":[0.932119,0.05553478,0.003222069,0.003528567,0.005217033,0.0003785641],"domain_scores_gemma":[0.7956067,0.1743165,0.00991007,0.01320142,0.006359062,0.0006063514],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001839229,0.0003640537,0.05834553,0.00159954,0.005712131,0.0002257886,0.0008432587,0.3148263,0.002208655,0.120062,0.003839166,0.4901344],"study_design_scores_gemma":[0.0006096481,0.0006888587,0.02177115,0.0005251091,0.0008459648,0.0005154789,0.0001579737,0.8345155,0.004067289,0.1289049,0.007118532,0.0002795107],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01834768,0.002233122,0.97727,0.0003114551,0.00008146587,0.0003295512,0.0002892025,0.0002946577,0.0008429358],"genre_scores_gemma":[0.2071425,0.003194039,0.7859131,0.0002349211,0.0001644628,0.001298422,0.0009474799,0.0002972546,0.0008077674],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9223462,"threshold_uncertainty_score":0.4106771,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Paul Grootendorst","is_ca":true},{"name":"Sharon‐Lise T. Normand","is_ca":false},{"name":"Geoffrey M. Anderson","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2038916704515713,"gpt":0.4246693777064753,"spread":0.2207777072549041,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1084491,0.0009352491,0.002185585,0.001543988,0.001244829,0.002388433,0.001982056,0.002393341,0.003890577],"category_scores_gemma":[0.3528894,0.0009293345,0.002563277,0.002061693,0.003917318,0.003463022,0.002268258,0.003538779,0.0003475638],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00214579,"about_ca_system_score_gemma":0.002718054,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004296117,"about_ca_topic_score_gemma":0.003023861,"domain_scores_codex":[0.9323142,0.06045049,0.001389781,0.002586807,0.002556632,0.0007019682],"domain_scores_gemma":[0.3842634,0.5775942,0.01191864,0.020674,0.004791589,0.0007581282],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001617984,0.0006440202,0.05971384,0.000702092,0.002926597,0.0008825893,0.001286864,0.2971936,0.001187016,0.5228561,0.004291138,0.1066982],"study_design_scores_gemma":[0.0004369841,0.0004143618,0.007276176,0.0003827135,0.000696883,0.0003848098,0.0001312093,0.7206703,0.001623986,0.2637204,0.004152644,0.0001095225],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1327234,0.003267655,0.8559098,0.003077977,0.0001243874,0.0007136401,0.0001797498,0.0001850433,0.003818324],"genre_scores_gemma":[0.7914546,0.002234229,0.2018158,0.001163652,0.000222396,0.001207999,0.0002587823,0.0001318442,0.001510751],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8915508,"threshold_uncertainty_score":0.5735407,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Muhammad Mamdani","is_ca":true},{"name":"Thérèse A. Stukel","is_ca":true},{"name":"Geoffrey M. Anderson","is_ca":true},{"name":"Jack V. Tu","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7319400186447765,"gpt":0.5861486378777672,"spread":0.1457913807670094,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09515899,0.0008925841,0.002031899,0.006047806,0.0007102516,0.00321476,0.001651156,0.002075332,0.002356262],"category_scores_gemma":[0.3099065,0.0006148887,0.001813206,0.009455675,0.003070486,0.003891934,0.002778591,0.002793218,0.0003883479],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001237878,"about_ca_system_score_gemma":0.00232789,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002096217,"about_ca_topic_score_gemma":0.00147036,"domain_scores_codex":[0.8547019,0.1300315,0.004255374,0.003354359,0.00718459,0.0004722185],"domain_scores_gemma":[0.7228,0.2267459,0.02321887,0.02204198,0.004294283,0.0008990102],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001110565,0.0004409205,0.1815073,0.001705734,0.005063971,0.0003122253,0.001178518,0.04542626,0.001009637,0.3676823,0.005020197,0.3895423],"study_design_scores_gemma":[0.001220227,0.002130459,0.1126904,0.001217465,0.001685187,0.00111367,0.0007488013,0.1997174,0.002914384,0.6483116,0.02793487,0.000315627],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0364692,0.004108901,0.9514685,0.003570345,0.0002758531,0.0005688406,0.0009930548,0.0001362905,0.002409021],"genre_scores_gemma":[0.5472874,0.005537903,0.4398868,0.001572934,0.0008182395,0.002021267,0.001690907,0.00009622685,0.001088381],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.904841,"threshold_uncertainty_score":0.5032549,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Nicole Blackman","is_ca":false},{"name":"John J. Koval","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1194851870437682,"gpt":0.4260948892710158,"spread":0.3066097022272476,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0671707,0.0009194962,0.002020541,0.01051445,0.001272814,0.003126918,0.004042391,0.001764458,0.005610983],"category_scores_gemma":[0.351872,0.0004554826,0.002152083,0.007647032,0.002175091,0.003753557,0.00385122,0.003072929,0.001801827],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001567063,"about_ca_system_score_gemma":0.002102059,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001530363,"about_ca_topic_score_gemma":0.0008842369,"domain_scores_codex":[0.8824344,0.07347108,0.009351484,0.008111298,0.0255265,0.001105324],"domain_scores_gemma":[0.7267736,0.220119,0.01426846,0.01160473,0.02636991,0.0008644462],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001362526,0.0002520107,0.02050437,0.004183941,0.001781927,0.0003224274,0.004410565,0.02588647,0.003158843,0.1969858,0.02644013,0.714711],"study_design_scores_gemma":[0.0003359136,0.002555149,0.05820322,0.004742139,0.001664742,0.003483329,0.0034158,0.2608244,0.01144719,0.5630474,0.08938649,0.0008941182],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01449865,0.004029885,0.9684323,0.0003412143,0.0004213856,0.0004896483,0.0006398733,0.0007902564,0.01035678],"genre_scores_gemma":[0.2301384,0.002170677,0.7607723,0.0002709093,0.0004029033,0.003491173,0.001069422,0.0004346769,0.001249576],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.0671707,"threshold_uncertainty_score":0.3552369,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Yan Zhou","is_ca":true},{"name":"Nandini Dendukuri","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9168580576152456,"gpt":0.6467614652057476,"spread":0.270096592409498,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2915782,0.003118136,0.009074728,0.01577653,0.0006926995,0.004729509,0.006194029,0.005678514,0.001385939],"category_scores_gemma":[0.5997665,0.001465483,0.01525689,0.01556879,0.004140815,0.00508566,0.004635103,0.00800042,0.0004195361],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002628387,"about_ca_system_score_gemma":0.00236498,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001248588,"about_ca_topic_score_gemma":0.0009457388,"domain_scores_codex":[0.5952849,0.3512408,0.02367696,0.01161998,0.01743191,0.0007453919],"domain_scores_gemma":[0.2731777,0.6699879,0.02254905,0.02859644,0.005335911,0.0003529598],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002026672,0.000151967,0.03362248,0.04316075,0.1855849,0.001833632,0.001911853,0.1576136,0.002185189,0.1840505,0.01812406,0.3697345],"study_design_scores_gemma":[0.0008933693,0.0009877877,0.0123338,0.007532652,0.03864757,0.002275185,0.0004571581,0.2165527,0.00440182,0.6976382,0.01755423,0.0007255357],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.004810434,0.02718587,0.9632841,0.001523655,0.0004548693,0.0005473736,0.001023821,0.0006022661,0.0005676493],"genre_scores_gemma":[0.237032,0.01152454,0.7417206,0.001713071,0.001037422,0.004821503,0.001241934,0.0006039456,0.0003050239],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7084218,"threshold_uncertainty_score":0.8736101,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Frank E. Harrell","is_ca":false},{"name":"David van Klaveren","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3271472948171102,"gpt":0.4229108498472749,"spread":0.09576355503016465,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06051616,0.002142106,0.001582612,0.01102796,0.001029544,0.004202735,0.003037936,0.003744332,0.005373322],"category_scores_gemma":[0.3301638,0.0009500146,0.002436946,0.009534793,0.004331253,0.006786434,0.004423625,0.005766712,0.001185991],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003516067,"about_ca_system_score_gemma":0.002650217,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002930678,"about_ca_topic_score_gemma":0.001288692,"domain_scores_codex":[0.9574413,0.02957649,0.001849129,0.003216896,0.007120054,0.0007960971],"domain_scores_gemma":[0.6804036,0.2550104,0.0265016,0.02305304,0.01370716,0.001324145],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000244765,0.0001256169,0.02308432,0.0005632677,0.000404629,0.0002132832,0.0007282507,0.6102901,0.0008769343,0.2185664,0.008155718,0.1367467],"study_design_scores_gemma":[0.00005125858,0.0002243876,0.009718898,0.0003875072,0.00009001567,0.0004172049,0.0001961912,0.7303763,0.001541395,0.2469802,0.009764533,0.0002520962],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01420968,0.001477859,0.9769179,0.0007425866,0.00009539034,0.0001919285,0.000739555,0.001220329,0.004404936],"genre_scores_gemma":[0.5083797,0.00220189,0.4801441,0.0006662584,0.0003850683,0.001517783,0.003261875,0.001514686,0.00192857],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.06051616,"threshold_uncertainty_score":0.3200439,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Henrik Stryhn","is_ca":true},{"name":"George Leckie","is_ca":false},{"name":"Juan Merlo","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4042531616015181,"gpt":0.5285301629397802,"spread":0.1242770013382621,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02922282,0.0004972554,0.001091739,0.004721245,0.0007556032,0.001975258,0.001761597,0.001045068,0.004373983],"category_scores_gemma":[0.1708047,0.0003801024,0.002338981,0.006826541,0.002038914,0.002046781,0.002158795,0.001754103,0.0005357368],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001563419,"about_ca_system_score_gemma":0.001049616,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005787776,"about_ca_topic_score_gemma":0.003882023,"domain_scores_codex":[0.9649846,0.02073218,0.003008292,0.004096145,0.006469579,0.000709293],"domain_scores_gemma":[0.8146182,0.1374074,0.02403751,0.01803542,0.005137029,0.0007643289],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003745402,0.0001806115,0.5183472,0.001366325,0.006040739,0.001478373,0.003140153,0.08366922,0.001370063,0.1863996,0.02021203,0.1774211],"study_design_scores_gemma":[0.00008901799,0.0003402685,0.3625159,0.0008440107,0.0008011307,0.001699906,0.002007382,0.2473398,0.002280916,0.3585367,0.02327658,0.0002684401],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.2321128,0.003102074,0.7388986,0.004013534,0.0003837951,0.0007307314,0.008288891,0.0009012612,0.01156824],"genre_scores_gemma":[0.897347,0.0005351271,0.09524191,0.0004604709,0.0002230253,0.001180188,0.003053151,0.0002119242,0.00174717],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.02922282,"threshold_uncertainty_score":0.1545469,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Lucinda Archer","is_ca":false},{"name":"Kym I E Snell","is_ca":false},{"name":"Joie Ensor","is_ca":false},{"name":"Mohammed T Hudda","is_ca":false},{"name":"Gary S. Collins","is_ca":false},{"name":"Richard D Riley","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1267781237050936,"gpt":0.4179327851487913,"spread":0.2911546614436977,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1428288,0.0009088044,0.001737079,0.001275561,0.0009832826,0.001906651,0.00243924,0.002417681,0.002930752],"category_scores_gemma":[0.3723291,0.0007118922,0.001677879,0.0007819307,0.001970153,0.002856064,0.003659992,0.00335308,0.0009766378],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001102595,"about_ca_system_score_gemma":0.003843283,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001250652,"about_ca_topic_score_gemma":0.001354559,"domain_scores_codex":[0.9323233,0.0548496,0.003464815,0.003620777,0.005118839,0.0006226348],"domain_scores_gemma":[0.7053518,0.2531127,0.007312169,0.02086695,0.01230515,0.001051345],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.008088199,0.00133444,0.09902495,0.001792413,0.001149757,0.001179203,0.002659313,0.1609408,0.01448099,0.1446785,0.0164551,0.5482163],"study_design_scores_gemma":[0.003079039,0.006134079,0.04557443,0.001768026,0.0006428738,0.001250189,0.0009833634,0.682891,0.02878288,0.2039205,0.02468001,0.0002936177],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03740253,0.0003296148,0.9529772,0.002138434,0.0002437103,0.002047156,0.0004103005,0.0003793359,0.004071654],"genre_scores_gemma":[0.3932779,0.0002229353,0.5962088,0.001209637,0.0001929738,0.007038581,0.000737118,0.0001883392,0.0009237418],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8571712,"threshold_uncertainty_score":0.7553599,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.06665911353813758,"gpt":0.3846642828112441,"spread":0.3180051692731065,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01374291,0.0008771619,0.000987488,0.001354776,0.0004737274,0.001190007,0.002086062,0.001555863,0.00673787],"category_scores_gemma":[0.06181479,0.0006107449,0.001374934,0.001330083,0.001095465,0.001402444,0.001169382,0.002853708,0.0009475405],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001407145,"about_ca_system_score_gemma":0.001882546,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003922246,"about_ca_topic_score_gemma":0.003447239,"domain_scores_codex":[0.9957432,0.003089623,0.0002076641,0.0003533555,0.0004378846,0.0001682368],"domain_scores_gemma":[0.9463238,0.0474557,0.002166143,0.002000666,0.001710204,0.0003433941],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002301606,0.0001128695,0.004467131,0.0001000767,0.0000728067,0.0001582544,0.0002469119,0.8820576,0.000572374,0.08739276,0.001859971,0.02272916],"study_design_scores_gemma":[0.0000780585,0.00006864387,0.0003253694,0.00003440316,0.00002029539,0.00004918287,0.00003284147,0.9577894,0.000818176,0.03936449,0.001396585,0.00002252334],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03031676,0.0001402072,0.9659995,0.0003640405,0.00006855208,0.0004540903,0.0005706556,0.000602116,0.001484139],"genre_scores_gemma":[0.3766026,0.000540782,0.6126078,0.0004556423,0.00006689229,0.00348267,0.002294208,0.0002826383,0.003666778],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01374291,"threshold_uncertainty_score":0.07268029,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Andrew R. Willan","is_ca":true},{"name":"Eleanor M. Pinto","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5587308049738138,"gpt":0.6187845387614808,"spread":0.06005373378766699,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2538993,0.00278958,0.0084427,0.005413027,0.001588863,0.009079531,0.004238263,0.01007982,0.004927876],"category_scores_gemma":[0.4715581,0.002269035,0.002745816,0.00419071,0.01411737,0.01009675,0.005995966,0.009394662,0.001605662],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007548877,"about_ca_system_score_gemma":0.01306581,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008697406,"about_ca_topic_score_gemma":0.0006673049,"domain_scores_codex":[0.4880716,0.4784879,0.01011296,0.006774519,0.01470176,0.00185132],"domain_scores_gemma":[0.5101001,0.457734,0.01219265,0.01107191,0.00668196,0.002219413],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002815019,0.0002114406,0.001479882,0.002676286,0.001011283,0.0002760562,0.0005277171,0.04480235,0.0002463675,0.795221,0.007363433,0.1433693],"study_design_scores_gemma":[0.001237931,0.0005935722,0.0003101188,0.001060911,0.0002873836,0.0001344085,0.00006181325,0.02628577,0.0002805321,0.9629857,0.006681844,0.00007989151],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.004441577,0.01474317,0.9270771,0.03322632,0.001123221,0.003228072,0.0003702887,0.0002288017,0.01556136],"genre_scores_gemma":[0.2215912,0.01192722,0.7372145,0.009704753,0.002272189,0.01485978,0.0002437704,0.0001472111,0.002039407],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7461007,"threshold_uncertainty_score":0.920075,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Lawrence C. McCandless","is_ca":true},{"name":"Paul Gustafson","is_ca":true},{"name":"Adrian R. Levy","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2203225194206481,"gpt":0.4762013753441512,"spread":0.2558788559235031,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.16287,0.001385169,0.00326694,0.004006319,0.001006635,0.003157167,0.003028102,0.003522776,0.002896078],"category_scores_gemma":[0.4815637,0.0009636616,0.003680089,0.003021321,0.003157869,0.003372162,0.003611556,0.003613611,0.0001784923],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003012592,"about_ca_system_score_gemma":0.002308714,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003945926,"about_ca_topic_score_gemma":0.001653432,"domain_scores_codex":[0.8365991,0.1500759,0.002555783,0.003495183,0.006378083,0.0008960147],"domain_scores_gemma":[0.3980834,0.5747087,0.01162935,0.01094235,0.004079066,0.0005570442],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0006708601,0.00009664701,0.01056792,0.001535167,0.003435847,0.0009178739,0.0007600688,0.5666932,0.0006471625,0.3567455,0.002211173,0.0557185],"study_design_scores_gemma":[0.0001885189,0.0001749072,0.002877858,0.0004877764,0.0008702548,0.0003696584,0.0001155269,0.5058632,0.0005926207,0.4853497,0.003016271,0.00009380565],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01751737,0.002340055,0.9750808,0.002014596,0.0001175295,0.0004227799,0.0002281639,0.0001073074,0.002171356],"genre_scores_gemma":[0.739243,0.002951466,0.2530017,0.001303766,0.0002962563,0.001983278,0.0002667602,0.00009065766,0.0008630661],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.16287,"threshold_uncertainty_score":0.8613495,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Melania Pintilie","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1114147099766471,"gpt":0.460709521559191,"spread":0.349294811582544,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1375142,0.002319782,0.004041575,0.01061008,0.00119419,0.00761741,0.004840997,0.00507532,0.004836836],"category_scores_gemma":[0.3963928,0.001136466,0.003834578,0.0075119,0.004155857,0.005939,0.004699115,0.005284426,0.0007713976],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001865752,"about_ca_system_score_gemma":0.003359849,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001683446,"about_ca_topic_score_gemma":0.001078071,"domain_scores_codex":[0.8442545,0.1279186,0.008531388,0.004872096,0.01352271,0.0009007505],"domain_scores_gemma":[0.3592251,0.596873,0.01804977,0.01848089,0.006310039,0.001061277],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001506484,0.0003517538,0.03594054,0.008025832,0.003150103,0.004095989,0.01151406,0.05320172,0.003272556,0.4905457,0.02258072,0.3658146],"study_design_scores_gemma":[0.0002659672,0.000339383,0.007570455,0.001939999,0.0006193147,0.00193872,0.002429207,0.06237675,0.002365742,0.8804165,0.03948367,0.0002543287],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01801375,0.004515182,0.9624125,0.006039371,0.0006741503,0.0009587769,0.002850536,0.0006357813,0.003899962],"genre_scores_gemma":[0.2000431,0.003594876,0.7856989,0.002593644,0.001402114,0.002168255,0.003105868,0.0003927074,0.001000556],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1375142,"threshold_uncertainty_score":0.7272532,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Gary S. Collins","is_ca":false},{"name":"Emmanuel Ogundimu","is_ca":false},{"name":"Jonathan Cook","is_ca":false},{"name":"Yannick Le Manach","is_ca":true},{"name":"Douglas G. Altman","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6090490742207253,"gpt":0.5329850486829449,"spread":0.07606402553778036,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1710393,0.002825563,0.003210726,0.003971879,0.001252538,0.005826093,0.002374124,0.003255126,0.001776205],"category_scores_gemma":[0.3729456,0.000867778,0.004497413,0.004331924,0.004047245,0.005243704,0.004891267,0.007014605,0.0006662454],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001883448,"about_ca_system_score_gemma":0.002890497,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002745558,"about_ca_topic_score_gemma":0.00274145,"domain_scores_codex":[0.8925464,0.0866133,0.006134266,0.0051504,0.008396836,0.001158817],"domain_scores_gemma":[0.4730465,0.4878442,0.01104025,0.02044895,0.006271353,0.001348739],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.005366056,0.0005530912,0.1864207,0.00246386,0.009841758,0.0005390573,0.001757809,0.4737023,0.003253051,0.02829788,0.003941333,0.2838632],"study_design_scores_gemma":[0.0006781768,0.004829661,0.06508611,0.002078689,0.003133206,0.00122357,0.001187027,0.7559267,0.007873443,0.1505599,0.00673961,0.0006840514],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.2596059,0.0139263,0.7102767,0.006599986,0.000547694,0.00109318,0.001737014,0.0007261062,0.005487072],"genre_scores_gemma":[0.7334692,0.003137388,0.2587851,0.000932119,0.0002419616,0.0009148062,0.001612097,0.000270162,0.0006371347],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1710393,"threshold_uncertainty_score":0.9045531,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Ruth H. Keogh","is_ca":false},{"name":"Pamela A. Shaw","is_ca":false},{"name":"Paul Gustafson","is_ca":true},{"name":"Raymond J. Carroll","is_ca":false},{"name":"Veronika Deffner","is_ca":false},{"name":"Kevin W. Dodd","is_ca":false},{"name":"Helmut Küchenhoff","is_ca":false},{"name":"Janet A. Tooze","is_ca":false},{"name":"Michael P. Wallace","is_ca":true},{"name":"Victor Kipnis","is_ca":false},{"name":"Laurence S. Freedman","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3319225737265194,"gpt":0.4910995069651478,"spread":0.1591769332386284,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04107296,0.003737472,0.003395579,0.007812687,0.001719427,0.005003806,0.008268571,0.01593667,0.03294725],"category_scores_gemma":[0.1241395,0.002741273,0.003650377,0.006735467,0.003169691,0.004075933,0.005021681,0.01111518,0.02903147],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005187134,"about_ca_system_score_gemma":0.02717891,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.05959269,"about_ca_topic_score_gemma":0.04951409,"domain_scores_codex":[0.9597526,0.01685348,0.004564077,0.001662215,0.01597605,0.001191549],"domain_scores_gemma":[0.8965233,0.06185387,0.005866391,0.006502762,0.02746163,0.001791974],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00008232251,0.0001133855,0.001049796,0.001851482,0.00005609439,0.0001447572,0.000321007,0.0030415,0.0003396651,0.03927378,0.882594,0.07113214],"study_design_scores_gemma":[0.0001386939,0.0001314117,0.003251258,0.00450722,0.00007636485,0.0002858614,0.000151024,0.001745588,0.0004918259,0.03203489,0.9570813,0.0001043916],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.003208721,0.05811043,0.3147441,0.1957342,0.03118744,0.01409471,0.1402037,0.008965378,0.2337514],"genre_scores_gemma":[0.01334935,0.08041561,0.564253,0.07881773,0.0110114,0.0210014,0.08651593,0.0040613,0.1405743],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.958927,"threshold_uncertainty_score":0.2172171,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Neil Klar","is_ca":true},{"name":"Allan Donner","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6291099124678506,"gpt":0.5903951121051839,"spread":0.03871480036266672,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7592224,0.003189792,0.01442285,0.007825679,0.003546305,0.01503239,0.01685111,0.01387118,0.008687127],"category_scores_gemma":[0.8520404,0.003495087,0.008537116,0.01331674,0.0252594,0.02081639,0.008705618,0.02487747,0.003584854],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0131016,"about_ca_system_score_gemma":0.04100379,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01031172,"about_ca_topic_score_gemma":0.01094882,"domain_scores_codex":[0.2266357,0.6802186,0.04120589,0.01342903,0.03689036,0.001620416],"domain_scores_gemma":[0.04306266,0.8924441,0.01204868,0.019699,0.02967187,0.003073598],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.003178697,0.0004592709,0.005025617,0.01955009,0.003112853,0.0002845218,0.003198725,0.01253289,0.0003447872,0.2930239,0.06243448,0.5968542],"study_design_scores_gemma":[0.002261056,0.0014871,0.004102049,0.0188261,0.001355597,0.0007618329,0.001664709,0.02952649,0.0004660119,0.8329231,0.1060193,0.0006067169],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.002299597,0.1012385,0.4713482,0.403703,0.01170131,0.00298208,0.0008653067,0.0009692732,0.004892698],"genre_scores_gemma":[0.03071493,0.03482444,0.8564803,0.04693514,0.01574541,0.01309388,0.000527184,0.000587109,0.001091723],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2407776,"threshold_uncertainty_score":0.2969217,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Mei‐Ling Ting Lee","is_ca":false},{"name":"G. À. Whitmore","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03013445253274159,"gpt":0.3339330755395447,"spread":0.303798623006803,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2050592,0.001682499,0.004931843,0.003897263,0.00191448,0.004523454,0.004492934,0.005824177,0.007486513],"category_scores_gemma":[0.6597241,0.001110198,0.003349763,0.004903579,0.007602415,0.007041269,0.004763548,0.006433833,0.001697756],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002209102,"about_ca_system_score_gemma":0.002826592,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005258063,"about_ca_topic_score_gemma":0.0003046882,"domain_scores_codex":[0.7735267,0.1806717,0.009004603,0.0134732,0.02219833,0.00112542],"domain_scores_gemma":[0.3517404,0.6071079,0.01020705,0.02326674,0.006623957,0.001053993],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00271926,0.0002424723,0.01744722,0.005484432,0.001621411,0.0009545673,0.001911591,0.02261817,0.003800031,0.6301864,0.02345006,0.2895643],"study_design_scores_gemma":[0.001571828,0.001979613,0.01110736,0.001320853,0.0007294107,0.001587705,0.0004707729,0.06323206,0.00608895,0.8570251,0.05467336,0.0002128752],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.008290155,0.003994542,0.9712809,0.005696645,0.001298278,0.002177747,0.0008089059,0.0004261417,0.00602676],"genre_scores_gemma":[0.2887715,0.002981258,0.6681011,0.005785738,0.002497052,0.02730529,0.001098087,0.0008980046,0.002561924],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2050592,"threshold_uncertainty_score":0.9803035,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Tim Cole","is_ca":false},{"name":"Sanja Stanojevic","is_ca":false},{"name":"Janet Stocks","is_ca":false},{"name":"Allan L. Coates","is_ca":true},{"name":"J. Hankinson","is_ca":false},{"name":"Angela Wade","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.01751881853312985,"gpt":0.2850192669584043,"spread":0.2675004484252745,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008070083,0.0004707761,0.0009312149,0.001988773,0.001100025,0.00122085,0.001093092,0.001734864,0.001166984],"category_scores_gemma":[0.01779088,0.0003404133,0.001288747,0.003878,0.0008225286,0.001326453,0.001992473,0.001410359,0.0002710758],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001446826,"about_ca_system_score_gemma":0.001165355,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.05107648,"about_ca_topic_score_gemma":0.04747558,"domain_scores_codex":[0.9947208,0.002745727,0.0003761871,0.0006585109,0.0009129841,0.0005858532],"domain_scores_gemma":[0.988552,0.006541682,0.00167753,0.001200941,0.001590401,0.0004374444],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002946458,0.000284525,0.9408535,0.0001886251,0.0001397588,0.009860576,0.007228293,0.001750117,0.0002983449,0.002924887,0.001290753,0.0348861],"study_design_scores_gemma":[0.00002974939,0.0007149508,0.942633,0.0003850132,0.0002153833,0.01897813,0.01509727,0.01059856,0.0009298885,0.001547216,0.008782686,0.0000881626],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9877759,0.002737194,0.005187242,0.0005476645,0.00002289825,0.00009779348,0.0007832579,0.0000328625,0.002815197],"genre_scores_gemma":[0.9903373,0.00177197,0.006263506,0.0001099554,0.000026911,0.00008315886,0.0004931192,0.00002671574,0.0008874094],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05107648,"threshold_uncertainty_score":0.1015583,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Marie‐Pierre Sylvestre","is_ca":true},{"name":"Michał Abrahamowicz","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2270755262068856,"gpt":0.4336685487184045,"spread":0.206593022511519,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01396189,0.0014971,0.002039848,0.001821773,0.0007291731,0.002448532,0.005424262,0.003134058,0.005413376],"category_scores_gemma":[0.02834576,0.001189548,0.002912734,0.002268343,0.002572824,0.002583056,0.002407372,0.003824927,0.0007486801],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002155215,"about_ca_system_score_gemma":0.002327314,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02364122,"about_ca_topic_score_gemma":0.01249448,"domain_scores_codex":[0.9953279,0.002565495,0.0001742532,0.0008372876,0.0004681582,0.0006267706],"domain_scores_gemma":[0.982296,0.01320222,0.001686016,0.001486652,0.0008146168,0.0005144114],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001292992,0.00006000418,0.004327074,0.00005667072,0.0001655538,0.0003235971,0.0002193924,0.9029258,0.0004731421,0.07841695,0.0009841643,0.01191837],"study_design_scores_gemma":[0.0000507795,0.00006377151,0.001899271,0.00002500841,0.00008410385,0.00009190079,0.00004145297,0.941757,0.0001431333,0.05445302,0.001348403,0.00004208352],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05596634,0.0006939931,0.9376453,0.001432591,0.0001487064,0.000155844,0.001190264,0.0003319378,0.002435018],"genre_scores_gemma":[0.8941261,0.001257488,0.07978477,0.0003756485,0.0002512277,0.0007009439,0.001100051,0.0001828962,0.02222088],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.02364122,"threshold_uncertainty_score":0.07383847,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Nico Nagelkerke","is_ca":true},{"name":"V. Fidler","is_ca":false},{"name":"Roos Bernsen","is_ca":false},{"name":"Martien W. Borgdorff","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2897409135338255,"gpt":0.5720931192616544,"spread":0.2823522057278289,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4671667,0.004079015,0.01291666,0.007541639,0.001307113,0.006231173,0.005770726,0.01114533,0.004301752],"category_scores_gemma":[0.7364189,0.003865177,0.01011864,0.008388656,0.009720949,0.008364313,0.005010463,0.009223339,0.001020137],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004609292,"about_ca_system_score_gemma":0.007826042,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002074359,"about_ca_topic_score_gemma":0.001322898,"domain_scores_codex":[0.2924678,0.6663992,0.01815315,0.009995091,0.01180628,0.001178416],"domain_scores_gemma":[0.1542044,0.8043857,0.02354878,0.01465719,0.002617696,0.0005862531],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.009097969,0.0007565379,0.01848144,0.02499821,0.02772267,0.00170284,0.00241937,0.1866444,0.001169532,0.4153221,0.006344988,0.30534],"study_design_scores_gemma":[0.004593182,0.002670834,0.003192597,0.002905105,0.005243473,0.0005121393,0.0001967368,0.1659033,0.001334171,0.8043222,0.008910054,0.0002161842],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.008835229,0.01225267,0.968828,0.003233958,0.0008413474,0.003873531,0.0003565526,0.0004388104,0.001339921],"genre_scores_gemma":[0.2296879,0.007883799,0.7375632,0.003399708,0.001130112,0.01817263,0.0007097764,0.000160563,0.001292325],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5328333,"threshold_uncertainty_score":0.6570783,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1183073088240789,"gpt":0.4368427016114135,"spread":0.3185353927873346,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04429983,0.001908723,0.002176515,0.006865692,0.000596438,0.001533744,0.001508949,0.001436658,0.0005729737],"category_scores_gemma":[0.08113194,0.0006781459,0.002478045,0.003767943,0.0006928474,0.002683205,0.001602027,0.002051917,0.0003255075],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00117708,"about_ca_system_score_gemma":0.002283043,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0121846,"about_ca_topic_score_gemma":0.01029122,"domain_scores_codex":[0.9677253,0.02827151,0.0007596565,0.0009085169,0.001904718,0.000430193],"domain_scores_gemma":[0.8966647,0.09106693,0.003714897,0.002375723,0.00508479,0.001092923],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.006199951,0.0007636962,0.1848536,0.000863904,0.003513039,0.0002323017,0.00102408,0.4880327,0.0006405436,0.007083002,0.003888178,0.3029049],"study_design_scores_gemma":[0.0001353612,0.0009911947,0.02006231,0.0001232284,0.0002353832,0.00009605486,0.0001740187,0.971915,0.0001324834,0.005515409,0.00054143,0.00007817632],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7057297,0.01430809,0.2714722,0.002650687,0.0004022298,0.0004348328,0.001093262,0.0009278963,0.002981069],"genre_scores_gemma":[0.9009328,0.003975611,0.09269816,0.0002283575,0.0002333058,0.0002744752,0.0009268908,0.0001486131,0.0005817569],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04429983,"threshold_uncertainty_score":0.2342827,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Jason P. Fine","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8315391831852751,"gpt":0.7107421776711227,"spread":0.1207970055141524,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3513434,0.005622614,0.02183607,0.01860244,0.001670889,0.01033077,0.01192326,0.009319861,0.01031818],"category_scores_gemma":[0.5422609,0.004306477,0.02338278,0.01813901,0.005504664,0.01613263,0.005134323,0.01334883,0.003187943],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01346814,"about_ca_system_score_gemma":0.04762144,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01037487,"about_ca_topic_score_gemma":0.01738652,"domain_scores_codex":[0.7053757,0.1928042,0.07002151,0.00586092,0.02467005,0.001267606],"domain_scores_gemma":[0.2374314,0.6514736,0.04511706,0.01325145,0.05076016,0.001966418],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0004559408,0.00009593748,0.0009696681,0.3755848,0.008218248,0.0001611066,0.0008050808,0.001453457,0.0001885705,0.01759632,0.04566472,0.5488063],"study_design_scores_gemma":[0.001216689,0.0003060124,0.00166694,0.7318075,0.02013821,0.0003647647,0.0005884843,0.00355542,0.0004002464,0.0559381,0.1836321,0.000385555],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0001365877,0.9454271,0.02494617,0.02348626,0.002803264,0.001763584,0.000404929,0.0003140544,0.0007181158],"genre_scores_gemma":[0.0051692,0.8178201,0.1543905,0.01087461,0.002311587,0.008116662,0.0006504463,0.0001480019,0.0005188343],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.6486566,"threshold_uncertainty_score":0.7999091,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Richard M. Nixon","is_ca":false},{"name":"Nick Bansback","is_ca":true},{"name":"Alan Brennan","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1452751148105242,"gpt":0.4591348507660336,"spread":0.3138597359555093,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1765714,0.00534715,0.01409376,0.01290789,0.0007980403,0.005786788,0.005661392,0.004038781,0.005974326],"category_scores_gemma":[0.2701223,0.002435141,0.04110869,0.01061069,0.001315757,0.004351894,0.003830329,0.005665165,0.0007165191],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003606623,"about_ca_system_score_gemma":0.003310202,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003484391,"about_ca_topic_score_gemma":0.003489045,"domain_scores_codex":[0.6938735,0.2852209,0.008253265,0.006765189,0.005315279,0.0005719929],"domain_scores_gemma":[0.7172419,0.2608842,0.008715756,0.01055804,0.002168039,0.000432037],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.005457101,0.0001846281,0.00607146,0.05274395,0.7004352,0.0005265838,0.0004813086,0.07927027,0.000788894,0.03551523,0.0082266,0.1102988],"study_design_scores_gemma":[0.007329436,0.003212956,0.006324171,0.01014114,0.4618016,0.0006451927,0.0003353255,0.2393362,0.003253161,0.2324957,0.03434834,0.0007767325],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","genre_codex":"methods","genre_gemma":"review","genre_scores_codex":[0.0103747,0.1427235,0.8267446,0.002987561,0.003391569,0.005190162,0.003863059,0.002486928,0.002237841],"genre_scores_gemma":[0.1483975,0.02490475,0.8024248,0.001772129,0.001015743,0.0172187,0.001993714,0.0005846937,0.001688016],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.1765714,"threshold_uncertainty_score":0.9338101,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Aurélien Latouche","is_ca":false},{"name":"Jason P. Fine","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2756750443033474,"gpt":0.4708426292700219,"spread":0.1951675849666745,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01199275,0.001342431,0.002932295,0.004783782,0.0003318299,0.0021992,0.001962015,0.002108575,0.004987721],"category_scores_gemma":[0.04326431,0.000820516,0.004896664,0.007907396,0.000848677,0.0023252,0.0008726228,0.002515803,0.001279458],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001904912,"about_ca_system_score_gemma":0.004843115,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006297919,"about_ca_topic_score_gemma":0.005723043,"domain_scores_codex":[0.9916428,0.004507151,0.001535216,0.0007480904,0.001441807,0.0001248242],"domain_scores_gemma":[0.9571891,0.03816672,0.002017598,0.0004377612,0.00202445,0.0001643428],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003074879,0.0001030334,0.003479604,0.1408321,0.002808913,0.0004380919,0.0004406197,0.005956118,0.0003290259,0.02605133,0.04148019,0.7777736],"study_design_scores_gemma":[0.0001480986,0.0004543265,0.008825173,0.1318525,0.007154505,0.00282291,0.0003072884,0.007570779,0.0006302115,0.03172141,0.8082232,0.0002896828],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0002415353,0.9921955,0.005271917,0.001057227,0.0003252532,0.00002857615,0.0001892568,0.00001934813,0.0006713343],"genre_scores_gemma":[0.004464666,0.9879238,0.005253375,0.001023256,0.0006491804,0.00009590162,0.0002469714,0.00002806671,0.000314852],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9880072,"threshold_uncertainty_score":0.06342447,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Jason P. Fine","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2384664040620321,"gpt":0.4583996193977982,"spread":0.2199332153357661,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04863184,0.001436908,0.002671584,0.004887186,0.0008591363,0.002422837,0.003223869,0.002302758,0.00674805],"category_scores_gemma":[0.1402349,0.001348439,0.0033089,0.006631045,0.002387093,0.002691287,0.003825925,0.004714081,0.001124426],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001827212,"about_ca_system_score_gemma":0.004349089,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004160546,"about_ca_topic_score_gemma":0.002585899,"domain_scores_codex":[0.9536024,0.04015297,0.001277481,0.002049778,0.00252634,0.000390849],"domain_scores_gemma":[0.9225355,0.06704519,0.003771637,0.004637277,0.001638419,0.000372027],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003073057,0.0002304638,0.01294104,0.000959094,0.001481025,0.0004607292,0.0004223752,0.2120805,0.0005181602,0.5234457,0.01101737,0.2361364],"study_design_scores_gemma":[0.0001579486,0.0001154555,0.002073309,0.0001922797,0.0001377409,0.0002254706,0.00004824457,0.5181597,0.0004674872,0.4695182,0.008840635,0.00006357118],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.001168015,0.0004890441,0.9971131,0.0002650995,0.00005880431,0.0002051022,0.0001053991,0.0001911825,0.0004042346],"genre_scores_gemma":[0.07508327,0.002011346,0.9173923,0.0002939723,0.0004179882,0.002275823,0.000695556,0.000231573,0.001598205],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9513682,"threshold_uncertainty_score":0.2571929,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"William A. Ghali","is_ca":true},{"name":"Jack V. Tu","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5450433319250734,"gpt":0.5321256663837163,"spread":0.01291766554135709,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07162116,0.003231179,0.00303204,0.00669043,0.0008423296,0.002765519,0.004741076,0.002009323,0.002363903],"category_scores_gemma":[0.1407545,0.00148015,0.00806969,0.006737764,0.001003042,0.004051266,0.002209914,0.003708849,0.000853183],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003854602,"about_ca_system_score_gemma":0.003656747,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03279937,"about_ca_topic_score_gemma":0.02178242,"domain_scores_codex":[0.9443202,0.04833979,0.001356786,0.002138342,0.002748752,0.001096085],"domain_scores_gemma":[0.774725,0.2068212,0.005804945,0.005441033,0.006478145,0.0007297062],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.004755767,0.0008966532,0.1235605,0.001171193,0.006838765,0.0002919804,0.0008241464,0.678712,0.0004743247,0.02829622,0.006466601,0.1477119],"study_design_scores_gemma":[0.0003329308,0.001097135,0.01974322,0.0002333884,0.0008404417,0.0001487519,0.0004614546,0.9630321,0.000231356,0.01211113,0.00161431,0.0001537734],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3656444,0.007835736,0.6096941,0.004132417,0.0004938952,0.001463691,0.003126333,0.001827068,0.005782469],"genre_scores_gemma":[0.7254409,0.005480292,0.2580738,0.0005935769,0.0002844155,0.002051379,0.004177182,0.0005627392,0.003335751],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.07162116,"threshold_uncertainty_score":0.3787734,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Karen Leffondré","is_ca":true},{"name":"Michał Abrahamowicz","is_ca":true},{"name":"Yongling Xiao","is_ca":true},{"name":"Jack Siemiatycki","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05194383189967641,"gpt":0.3575085325930135,"spread":0.3055647006933371,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008474787,0.001171204,0.001914388,0.002078319,0.0005027829,0.001142397,0.001489737,0.0008916958,0.0008179024],"category_scores_gemma":[0.02667758,0.0004933874,0.001903323,0.003069831,0.0007728375,0.0009267653,0.001223791,0.0010421,0.0001252169],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001845386,"about_ca_system_score_gemma":0.002081751,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.06432156,"about_ca_topic_score_gemma":0.04271337,"domain_scores_codex":[0.9980253,0.00118602,0.00009613569,0.0003483378,0.0002281561,0.0001160783],"domain_scores_gemma":[0.9823292,0.01471891,0.001147011,0.0008168314,0.0006559248,0.0003321964],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0004633338,0.0001567222,0.1113153,0.0001748929,0.001209663,0.0003200606,0.000446721,0.7811418,0.001323649,0.01195604,0.0006977302,0.09079408],"study_design_scores_gemma":[0.00002928364,0.0001880622,0.01742512,0.00001457791,0.0001139549,0.0000836171,0.00004273714,0.9740849,0.000260228,0.007134887,0.0005819264,0.00004064591],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4009954,0.001509202,0.5937659,0.0006976315,0.00005160148,0.0002642708,0.0008078245,0.0005561002,0.001351994],"genre_scores_gemma":[0.871875,0.0008171824,0.1247599,0.00008972429,0.0000601579,0.0001812942,0.0007613877,0.00005607601,0.001399275],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.06432156,"threshold_uncertainty_score":0.1278943,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","authors":[{"name":"Willi Sauerbrei","is_ca":false},{"name":"Michał Abrahamowicz","is_ca":true},{"name":"Douglas G. Altman","is_ca":false},{"name":"Saskia le Cessie","is_ca":false},{"name":"James R. Carpenter","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8755938676556354,"gpt":0.6135030834304239,"spread":0.2620907842252115,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5997813,0.002794118,0.00417282,0.01523635,0.002793467,0.01833935,0.008295348,0.0107397,0.00400746],"category_scores_gemma":[0.6437423,0.002593274,0.00640218,0.01095081,0.013801,0.01450614,0.03375689,0.02477838,0.002251358],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01230229,"about_ca_system_score_gemma":0.1083373,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00433046,"about_ca_topic_score_gemma":0.006466005,"domain_scores_codex":[0.3701505,0.507157,0.04951082,0.009240393,0.06025872,0.00368254],"domain_scores_gemma":[0.1374005,0.7038296,0.03305851,0.03795265,0.0703753,0.01738334],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004732905,0.0001988907,0.00192124,0.0132145,0.001172517,0.0004046501,0.007833522,0.002989017,0.0007014501,0.2701397,0.2698875,0.4310637],"study_design_scores_gemma":[0.0006049604,0.0002553334,0.001488122,0.03272739,0.0006076242,0.0003349915,0.001425595,0.003613395,0.000770407,0.3570108,0.6009552,0.0002062195],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.002082877,0.05546195,0.290276,0.6112762,0.01956657,0.003138135,0.001164003,0.003367967,0.01366637],"genre_scores_gemma":[0.01205698,0.02022761,0.8947452,0.05498524,0.007174719,0.006468165,0.00119971,0.001034974,0.002107276],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4002187,"threshold_uncertainty_score":0.493541,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}