{"meta":{"query_hash":"65f19e95c375","filters":{"venue":"Pharmaceutical Statistics"},"cohort_total":55,"direct_labels_cover":2,"predictions_cover":55,"exported":55,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/65f19e95c375","api":"https://metacan.xera.ac/api/v1/cohort?venue=Pharmaceutical+Statistics"},"results":[{"id":"W1905891062","doi":"10.1002/pst.1704","title":"Bayesian inference in two‐arm trials using relative belief ratios","year":2015,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; University of Manitoba","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Bayesian probability; Inference; Statistical inference; Bayesian inference; Computer science; Artificial intelligence; Bayesian statistics; Fiducial inference; Measure (data warehouse); Machine learning; Frequentist inference; Econometrics; Statistics; Mathematics; Data mining","score_opus":0.8804641098560438,"score_gpt":0.7012657224618299,"score_spread":0.1791983873942139,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1905891062","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0017600701,0.0007938086,0.9959997,0.000532273,0.000059362268,0.00012417798,0.00005530772,0.00011902343,0.0005562444],"genre_scores_gemma":[0.1472335,0.0020103932,0.84659976,0.00078355294,0.00053578994,0.0013087699,0.00022491402,0.00013728961,0.001166079],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.81968695,0.15813395,0.005580303,0.006772866,0.008652856,0.0011730659],"domain_scores_gemma":[0.49336314,0.47696942,0.013159024,0.011396501,0.0042385515,0.0008733242],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.19699831,0.003274769,0.0072777313,0.006148022,0.0009956638,0.00641595,0.0066589005,0.0048775743,0.0050639054],"category_scores_gemma":[0.44064647,0.0026401782,0.005496529,0.005342083,0.0062162797,0.009639719,0.0049415245,0.009507831,0.0009463315],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00080843293,0.000116335366,0.0015662769,0.0012005578,0.001472737,0.00029257144,0.0005040916,0.19484662,0.00052777893,0.7051605,0.0019661866,0.091538034],"study_design_scores_gemma":[0.0002909316,0.00018353954,0.0003094533,0.00017029046,0.00024626116,0.0001082367,0.0000258595,0.35701942,0.000614592,0.6394566,0.0015085859,0.00006623562],"about_ca_topic_score_codex":0.0024271612,"about_ca_topic_score_gemma":0.0015998572,"teacher_disagreement_score":0.8030017,"about_ca_system_score_codex":0.0033933292,"about_ca_system_score_gemma":0.003413162,"threshold_uncertainty_score":0.99024403},"labels":[],"label_agreement":null},{"id":"W1979781274","doi":"10.1002/pst.200","title":"Power and sample size considerations in clinical trials with competing risk endpoints","year":2006,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Statistics Canada","funders":"","keywords":"Sample size determination; Accrual; Event (particle physics); Weibull distribution; Clinical trial; Econometrics; Parametric statistics; Clinical endpoint; Sample (material); Statistics; Actuarial science; Computer science; Medicine; Economics; Mathematics; Internal medicine; Accounting","score_opus":0.6618107269419979,"score_gpt":0.6408193264275303,"score_spread":0.020991400514467684,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1979781274","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0047566127,0.011455918,0.96512824,0.00904941,0.0007798724,0.0016932217,0.00013961928,0.00019548065,0.0068015554],"genre_scores_gemma":[0.20865755,0.010390241,0.7523168,0.0074419323,0.003549047,0.014064424,0.00027422464,0.00047195607,0.0028338411],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7083338,0.26394016,0.0062866183,0.0036842013,0.016782943,0.0009722554],"domain_scores_gemma":[0.35193992,0.6285859,0.0057753883,0.007831225,0.005248216,0.00061941636],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.24273075,0.002955537,0.0047429954,0.0046851346,0.001042834,0.003863648,0.00442115,0.007812739,0.0034207725],"category_scores_gemma":[0.55114275,0.0014913128,0.0027901821,0.0035533751,0.006546679,0.006910174,0.00368984,0.0085321115,0.0010238537],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004087062,0.00020466885,0.002782595,0.0035631247,0.00083079084,0.001238943,0.0014855502,0.04221853,0.0026255513,0.6877211,0.008436469,0.2448057],"study_design_scores_gemma":[0.002214097,0.0023031288,0.0015058212,0.0017102103,0.00048587396,0.0014263365,0.00024192815,0.10617693,0.0032985639,0.86047375,0.020022372,0.00014095566],"about_ca_topic_score_codex":0.0005004308,"about_ca_topic_score_gemma":0.00042257496,"teacher_disagreement_score":0.75726926,"about_ca_system_score_codex":0.0020455886,"about_ca_system_score_gemma":0.0034126167,"threshold_uncertainty_score":0.9338478},"labels":[],"label_agreement":null},{"id":"W1980598266","doi":"10.1002/pst.338","title":"Impact of baseline ECG collection on the planning, analysis and interpretation of ‘thorough’ QT trials","year":2008,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Cardiac electrophysiology and arrhythmias","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"Health Canada","keywords":"Baseline (sea); Medicine; QT interval; Replicate; Time point; Statistics; Internal medicine; Mathematics","score_opus":0.08796446487793805,"score_gpt":0.43880973624290165,"score_spread":0.3508452713649636,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1980598266","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.031457365,0.02899195,0.8650373,0.029844709,0.0036545512,0.015217608,0.0037229988,0.0022350939,0.019838398],"genre_scores_gemma":[0.2350663,0.010163862,0.7206643,0.013129009,0.0023241849,0.014556611,0.0012944996,0.0009812269,0.0018198928],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.32104522,0.59924114,0.02921731,0.0075200917,0.041916538,0.0010597679],"domain_scores_gemma":[0.18646057,0.6932843,0.050313756,0.036985155,0.03155851,0.0013977612],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.48400852,0.0014216988,0.003581359,0.0017525554,0.001365818,0.0048625856,0.0029482027,0.003394905,0.0032702256],"category_scores_gemma":[0.64204,0.0013983427,0.0030927716,0.0030062618,0.0028643627,0.003724535,0.0031728863,0.0062974114,0.0014566144],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.014061754,0.0006241255,0.022057652,0.010614069,0.003574279,0.0009834588,0.004133576,0.017810084,0.0062595215,0.032148108,0.05388021,0.8338532],"study_design_scores_gemma":[0.013701193,0.030037684,0.2357358,0.025018452,0.010533751,0.004904725,0.0024214638,0.08690588,0.046008445,0.18933396,0.35363752,0.0017612056],"about_ca_topic_score_codex":0.0029137156,"about_ca_topic_score_gemma":0.0041710804,"teacher_disagreement_score":0.51599145,"about_ca_system_score_codex":0.0036672421,"about_ca_system_score_gemma":0.008479051,"threshold_uncertainty_score":0.6363094},"labels":[],"label_agreement":null},{"id":"W2019978533","doi":"10.1002/pst.244","title":"Assessment of the Gould‐Shih procedure for sample size re‐estimation","year":2006,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Women's Health Research Institute","funders":"","keywords":"Standard deviation; Sample size determination; Statistics; Mathematics; Estimator; Standard error; Nominal level; Confidence interval","score_opus":0.5327868823583967,"score_gpt":0.6279960254321367,"score_spread":0.09520914307374007,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2019978533","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009628154,0.0021631399,0.9759675,0.0031031445,0.00057662826,0.003046959,0.0002209442,0.0005828109,0.0047106366],"genre_scores_gemma":[0.12740429,0.0009402921,0.8612235,0.0020109375,0.00024466022,0.0060520475,0.00018143308,0.00037459083,0.0015682937],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.7076776,0.25528768,0.0063730828,0.004884663,0.024824288,0.00095268886],"domain_scores_gemma":[0.31844965,0.6277888,0.012041748,0.025126453,0.015461723,0.0011317253],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.29206705,0.0018423278,0.00350974,0.0043140734,0.001974338,0.0025971269,0.003965171,0.006221835,0.007046809],"category_scores_gemma":[0.5780246,0.0010926146,0.004413056,0.004055732,0.006574908,0.0036407837,0.0054497193,0.006109966,0.00077178655],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0041293255,0.00036718592,0.011755691,0.0031753222,0.0037961053,0.0004952377,0.0023584957,0.06921575,0.0025637131,0.4422564,0.019069728,0.4408169],"study_design_scores_gemma":[0.0029789836,0.005311289,0.012810754,0.002203067,0.0020853467,0.0013707904,0.00073708163,0.5319438,0.008433899,0.368279,0.063115895,0.00073008524],"about_ca_topic_score_codex":0.0021831403,"about_ca_topic_score_gemma":0.0019722513,"teacher_disagreement_score":0.70793295,"about_ca_system_score_codex":0.0032340765,"about_ca_system_score_gemma":0.008778581,"threshold_uncertainty_score":0.87300736},"labels":[],"label_agreement":null},{"id":"W2024015933","doi":"10.1002/pst.294","title":"Sequential design approaches for bioequivalence studies with crossover designs","year":2007,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":95,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Theratechnologies (Canada)","funders":"","keywords":"Sample size determination; Bioequivalence; Crossover; Variance (accounting); Crossover study; Statistics; Type I and type II errors; Statistical power; Clinical study design; A priori and a posteriori; Computer science; Mathematics; Econometrics; Clinical trial; Medicine; Machine learning; Placebo; Pharmacology; Pharmacokinetics","score_opus":0.9499294280480977,"score_gpt":0.6769455700599387,"score_spread":0.27298385798815894,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2024015933","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0008692722,0.0002738079,0.99565244,0.0001025662,0.00017185634,0.0021394142,0.000054672244,0.00011710602,0.0006188018],"genre_scores_gemma":[0.018281154,0.00044638963,0.9669777,0.00020452219,0.00016438428,0.013159726,0.0000736445,0.00006171911,0.0006306703],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8137314,0.16292134,0.0047673522,0.005721753,0.012068075,0.0007899709],"domain_scores_gemma":[0.8719517,0.10963437,0.005076703,0.007642461,0.005108746,0.00058597897],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.12751198,0.0030692036,0.0037740094,0.0035717802,0.0011318555,0.0021286902,0.003797135,0.0032761327,0.01313451],"category_scores_gemma":[0.18069798,0.00156412,0.0039506946,0.004106804,0.0037894154,0.0029477053,0.0027619686,0.00622269,0.0016750029],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002816554,0.00061659433,0.0012157932,0.0032127153,0.0014616901,0.00021656512,0.0016953814,0.054782152,0.0021043357,0.6412054,0.0047301836,0.2859427],"study_design_scores_gemma":[0.006112018,0.007127528,0.0012116439,0.00073024275,0.0007821782,0.00025963815,0.00016889225,0.257008,0.0033411705,0.6835752,0.039462607,0.00022097999],"about_ca_topic_score_codex":0.0011327628,"about_ca_topic_score_gemma":0.0011222418,"teacher_disagreement_score":0.872488,"about_ca_system_score_codex":0.0025590153,"about_ca_system_score_gemma":0.0054191914,"threshold_uncertainty_score":0.67435586},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2030706697","doi":"10.1002/pst.368","title":"The Rheumatoid Arthritis Drug Development Model: a case study in Bayesian clinical trial simulation","year":2009,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes","funders":"AstraZeneca","keywords":"Drug development; Clinical trial; Medicine; Rheumatoid arthritis; Bayesian probability; Outcome (game theory); Population; Intensive care medicine; Drug; Risk analysis (engineering); Medical physics; Computer science; Internal medicine; Pharmacology; Artificial intelligence; Mathematics","score_opus":0.6679110266087562,"score_gpt":0.6478830317170668,"score_spread":0.020027994891689382,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2030706697","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.48033154,0.004844137,0.42609927,0.017439548,0.00033235134,0.0019162917,0.0028109513,0.0005785185,0.065647416],"genre_scores_gemma":[0.9058403,0.001559361,0.08107905,0.00076511246,0.000105452535,0.001219323,0.0005558307,0.00006230238,0.008813242],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9926267,0.0057209674,0.00020504871,0.0003564808,0.0006316686,0.00045908196],"domain_scores_gemma":[0.9195435,0.07446964,0.0019916394,0.00086793676,0.0018128775,0.00131442],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012768201,0.0011587457,0.0017404527,0.0014278705,0.0009927904,0.0027680893,0.0021725076,0.004884664,0.0064129657],"category_scores_gemma":[0.03690444,0.00092611654,0.0014531908,0.0019285593,0.0020485509,0.0013961212,0.0018361037,0.003005683,0.00051413925],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00038460395,0.00025590183,0.0035344388,0.00012364765,0.00007890322,0.00088804064,0.00019400798,0.94409865,0.00016759531,0.04308526,0.0013603396,0.0058286604],"study_design_scores_gemma":[0.00032587623,0.00019782194,0.0006900803,0.0000642941,0.00005195564,0.00019426378,0.00011822676,0.97576034,0.00016134474,0.020507867,0.0018929626,0.000034979174],"about_ca_topic_score_codex":0.024567584,"about_ca_topic_score_gemma":0.013807067,"teacher_disagreement_score":0.024567584,"about_ca_system_score_codex":0.0037032044,"about_ca_system_score_gemma":0.004131678,"threshold_uncertainty_score":0.06752545},"labels":[],"label_agreement":null},{"id":"W2042319973","doi":"10.1002/pst.443","title":"On assessing the presence of evaluation‐time bias in progression‐free survival in randomized trials","year":2010,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Randomized controlled trial; Computer science; Test (biology); Log-rank test; Randomized experiment; Progression-free survival; Statistics; Econometrics; Overall survival; Survival analysis; Medical physics; Medicine; Oncology; Mathematics; Internal medicine","score_opus":0.7951823931859413,"score_gpt":0.6931044078284545,"score_spread":0.1020779853574868,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2042319973","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.074144065,0.014947244,0.88253665,0.0141162295,0.0006249967,0.0013807964,0.00025582573,0.0003702614,0.011623819],"genre_scores_gemma":[0.7227518,0.0055658533,0.26378334,0.0042525046,0.00080806867,0.0014507304,0.00018817163,0.00017685839,0.0010226768],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.3997254,0.56504244,0.010393626,0.005057998,0.018477544,0.0013030051],"domain_scores_gemma":[0.03374223,0.9466682,0.010929172,0.006096372,0.0022945227,0.00026951477],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.47936657,0.0021838872,0.0035005365,0.003181541,0.0010274906,0.0053873216,0.002660385,0.0047289906,0.002506437],"category_scores_gemma":[0.8478593,0.00090635894,0.0032153744,0.0042000897,0.010933794,0.011624986,0.0034485727,0.0057954066,0.0004181407],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008613166,0.000353588,0.03662808,0.0048673116,0.0036211242,0.0007854769,0.002517009,0.12569664,0.0015964574,0.58858943,0.0036776224,0.22305416],"study_design_scores_gemma":[0.0016606523,0.0031403569,0.0068967817,0.0024811004,0.0013232843,0.0012084114,0.00034843409,0.25453395,0.0040088897,0.7174577,0.0066655627,0.00027488483],"about_ca_topic_score_codex":0.0013399618,"about_ca_topic_score_gemma":0.0006819099,"teacher_disagreement_score":0.52063346,"about_ca_system_score_codex":0.004153239,"about_ca_system_score_gemma":0.0061910325,"threshold_uncertainty_score":0.6420337},"labels":[],"label_agreement":null},{"id":"W2049564589","doi":"10.1002/pst.289","title":"Comparing two independent incidence rates using conditional and unconditional exact tests","year":2007,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western Forest Products","funders":"","keywords":"Exact statistics; Mathematics; Statistics; Exact test; Wald test; Type I and type II errors; Score test; Poisson distribution; Likelihood-ratio test; Binomial distribution; p-value; Chi-square test; Negative binomial distribution; Binomial test; Nominal level; Statistic; Test statistic; Statistical hypothesis testing; Confidence interval","score_opus":0.7041238095391115,"score_gpt":0.6470359725829649,"score_spread":0.05708783695614661,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2049564589","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.11948941,0.001459183,0.86476386,0.00076696754,0.00047374098,0.0010924088,0.0027548247,0.00081214256,0.008387466],"genre_scores_gemma":[0.728619,0.0010132067,0.26215518,0.00055225275,0.00037027485,0.0024515444,0.0027332734,0.00017582126,0.0019294387],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8825463,0.08509575,0.0069657103,0.0074997996,0.015956009,0.0019364662],"domain_scores_gemma":[0.56924397,0.3772657,0.021793038,0.022592247,0.007713152,0.0013919164],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09601701,0.00081517757,0.002427317,0.0050560962,0.0004966462,0.0023698742,0.002903289,0.001865599,0.01153271],"category_scores_gemma":[0.3702819,0.0004837105,0.003161478,0.0046413415,0.0028378316,0.0037820532,0.0029029343,0.0026663053,0.001116588],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.007222709,0.00062977197,0.19329771,0.0031890736,0.005617892,0.0010259378,0.0014105798,0.031374376,0.002532867,0.2207238,0.008655698,0.5243195],"study_design_scores_gemma":[0.0014447757,0.006478448,0.23374134,0.001401632,0.0052005304,0.0036077688,0.0013892628,0.26320878,0.015277361,0.44657835,0.020861285,0.0008105264],"about_ca_topic_score_codex":0.00067726785,"about_ca_topic_score_gemma":0.00055997056,"teacher_disagreement_score":0.09601701,"about_ca_system_score_codex":0.0009784712,"about_ca_system_score_gemma":0.0018868076,"threshold_uncertainty_score":0.5077926},"labels":[],"label_agreement":null},{"id":"W2064381090","doi":"10.1002/pst.111","title":"Carry‐over in cross‐over trials in bioequivalence: theoretical concerns and empirical evidence","year":2004,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":50,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Pfizer (Canada)","funders":"","keywords":"Carry (investment); Bioequivalence; Estimator; Context (archaeology); Econometrics; Computer science; Test (biology); Order (exchange); Mathematics; Statistics; Economics; Medicine; Pharmacology","score_opus":0.8455461255403379,"score_gpt":0.7076831575087411,"score_spread":0.1378629680315968,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2064381090","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.24061982,0.13439037,0.5540012,0.04154741,0.0019923574,0.0020651307,0.0006581586,0.00074466824,0.02398084],"genre_scores_gemma":[0.9301066,0.009972943,0.04750623,0.0077495007,0.0017039028,0.0008856012,0.00021086542,0.000103954204,0.0017605314],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6329117,0.31305638,0.014197526,0.009958212,0.028374558,0.0015016189],"domain_scores_gemma":[0.090273075,0.83996207,0.036579553,0.026908962,0.0056254277,0.0006508993],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.392063,0.0017662344,0.0039679743,0.0023879951,0.0012060301,0.0055642542,0.0025988459,0.008174349,0.0051989853],"category_scores_gemma":[0.61124295,0.0012150382,0.0064014303,0.0029571252,0.015479896,0.008921876,0.0036911534,0.00543928,0.00052975916],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.051786784,0.0014200684,0.048239905,0.014051192,0.029212708,0.0017495183,0.003872057,0.029016567,0.0038232787,0.39504638,0.006273116,0.41550845],"study_design_scores_gemma":[0.0073304367,0.020951964,0.036809664,0.005853319,0.021228068,0.0028812462,0.0006277584,0.0649729,0.009313451,0.80209225,0.027459208,0.00047977784],"about_ca_topic_score_codex":0.0011799021,"about_ca_topic_score_gemma":0.0006703521,"teacher_disagreement_score":0.607937,"about_ca_system_score_codex":0.0031774624,"about_ca_system_score_gemma":0.002595361,"threshold_uncertainty_score":0.7496946},"labels":[],"label_agreement":null},{"id":"W2104082640","doi":"10.1002/pst.483","title":"Additional results for ‘Sequential design approaches for bioequivalence studies with crossover designs’","year":2011,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":46,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Theratechnologies (Canada)","funders":"","keywords":"Bioequivalence; Crossover; Sample size determination; Statistics; Crossover study; Type I and type II errors; Econometrics; Mathematics; Computer science; Medicine; Pharmacology; Artificial intelligence; Pharmacokinetics","score_opus":0.8859019112477156,"score_gpt":0.5762477195149607,"score_spread":0.3096541917327549,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2104082640","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0027928895,0.01309426,0.9052243,0.013272235,0.0075499616,0.0017781118,0.0018444968,0.0009258583,0.053517763],"genre_scores_gemma":[0.05487785,0.013240094,0.8710603,0.012575072,0.0054513817,0.0063033137,0.002574557,0.0010071481,0.032910287],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9185553,0.056944307,0.0033642284,0.0028020283,0.01713886,0.001195305],"domain_scores_gemma":[0.6730087,0.284691,0.0042459276,0.011139587,0.025820296,0.0010944617],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.10602961,0.002756048,0.0029662016,0.004615356,0.0014154399,0.001802661,0.0022729782,0.0036649357,0.064219244],"category_scores_gemma":[0.2563116,0.0012308664,0.0066641252,0.0058428347,0.0022933085,0.004404686,0.0021623909,0.006972219,0.00974714],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020718528,0.0013412564,0.00087997195,0.006080028,0.00075796206,0.00035381305,0.00055799144,0.010422769,0.002452469,0.32653773,0.16170092,0.48684323],"study_design_scores_gemma":[0.0022293348,0.0038827115,0.0040534404,0.0029899355,0.0009985289,0.0008330821,0.00015278335,0.04676232,0.0109756375,0.5031928,0.42353082,0.0003986343],"about_ca_topic_score_codex":0.0027487564,"about_ca_topic_score_gemma":0.0019402464,"teacher_disagreement_score":0.10602961,"about_ca_system_score_codex":0.0029713644,"about_ca_system_score_gemma":0.0048500793,"threshold_uncertainty_score":0.5607449},"labels":[],"label_agreement":null},{"id":"W2107812146","doi":"10.1002/pst.351","title":"Using short‐term evidence to predict six‐month outcomes in clinical trials of signs and symptoms in rheumatoid arthritis","year":2008,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Rheumatoid Arthritis Research and Therapies","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; St. Paul's Hospital","funders":"","keywords":"Rheumatoid arthritis; Medicine; Clinical trial; Logistic regression; Bayesian probability; Randomized controlled trial; Internal medicine; Physical therapy; Statistics; Mathematics","score_opus":0.31636016439954256,"score_gpt":0.5100445254687855,"score_spread":0.19368436106924297,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2107812146","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3057835,0.03437543,0.63111126,0.011104369,0.00047916346,0.0034587432,0.0062553966,0.0012652376,0.0061668176],"genre_scores_gemma":[0.9154876,0.004807436,0.06999486,0.0011465446,0.00026461974,0.0031582688,0.0029910188,0.00007321504,0.0020765576],"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","domain_scores_codex":[0.89298457,0.09283258,0.0049368315,0.0039637396,0.004319725,0.0009625157],"domain_scores_gemma":[0.3133272,0.65251684,0.023551207,0.0056283567,0.003223711,0.0017526946],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1528782,0.0023396348,0.0055718957,0.0046455706,0.000741655,0.005060608,0.0033955425,0.0047940523,0.0054866695],"category_scores_gemma":[0.33446002,0.0014582542,0.0055322754,0.0036064084,0.0027248284,0.004418987,0.002711529,0.0050394908,0.000926783],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.011159764,0.00052175426,0.081399575,0.005640084,0.009258778,0.0008469078,0.0007322794,0.71409523,0.0005033723,0.049700987,0.0034602622,0.12268101],"study_design_scores_gemma":[0.0023358662,0.0045937453,0.015625475,0.0016024137,0.0034693733,0.0005003507,0.00013305219,0.87685555,0.00041874338,0.090374276,0.0039161285,0.00017503649],"about_ca_topic_score_codex":0.0052540116,"about_ca_topic_score_gemma":0.004303891,"teacher_disagreement_score":0.1528782,"about_ca_system_score_codex":0.0033406168,"about_ca_system_score_gemma":0.0039543253,"threshold_uncertainty_score":0.80850685},"labels":[],"label_agreement":null},{"id":"W2111635289","doi":"10.1002/pst.433","title":"Optimal caliper widths for propensity‐score matching when estimating differences in means and differences in proportions in observational studies","year":2010,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":3895,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Propensity score matching; Covariate; Calipers; Statistics; Estimator; Mathematics; Logit; Matching (statistics); Observational study; Econometrics; Standard deviation; Confidence interval; Logistic regression; Medicine","score_opus":0.5850172035475715,"score_gpt":0.5093227869028153,"score_spread":0.07569441664475618,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2111635289","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.002250815,0.0005286876,0.99541795,0.00027136042,0.00007565333,0.0003410179,0.00008470524,0.00024608904,0.0007836892],"genre_scores_gemma":[0.03348788,0.0004953936,0.9635423,0.00023668988,0.00010187675,0.0014498917,0.00016161289,0.00016032356,0.00036403758],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9359957,0.053232953,0.0027454912,0.0036741116,0.0038319442,0.0005197449],"domain_scores_gemma":[0.8081051,0.16614056,0.009309796,0.012038122,0.003731029,0.00067545235],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08407174,0.001653233,0.0021917901,0.003623212,0.0012450062,0.002821066,0.0031498307,0.00283857,0.0055735684],"category_scores_gemma":[0.35259444,0.0016904122,0.0024485327,0.004770826,0.0027041375,0.0046465937,0.003786211,0.0059000757,0.0013438899],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009411592,0.0002998363,0.008877102,0.0012399664,0.0005341051,0.00042425597,0.0010601606,0.13590644,0.002644177,0.495127,0.009719002,0.34322676],"study_design_scores_gemma":[0.000567504,0.00043505727,0.003966538,0.0007937132,0.00022062175,0.00040449735,0.00020577048,0.36192673,0.004012359,0.6027439,0.024579884,0.00014338098],"about_ca_topic_score_codex":0.0023298142,"about_ca_topic_score_gemma":0.002309705,"teacher_disagreement_score":0.91592824,"about_ca_system_score_codex":0.0017164839,"about_ca_system_score_gemma":0.0037810132,"threshold_uncertainty_score":0.44461918},"labels":[],"label_agreement":null},{"id":"W2116813387","doi":"10.1002/pst.1680","title":"Assessing the treatment effect in a randomized controlled trial with extensive non‐adherence: the EVOLVE trial","year":2015,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. John’s Health Sciences Centre","funders":"","keywords":"Randomized controlled trial; Medicine; Statistics; Medical physics; Mathematics; Internal medicine","score_opus":0.45167162741721173,"score_gpt":0.5365292550014282,"score_spread":0.0848576275842165,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2116813387","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8456926,0.042406175,0.06556473,0.0059378203,0.0040005976,0.022181176,0.0030917611,0.0008418272,0.010283236],"genre_scores_gemma":[0.9486949,0.0024932865,0.027087081,0.0027845276,0.00068600214,0.014440505,0.0012787973,0.00010860026,0.0024263558],"study_design_codex":"randomized_trial","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.81636655,0.17192252,0.0041348995,0.0032549393,0.0033638875,0.00095723436],"domain_scores_gemma":[0.9095805,0.071233265,0.008716234,0.008012893,0.0013677181,0.0010894143],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08760576,0.0018760461,0.00762912,0.00083627243,0.00055744586,0.0021726731,0.0013184751,0.0034750453,0.003728382],"category_scores_gemma":[0.101753056,0.0007278704,0.0082495585,0.00092882506,0.0022917173,0.0035329186,0.0016527945,0.004900641,0.00039029008],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.89560497,0.0015945453,0.006348718,0.005070684,0.05260471,0.00022785993,0.00022945223,0.0051646717,0.00077882956,0.0059239184,0.0019122121,0.024539419],"study_design_scores_gemma":[0.85063994,0.05893755,0.012338847,0.0012540084,0.026385522,0.0003296427,0.00011719656,0.027203387,0.0015496555,0.013917197,0.0071276943,0.00019928125],"about_ca_topic_score_codex":0.0003981514,"about_ca_topic_score_gemma":0.0005586189,"teacher_disagreement_score":0.9123942,"about_ca_system_score_codex":0.0009136208,"about_ca_system_score_gemma":0.0010256579,"threshold_uncertainty_score":0.46330905},"labels":[],"label_agreement":null},{"id":"W2134800676","doi":"10.1002/pst.1721","title":"Optimal adaptive sequential designs for crossover bioequivalence studies","year":2015,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Bioequivalence; Sequential analysis; Crossover; Sample size determination; Statistics; Mathematics; Crossover study; Adaptive design; Type I and type II errors; Nominal level; Computer science; Mathematical optimization; Confidence interval; Medicine; Machine learning; Clinical trial","score_opus":0.8262700275298006,"score_gpt":0.64127193821718,"score_spread":0.18499808931262063,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2134800676","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02360701,0.00042669915,0.9738461,0.00007065251,0.000054130775,0.0011976709,0.00004764558,0.00011734753,0.0006327531],"genre_scores_gemma":[0.21834087,0.00039120545,0.77605134,0.00013468275,0.000055648692,0.0044296696,0.000093135466,0.000041735464,0.000461799],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.96221274,0.03262319,0.0010890348,0.0016460433,0.002105734,0.0003233024],"domain_scores_gemma":[0.9718292,0.023267727,0.002040635,0.0014107925,0.0012354535,0.00021627858],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.029748818,0.0011783942,0.0017520047,0.0011951483,0.0003778538,0.00074197387,0.0011546693,0.0008938842,0.0027870568],"category_scores_gemma":[0.042258464,0.0007897335,0.0013584975,0.0009099173,0.0011370195,0.00095913355,0.00081932714,0.0016505836,0.00026437594],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.011420567,0.0015230463,0.003723551,0.0018268593,0.0010446443,0.00014127212,0.00057033467,0.3275423,0.01652265,0.09446922,0.0013686928,0.53984696],"study_design_scores_gemma":[0.004414033,0.017724797,0.005214215,0.00026697083,0.0004961858,0.00014183813,0.00007060759,0.82915056,0.012644819,0.120374314,0.009316883,0.0001848745],"about_ca_topic_score_codex":0.00040861845,"about_ca_topic_score_gemma":0.0005003645,"teacher_disagreement_score":0.029748818,"about_ca_system_score_codex":0.0009550123,"about_ca_system_score_gemma":0.0018348396,"threshold_uncertainty_score":0.15732867},"labels":[],"label_agreement":null},{"id":"W2745664880","doi":"10.1002/pst.1823","title":"Competing risk analysis in a large cardiovascular clinical trial: An <scp>APEX</scp> substudy","year":2017,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Thomas Hospital; University of Calgary","funders":"Portola Pharmaceuticals","keywords":"Medicine; Proportional hazards model; Clinical endpoint; Internal medicine; Pulmonary embolism; Hazard ratio; Venous thromboembolism; Deep vein; Thrombosis; Clinical trial; Univariate analysis; Confidence interval; Cardiology; Multivariate analysis","score_opus":0.6114267450329344,"score_gpt":0.5681218487714129,"score_spread":0.04330489626152145,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2745664880","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.59196275,0.3287546,0.020587062,0.010361128,0.0042280657,0.015025583,0.01657279,0.0005016262,0.012006416],"genre_scores_gemma":[0.8917128,0.050541367,0.012739186,0.0060913153,0.0068014124,0.020133574,0.010073947,0.00014422418,0.0017621695],"study_design_codex":"randomized_trial","study_design_gemma":"observational","domain_scores_codex":[0.96869814,0.02676884,0.0019914347,0.00064672745,0.0015402939,0.0003546061],"domain_scores_gemma":[0.94979227,0.036274783,0.0054226066,0.004726327,0.0026700832,0.0011139025],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.027753493,0.0013254086,0.0032452708,0.00087314006,0.0007305037,0.0017222302,0.00053294125,0.0017512034,0.0028947361],"category_scores_gemma":[0.03408072,0.0003910506,0.0064446786,0.0018347584,0.00068201654,0.00095090934,0.0009381022,0.0030083216,0.00031417818],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.41655207,0.0028833933,0.07523098,0.05115647,0.1622126,0.0012109517,0.0006873413,0.0033944529,0.004381916,0.0054438156,0.04846554,0.22838049],"study_design_scores_gemma":[0.4349831,0.062108584,0.20207068,0.009297286,0.19216394,0.002898257,0.00043928376,0.014062169,0.0031156226,0.008335666,0.070026405,0.00049902516],"about_ca_topic_score_codex":0.0020607333,"about_ca_topic_score_gemma":0.0042260136,"teacher_disagreement_score":0.027753493,"about_ca_system_score_codex":0.0007667859,"about_ca_system_score_gemma":0.0013314013,"threshold_uncertainty_score":0.1467762},"labels":[],"label_agreement":null},{"id":"W2789675304","doi":"10.1002/pst.1853","title":"Bayesian statistical models to estimate <scp>EQ‐5D</scp> utility scores from <scp>EORTC QLQ</scp> data in myeloma","year":2018,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Statistics; Regression; Regression analysis; Linear regression; Bayesian probability; Variance (accounting); Mathematics; Medicine","score_opus":0.5672172281859323,"score_gpt":0.5016580102366126,"score_spread":0.06555921794931974,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2789675304","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.16432421,0.003144002,0.8236694,0.0019789168,0.00015514717,0.0007584527,0.0022127293,0.00093543134,0.0028217775],"genre_scores_gemma":[0.80256766,0.0023294932,0.18004015,0.0006923703,0.00025120628,0.0027542308,0.004536427,0.00023859067,0.0065897955],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.989344,0.008734997,0.00031168826,0.00067990023,0.0005557965,0.00037354164],"domain_scores_gemma":[0.94446844,0.05047697,0.002468845,0.00088297675,0.0014125815,0.00029017133],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026314167,0.0015658769,0.0022739086,0.0024237332,0.00080993766,0.0023451787,0.0034027095,0.0021583284,0.0025833715],"category_scores_gemma":[0.06375375,0.0017384919,0.0028878518,0.0018610663,0.0013008787,0.0020799737,0.0018878711,0.0033515606,0.0009087831],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004938267,0.00016200996,0.013477746,0.00020379445,0.00068500655,0.00016509397,0.00032515306,0.9262401,0.00025348313,0.019399378,0.0018925032,0.03670203],"study_design_scores_gemma":[0.0000612772,0.000077702694,0.0024309803,0.000053462725,0.000066239576,0.00003953253,0.000035983485,0.9799049,0.000062880325,0.016622245,0.0006144348,0.000030294717],"about_ca_topic_score_codex":0.023852957,"about_ca_topic_score_gemma":0.017863598,"teacher_disagreement_score":0.026314167,"about_ca_system_score_codex":0.0023197853,"about_ca_system_score_gemma":0.0018398796,"threshold_uncertainty_score":0.13916427},"labels":[],"label_agreement":null},{"id":"W2897367577","doi":"10.1002/pst.1906","title":"Applications of Bayesian statistical methodology to clinical trial design: A case study of a phase 2 trial with an interim futility assessment in patients with knee osteoarthritis","year":2018,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Xenon Pharmaceuticals (Canada)","funders":"Eli Lilly and Company","keywords":"Interim; Interim analysis; Medicine; Placebo; Clinical trial; WOMAC; Osteoarthritis; Sample size determination; Celecoxib; Randomized controlled trial; Bayesian probability; Clinical study design; Physical therapy; Medical physics; Computer science; Surgery; Internal medicine; Alternative medicine; Statistics; Artificial intelligence","score_opus":0.7394911056247505,"score_gpt":0.6836660418249617,"score_spread":0.0558250637997888,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2897367577","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.22869948,0.012819953,0.69273114,0.03236174,0.0005304871,0.0075337156,0.0007032813,0.0003347023,0.024285471],"genre_scores_gemma":[0.67367226,0.003219283,0.31304088,0.0030660399,0.00030481024,0.005159503,0.00016773831,0.00006514644,0.0013043213],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.78315467,0.20715348,0.0024052758,0.0014867069,0.0047368654,0.0010629302],"domain_scores_gemma":[0.5091597,0.46989822,0.008163866,0.0056784856,0.0054239705,0.0016758532],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.19774142,0.0011093477,0.0021772927,0.0014349335,0.0014525938,0.0027297684,0.001744477,0.005846343,0.003109644],"category_scores_gemma":[0.2748119,0.0009702099,0.0035671347,0.0018988666,0.0033175193,0.0016242579,0.0020070882,0.0042880857,0.00043108585],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.015721774,0.002575187,0.04162678,0.005035825,0.0036797374,0.010316138,0.0048194956,0.34032825,0.0023060415,0.25271094,0.01009147,0.31078827],"study_design_scores_gemma":[0.013321095,0.0149218505,0.010417746,0.0023580296,0.0022036287,0.004960227,0.0012001323,0.5765181,0.0025352417,0.33841097,0.032750156,0.00040294012],"about_ca_topic_score_codex":0.0029486164,"about_ca_topic_score_gemma":0.003822691,"teacher_disagreement_score":0.19774142,"about_ca_system_score_codex":0.0028206164,"about_ca_system_score_gemma":0.007670737,"threshold_uncertainty_score":0.98932767},"labels":[],"label_agreement":null},{"id":"W2991547990","doi":"10.1002/pst.1984","title":"Comparisons of outlier tests for potency bioassays","year":2019,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"AstraZeneca (Canada)","funders":"","keywords":"Potency; Outlier; Bioassay; Statistics; Mathematics; Computer science; Chemistry; Biology","score_opus":0.2950893604694196,"score_gpt":0.5342374065051462,"score_spread":0.2391480460357266,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2991547990","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.14420313,0.002571831,0.8447823,0.00050456834,0.0004438613,0.0007470175,0.0008933133,0.0016257789,0.0042281286],"genre_scores_gemma":[0.64390564,0.0011147553,0.3495632,0.00030124508,0.00018747397,0.00077205204,0.002652989,0.0003133831,0.0011893314],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9459803,0.03128923,0.0043824245,0.003779097,0.01340628,0.0011626575],"domain_scores_gemma":[0.6897569,0.26081333,0.011580217,0.015464812,0.021081924,0.0013027955],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04783844,0.0014518166,0.0016433359,0.0039858934,0.0008886352,0.002177445,0.0026267075,0.0017389653,0.0019206665],"category_scores_gemma":[0.19262603,0.00038892106,0.0022318242,0.002981519,0.001701781,0.0027456968,0.0024301463,0.0025823358,0.0005080342],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0059356685,0.0010628556,0.056487523,0.0017786324,0.0015952387,0.00066651567,0.0008862521,0.41319016,0.016102491,0.054018673,0.0064221877,0.44185385],"study_design_scores_gemma":[0.00026911008,0.004485891,0.017323917,0.00018488824,0.0003486571,0.00065020047,0.00046145156,0.91250837,0.03414293,0.022908306,0.0064894296,0.00022683073],"about_ca_topic_score_codex":0.0019872028,"about_ca_topic_score_gemma":0.0017242276,"teacher_disagreement_score":0.04783844,"about_ca_system_score_codex":0.0018169571,"about_ca_system_score_gemma":0.0020253449,"threshold_uncertainty_score":0.25299686},"labels":[],"label_agreement":null},{"id":"W3006624376","doi":"10.1002/pst.2005","title":"START: single‐to‐double arm transition design for phase II clinical trials","year":2020,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"","keywords":"Randomization; Frequentist inference; Computer science; Restricted randomization; Null hypothesis; Clinical study design; Clinical trial; Randomized controlled trial; Statistics; Research design; Gold standard (test); Medical physics; Medicine; Mathematics; Surgery; Bayesian probability; Internal medicine; Bayesian inference","score_opus":0.9480752190546133,"score_gpt":0.7157514049968547,"score_spread":0.2323238140577586,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3006624376","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008760628,0.00052209076,0.9624518,0.00040355878,0.0004737186,0.02107174,0.0010453133,0.0029194255,0.002351647],"genre_scores_gemma":[0.09742336,0.00034437326,0.8217293,0.000788267,0.00017114158,0.075804986,0.00080232875,0.00036361863,0.0025726086],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9346755,0.057695106,0.0020521972,0.0021997858,0.002709554,0.00066787406],"domain_scores_gemma":[0.9444293,0.042550616,0.0035330541,0.0057767094,0.0025004721,0.0012098725],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.055873524,0.0020661445,0.00264572,0.0015983279,0.0006965814,0.0015912842,0.0021869056,0.0023006517,0.02486455],"category_scores_gemma":[0.09268068,0.0012578544,0.0028413746,0.0011526698,0.001918832,0.001514056,0.0024605046,0.0044997637,0.0032194834],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.083550535,0.0028861603,0.006113181,0.0062977374,0.003049143,0.00057513954,0.0009873129,0.11114166,0.0077361465,0.18674435,0.038066372,0.5528522],"study_design_scores_gemma":[0.061711244,0.028583102,0.0037581166,0.0011515733,0.0017774609,0.00049479876,0.00015451951,0.55734193,0.013820339,0.25742114,0.07338035,0.0004054603],"about_ca_topic_score_codex":0.0003543816,"about_ca_topic_score_gemma":0.00051757455,"teacher_disagreement_score":0.055873524,"about_ca_system_score_codex":0.0010144728,"about_ca_system_score_gemma":0.0033626887,"threshold_uncertainty_score":0.29549098},"labels":[],"label_agreement":null},{"id":"W3013414970","doi":"10.1002/pst.2012","title":"A critical review of graphics for subgroup analyses in clinical trials","year":2020,"lang":"en","type":"review","venue":"Pharmaceutical Statistics","topic":"Data Visualization and Analytics","field":"Computer Science","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Horizon 2020 Framework Programme; Medical Research Council Canada; European Commission; Marie Curie; National Institute for Health and Care Research","keywords":"Subgroup analysis; Computer science; Population; Plot (graphics); Identification (biology); Clinical trial; Visualization; Econometrics; Statistics; Data mining; Medicine; Mathematics; Pathology","score_opus":0.7492670943859359,"score_gpt":0.6880764956332216,"score_spread":0.061190598752714265,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3013414970","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00006515082,0.9677979,0.007389495,0.016047377,0.0067286612,0.00032160612,0.00024176501,0.00018656567,0.0012214395],"genre_scores_gemma":[0.0020359096,0.95161587,0.02399521,0.0134782195,0.0056138104,0.0015694137,0.00033833206,0.00028176166,0.0010713772],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9287829,0.046704732,0.0120396875,0.0017298297,0.010231607,0.00051136594],"domain_scores_gemma":[0.65193933,0.29278633,0.013893881,0.006500854,0.033421956,0.0014575656],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09056704,0.0024248231,0.0056132427,0.013343994,0.0010232531,0.0055688526,0.0063117263,0.005798146,0.011043763],"category_scores_gemma":[0.2993375,0.002281941,0.007898375,0.014794696,0.00443083,0.005341488,0.003279002,0.012208661,0.0050445385],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00026173648,0.000031420164,0.0001716243,0.09959248,0.0011795093,0.00014582174,0.0005546547,0.00093822257,0.00021493995,0.020860687,0.29267916,0.5833697],"study_design_scores_gemma":[0.0002026086,0.0001342217,0.00066294137,0.13590738,0.0014126,0.00047598386,0.00013770205,0.00064461044,0.0002919966,0.021671643,0.83831835,0.00013996115],"about_ca_topic_score_codex":0.004741128,"about_ca_topic_score_gemma":0.0067917584,"teacher_disagreement_score":0.90943295,"about_ca_system_score_codex":0.008161365,"about_ca_system_score_gemma":0.013461632,"threshold_uncertainty_score":0.47897},"labels":[],"label_agreement":null},{"id":"W3027176385","doi":"10.1002/pst.2029","title":"Investigating the appropriateness of different concordance measures in a time‐to‐event setting","year":2020,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Censoring (clinical trials); Concordance; Statistic; Statistics; Comparability; Parametric statistics; Event (particle physics); Mathematics; Econometrics; Computer science; Medicine; Internal medicine","score_opus":0.08646133817187293,"score_gpt":0.3480782208862438,"score_spread":0.2616168827143709,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3027176385","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8337283,0.0027689761,0.15360326,0.0011587502,0.00033406614,0.000977448,0.0008745011,0.00027272056,0.006281995],"genre_scores_gemma":[0.97631174,0.00014090993,0.022581106,0.0001611893,0.000032420816,0.00028213355,0.0002768286,0.00004193153,0.00017162129],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.78042907,0.17846905,0.012175576,0.009058074,0.018520122,0.0013481742],"domain_scores_gemma":[0.17856629,0.7607764,0.02433284,0.02039383,0.014540505,0.0013901134],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.21668035,0.00083140814,0.0013720029,0.0033303988,0.0013087963,0.0036665765,0.0027865057,0.0026234614,0.0015227107],"category_scores_gemma":[0.6167494,0.00050726265,0.0028825293,0.003304773,0.0036110003,0.002745501,0.002617929,0.0025444867,0.00027787779],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.01197435,0.0009156071,0.6822913,0.001851575,0.0068620457,0.00071666297,0.0050568655,0.095130526,0.002550803,0.02707499,0.004075713,0.16149954],"study_design_scores_gemma":[0.0015253274,0.012255638,0.44527996,0.0015317634,0.0026706276,0.0019144029,0.0038406271,0.4509202,0.01697612,0.05621318,0.0060710004,0.0008012081],"about_ca_topic_score_codex":0.0036683625,"about_ca_topic_score_gemma":0.0026412923,"teacher_disagreement_score":0.78331965,"about_ca_system_score_codex":0.002704675,"about_ca_system_score_gemma":0.003148177,"threshold_uncertainty_score":0.9659726},"labels":[],"label_agreement":null},{"id":"W3081849395","doi":"10.1002/pst.2067","title":"A stochastically curtailed two‐arm randomised phase<scp>II</scp>trial design for binary outcomes","year":2020,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada; Cancer Research UK","keywords":"Interim; Interim analysis; Research design; Early stopping; Sample size determination; Outcome (game theory); Clinical study design; Null hypothesis; Randomized controlled trial; Computer science; Statistics; Medicine; Mathematics; Clinical trial; Surgery; Machine learning; Internal medicine","score_opus":0.6908497330030037,"score_gpt":0.6098577142060511,"score_spread":0.08099201879695261,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3081849395","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007480949,0.0021594819,0.91243935,0.0022495934,0.002712216,0.06460018,0.00093908823,0.0006905325,0.0067285886],"genre_scores_gemma":[0.087554276,0.0017882051,0.704229,0.0046005324,0.0008652895,0.19551285,0.00059388013,0.00018981023,0.0046662376],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.91876507,0.068077885,0.004208633,0.0028865589,0.0054885065,0.00057335413],"domain_scores_gemma":[0.94416493,0.03892211,0.0059744543,0.0065055946,0.003406122,0.0010268012],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05911991,0.0013693867,0.002926586,0.0011758584,0.0007750357,0.0017782451,0.0020296508,0.0037342173,0.015788058],"category_scores_gemma":[0.0924496,0.0005824002,0.0030058702,0.0013020843,0.002382058,0.0016781746,0.001631183,0.00423995,0.003232082],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.035089396,0.0021401127,0.0027489683,0.023421327,0.0024748042,0.00038435977,0.00068955467,0.022938956,0.010088608,0.38149256,0.032626882,0.4859044],"study_design_scores_gemma":[0.098493524,0.083771735,0.0052279416,0.008253669,0.003979904,0.0017834905,0.00024491502,0.14626928,0.017758887,0.35660005,0.27681398,0.00080259494],"about_ca_topic_score_codex":0.00020625092,"about_ca_topic_score_gemma":0.00036853622,"teacher_disagreement_score":0.94088006,"about_ca_system_score_codex":0.0010429859,"about_ca_system_score_gemma":0.004967022,"threshold_uncertainty_score":0.31265968},"labels":[],"label_agreement":null},{"id":"W3087544668","doi":"10.1002/pst.2068","title":"Assessing the quality of studies in meta‐research: Review/guidelines on the most important quality assessment tools","year":2020,"lang":"en","type":"review","venue":"Pharmaceutical Statistics","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":225,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Jadad scale; Systematic review; Observational study; Meta-analysis; Quality (philosophy); Scale (ratio); Publication bias; Consolidated Standards of Reporting Trials; Randomized controlled trial; Computer science; MEDLINE; Management science; Medicine; Engineering; Cochrane Library; Pathology","score_opus":0.996400707501376,"score_gpt":0.8337696602368216,"score_spread":0.16263104726455435,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3087544668","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0005718577,0.84539765,0.100061186,0.020415358,0.005572525,0.01947234,0.0044337036,0.00092215027,0.003153182],"genre_scores_gemma":[0.012761801,0.46972725,0.43235254,0.009731347,0.002535853,0.0675188,0.0036736464,0.0007387636,0.0009600632],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.47138864,0.28929922,0.18239002,0.0073354407,0.04810904,0.0014775248],"domain_scores_gemma":[0.2838248,0.54226774,0.07634177,0.01906701,0.07671987,0.0017787792],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.38544455,0.0048178555,0.017291164,0.031915482,0.0026323337,0.012308858,0.011544332,0.009566892,0.009341692],"category_scores_gemma":[0.6114539,0.004661108,0.02781073,0.026595186,0.005804483,0.009362474,0.010029309,0.010592085,0.004231509],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005855572,0.00009424503,0.0014552671,0.54182553,0.018686928,0.00020343164,0.0017523657,0.0014555267,0.0005010311,0.013870394,0.08008064,0.33948904],"study_design_scores_gemma":[0.0013470612,0.00023013655,0.0026525592,0.7455384,0.027848665,0.00062119035,0.0004887681,0.0024374456,0.0014766944,0.034231704,0.1826509,0.00047641766],"about_ca_topic_score_codex":0.006713717,"about_ca_topic_score_gemma":0.0076762745,"teacher_disagreement_score":0.6145555,"about_ca_system_score_codex":0.00993762,"about_ca_system_score_gemma":0.033252973,"threshold_uncertainty_score":0.7578563},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W3092379046","doi":"10.1002/pst.2073","title":"Utilizing Bayesian predictive power in clinical trial design","year":2020,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact; University of British Columbia","funders":"","keywords":"Bayesian probability; Interim; Interim analysis; Computer science; Adaptive design; Machine learning; Computation; Clinical trial; Predictive power; Clinical study design; Artificial intelligence; Data mining; Algorithm; Medicine","score_opus":0.8404547068645039,"score_gpt":0.6558272472679741,"score_spread":0.1846274595965297,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3092379046","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0016978615,0.00045868804,0.9958528,0.00065426435,0.00003486702,0.00009477756,0.00002794122,0.00006909769,0.0011098373],"genre_scores_gemma":[0.37571162,0.0029958868,0.61602986,0.001124854,0.0005527595,0.0017604413,0.00018089966,0.0001204569,0.0015231651],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97002184,0.0251456,0.00068558863,0.0009521887,0.002857314,0.00033742137],"domain_scores_gemma":[0.91328037,0.07911476,0.002602187,0.0024619666,0.0021641476,0.00037662825],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04396248,0.0014023153,0.0017386458,0.0020381168,0.00053573895,0.0027175858,0.0017097798,0.0017808676,0.001970045],"category_scores_gemma":[0.16497146,0.0010164623,0.0010877507,0.0017974809,0.0031196778,0.0038237115,0.0027183224,0.0033770392,0.000533977],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002283383,0.000058827,0.0015733264,0.00034370096,0.00019915405,0.00011542685,0.00023067169,0.40644118,0.00043982477,0.48475876,0.0018265584,0.10378428],"study_design_scores_gemma":[0.00010073058,0.00011060337,0.00022757535,0.000107074266,0.000049071376,0.000036192938,0.000015297033,0.46709645,0.0004345378,0.5296903,0.0021086677,0.000023362616],"about_ca_topic_score_codex":0.0017826918,"about_ca_topic_score_gemma":0.0011894045,"teacher_disagreement_score":0.04396248,"about_ca_system_score_codex":0.0018709948,"about_ca_system_score_gemma":0.0042745946,"threshold_uncertainty_score":0.23249859},"labels":[],"label_agreement":null},{"id":"W3092430397","doi":"10.1002/pst.2072","title":"The current state of Bayesian methods in nonclinical pharmaceutical statistics: Survey results and recommendations from the <scp>DIA</scp>/<scp>ASA‐BIOP</scp> Nonclinical Bayesian Working Group","year":2020,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"AstraZeneca (Canada)","funders":"","keywords":"Bayesian probability; Bayesian statistics; Bayesian inference; Medicine; Computer science; Statistics; Mathematics; Artificial intelligence","score_opus":0.5640579803130207,"score_gpt":0.595575910905105,"score_spread":0.0315179305920843,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3092430397","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09603822,0.17230298,0.11139225,0.5781783,0.0020102118,0.00076586043,0.0032829344,0.00057700794,0.035452295],"genre_scores_gemma":[0.45869446,0.28197202,0.17566273,0.071497954,0.002740305,0.0014023925,0.0037240267,0.00041412204,0.0038920795],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.87491745,0.073085,0.016005594,0.0044177733,0.029676216,0.00189795],"domain_scores_gemma":[0.3115643,0.5519064,0.02728685,0.0108836675,0.09190416,0.0064546485],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.21777384,0.0005502202,0.0014331425,0.005949532,0.0015541754,0.009374616,0.0021999122,0.0030525678,0.0047055287],"category_scores_gemma":[0.39461192,0.0008366566,0.0013843648,0.009329591,0.0056703542,0.014930062,0.003177938,0.005971524,0.0016567815],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00034629813,0.00028528282,0.09858512,0.005947151,0.00019302724,0.00013386439,0.008278757,0.0028924195,0.0010099994,0.02906808,0.07240822,0.7808518],"study_design_scores_gemma":[0.00015293152,0.0011399419,0.1780497,0.034269735,0.00039956503,0.0013266699,0.048685737,0.019374603,0.0030220998,0.0633555,0.64950734,0.00071625086],"about_ca_topic_score_codex":0.010324469,"about_ca_topic_score_gemma":0.009145532,"teacher_disagreement_score":0.78222615,"about_ca_system_score_codex":0.005494236,"about_ca_system_score_gemma":0.01475672,"threshold_uncertainty_score":0.9646241},"labels":[],"label_agreement":null},{"id":"W3110632768","doi":"10.1002/pst.2082","title":"Joint analysis of longitudinal measurements and survival times with a cure fraction based on partly linear mixed and semiparametric cure models","year":2020,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Queen's University","funders":"Fundamental Research Funds for the Central Universities; Natural Sciences and Engineering Research Council of Canada; National Natural Science Foundation of China","keywords":"Random effects model; Mixed model; Semiparametric model; Inference; Clinical trial; Trajectory; Semiparametric regression; Survival analysis; Hazard ratio; Fraction (chemistry); Longitudinal study; Mathematics; Statistics; Medicine; Econometrics; Computer science; Regression analysis; Confidence interval; Internal medicine; Parametric statistics; Artificial intelligence; Meta-analysis","score_opus":0.42214573473327416,"score_gpt":0.4370171898770877,"score_spread":0.01487145514381355,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3110632768","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.061896145,0.00029261378,0.9367085,0.0003201177,0.000022610822,0.00006196758,0.00021441508,0.00017667218,0.00030700275],"genre_scores_gemma":[0.8422465,0.0005203079,0.15088579,0.00026492513,0.000095130505,0.0006646337,0.0011825061,0.0001142992,0.004026021],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9941186,0.004049567,0.00019574632,0.00083641737,0.0004727345,0.00032693485],"domain_scores_gemma":[0.97277224,0.021511259,0.0024724912,0.0018781804,0.0009525969,0.00041321624],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014502115,0.0011595521,0.0021938141,0.0014966989,0.000499219,0.0016327229,0.0022666242,0.0015842603,0.0016405197],"category_scores_gemma":[0.03504649,0.0010812923,0.0027920536,0.0013054569,0.002101276,0.0020806727,0.0023838403,0.002563052,0.0003534885],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00079677306,0.00024105606,0.019962952,0.00033853267,0.0006265397,0.00041416462,0.0009808101,0.77059096,0.0032659129,0.13071187,0.0010960731,0.07097431],"study_design_scores_gemma":[0.000022627326,0.000107682725,0.0020546552,0.000022309772,0.00006644614,0.000063057334,0.00004002324,0.96851695,0.0005047409,0.028089782,0.00047091802,0.00004077419],"about_ca_topic_score_codex":0.003144348,"about_ca_topic_score_gemma":0.0028880276,"teacher_disagreement_score":0.014502115,"about_ca_system_score_codex":0.001073111,"about_ca_system_score_gemma":0.0013055581,"threshold_uncertainty_score":0.07669538},"labels":[],"label_agreement":null},{"id":"W3120519245","doi":"10.1002/pst.2092","title":"A weighted log‐rank test and associated effect estimator for cancer trials with delayed treatment effect","year":2021,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"AstraZeneca (Canada)","funders":"National Institutes of Health","keywords":"Mathematics; Estimator; Statistics; Log-rank test; Test statistic; Statistic; Robustness (evolution); Hazard ratio; Proportional hazards model; Sample size determination; Statistical hypothesis testing; Confidence interval","score_opus":0.4828785734049401,"score_gpt":0.6096617833509836,"score_spread":0.12678320994604353,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3120519245","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.027405797,0.0008681907,0.9696714,0.00045666654,0.00010796913,0.0003210995,0.00018632616,0.00025253987,0.000729971],"genre_scores_gemma":[0.61274105,0.0005679856,0.3820203,0.00042910373,0.00023569383,0.0017425851,0.0003893914,0.0000865672,0.0017873202],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95109516,0.04188636,0.0012937747,0.002094731,0.0030463075,0.000583693],"domain_scores_gemma":[0.7381468,0.24175927,0.008277935,0.006642107,0.004405328,0.0007685403],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.061763138,0.000951512,0.0023792665,0.0033338477,0.00038785752,0.0013686058,0.0022486013,0.0020522343,0.0053866697],"category_scores_gemma":[0.18966685,0.00045058806,0.0019093488,0.0023009947,0.002158008,0.0023587134,0.0016245958,0.0022828802,0.00046050758],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0029881394,0.00044267546,0.03147146,0.0018584104,0.0024994237,0.000810065,0.00029647327,0.3035079,0.0038094723,0.26790082,0.005326853,0.3790883],"study_design_scores_gemma":[0.00045584224,0.0017688093,0.005599966,0.00016691191,0.00036070545,0.00030950113,0.00007345122,0.88568187,0.0025598202,0.09948877,0.0034546666,0.00007976248],"about_ca_topic_score_codex":0.0008820948,"about_ca_topic_score_gemma":0.000688845,"teacher_disagreement_score":0.93823683,"about_ca_system_score_codex":0.0011499242,"about_ca_system_score_gemma":0.0025249033,"threshold_uncertainty_score":0.32663858},"labels":[],"label_agreement":null},{"id":"W3137022898","doi":"10.1002/pst.2119","title":"Treatment allocation strategies for umbrella trials in the presence of multiple biomarkers: A comparison of methods","year":2021,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Fakultet Medicinskih Nauka, Univerziteta U Kragujevcu; Newcastle University; Medical Research Council Canada; Cancer Research UK","keywords":"Biomarker; Context (archaeology); Bayesian probability; Set (abstract data type); Medicine; Matching (statistics); Computer science; Hierarchy; Artificial intelligence; Pathology","score_opus":0.8703654930178174,"score_gpt":0.7163172398238357,"score_spread":0.15404825319398174,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3137022898","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06820066,0.011393475,0.8935993,0.0031613791,0.00056520436,0.014956265,0.000507492,0.0005559455,0.0070602493],"genre_scores_gemma":[0.35242766,0.00330128,0.6131823,0.0019202264,0.00016673139,0.025956033,0.0003648536,0.00023292721,0.0024480529],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.67255116,0.30646965,0.006755466,0.0048854155,0.008177061,0.0011612457],"domain_scores_gemma":[0.649094,0.3200395,0.012610565,0.012795554,0.0037273047,0.0017330799],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.17798503,0.0022038417,0.0053415657,0.0025474632,0.0006588023,0.0030933924,0.0035986125,0.0035304201,0.007017768],"category_scores_gemma":[0.26880687,0.0013704246,0.005315449,0.0016103957,0.0021505991,0.0047466503,0.0032474017,0.005757358,0.0010975238],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.037985634,0.0020245838,0.0070497054,0.0058335187,0.008922616,0.00018122682,0.0026366776,0.23623185,0.0013478086,0.18111795,0.004538728,0.5121297],"study_design_scores_gemma":[0.031017816,0.012498514,0.0057213344,0.0035248827,0.0031102144,0.00032289783,0.000516689,0.6502615,0.0020922644,0.27539706,0.015056188,0.0004805606],"about_ca_topic_score_codex":0.0015928987,"about_ca_topic_score_gemma":0.0013637027,"teacher_disagreement_score":0.822015,"about_ca_system_score_codex":0.0037430108,"about_ca_system_score_gemma":0.004489563,"threshold_uncertainty_score":0.941286},"labels":[],"label_agreement":null},{"id":"W3160632722","doi":"10.1002/pst.2132","title":"Selection bias, investment decisions and treatment effect distributions","year":2021,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Caprion (Canada)","funders":"","keywords":"Selection (genetic algorithm); Selection bias; Bayesian probability; Phase (matter); Investment (military); Econometrics; Distribution (mathematics); Computer science; Actuarial science; Economics; Risk analysis (engineering); Statistics; Business; Machine learning; Mathematics; Artificial intelligence; Political science","score_opus":0.5404197726327826,"score_gpt":0.5112801002807985,"score_spread":0.029139672351984114,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3160632722","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.12804833,0.0071068467,0.7828007,0.047542393,0.00059682154,0.0021905801,0.0015466192,0.00034753245,0.029820174],"genre_scores_gemma":[0.8396051,0.0037776856,0.14091301,0.006123572,0.0008280228,0.0020803225,0.0005990129,0.00014642258,0.0059267115],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.86046183,0.113504246,0.004757217,0.005901874,0.012581129,0.0027936872],"domain_scores_gemma":[0.43369102,0.51639795,0.02812806,0.015240539,0.0054424275,0.0011000697],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.20963952,0.0009619133,0.0024473171,0.0027341999,0.0010370886,0.006020559,0.0022156178,0.0061768726,0.008374845],"category_scores_gemma":[0.5107809,0.0010178423,0.001722505,0.002942677,0.0069369203,0.008352658,0.0026624356,0.0065765497,0.0008365434],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00084946037,0.00014880009,0.023437103,0.00052502955,0.0004482391,0.00042564157,0.0013341248,0.029935203,0.00046540046,0.8081269,0.0064902767,0.12781377],"study_design_scores_gemma":[0.00034311935,0.0003297343,0.010856217,0.0006034839,0.00020881837,0.000462783,0.00033951653,0.056965787,0.0010552917,0.92124414,0.0074849334,0.00010619351],"about_ca_topic_score_codex":0.0020881016,"about_ca_topic_score_gemma":0.0015973229,"teacher_disagreement_score":0.79036045,"about_ca_system_score_codex":0.0038472996,"about_ca_system_score_gemma":0.0022467587,"threshold_uncertainty_score":0.97465515},"labels":[],"label_agreement":null},{"id":"W3161958318","doi":"10.1002/pst.2129","title":"The detailed clinical objectives approach to designing clinical trials and choosing estimands","year":2021,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Boehringer Ingelheim (Canada)","funders":"","keywords":"Clinical trial; Medical physics; Computer science; Medicine; Econometrics; Mathematics; Internal medicine","score_opus":0.8294575017396123,"score_gpt":0.6433639861404327,"score_spread":0.18609351559917964,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3161958318","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0007179566,0.006074261,0.9396783,0.022415278,0.0017389867,0.010809468,0.0006193861,0.00035450875,0.017591862],"genre_scores_gemma":[0.015000995,0.0051001315,0.9461588,0.008327955,0.0012594411,0.02071227,0.0002647304,0.00018884818,0.002986847],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.55135095,0.39660114,0.021724962,0.0062140683,0.022689812,0.0014191665],"domain_scores_gemma":[0.46838245,0.46394217,0.019323334,0.024611926,0.020341258,0.0033987623],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.34121478,0.0037193934,0.0047570686,0.00574157,0.0019760767,0.011156143,0.0046500927,0.008581285,0.012440218],"category_scores_gemma":[0.40270734,0.0022838514,0.004844474,0.004287579,0.010226951,0.008983178,0.006724537,0.019628115,0.0065513174],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010743018,0.00024681914,0.0009205129,0.010358234,0.0010743161,0.00029471147,0.002793518,0.010874589,0.0010026284,0.6991033,0.029931571,0.24232553],"study_design_scores_gemma":[0.0011230254,0.0015255634,0.0008493752,0.010296635,0.0007356445,0.0003936215,0.000582969,0.010200077,0.0023148248,0.76031685,0.21138313,0.00027820384],"about_ca_topic_score_codex":0.0012127594,"about_ca_topic_score_gemma":0.001871384,"teacher_disagreement_score":0.6587852,"about_ca_system_score_codex":0.006004657,"about_ca_system_score_gemma":0.026678279,"threshold_uncertainty_score":0.8123995},"labels":[],"label_agreement":null},{"id":"W4200071936","doi":"10.1002/pst.2186","title":"Empirical likelihood confidence interval for sensitivity to the early disease stage","year":2021,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Institute of Biomedical Imaging and Bioengineering; Canadian Institutes of Health Research; National Institute on Aging; National Institutes of Health; Simons Foundation; U.S. Department of Defense","keywords":"Confidence interval; Disease; Inference; Stage (stratigraphy); Sensitivity (control systems); Statistics; Medicine; Computer science; Artificial intelligence; Internal medicine; Mathematics","score_opus":0.09480981289805816,"score_gpt":0.4347240200643359,"score_spread":0.33991420716627774,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4200071936","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.036707442,0.001959769,0.9553599,0.0008171114,0.00011730789,0.00016975119,0.00047856604,0.00052182336,0.0038683289],"genre_scores_gemma":[0.76594764,0.0015333918,0.2282842,0.0004237589,0.0002567322,0.0005930595,0.0012856065,0.00016277487,0.0015129199],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9830035,0.010897164,0.0008097895,0.0022738427,0.0024483313,0.00056748564],"domain_scores_gemma":[0.71709603,0.2549816,0.009381768,0.009242586,0.008133437,0.0011645561],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.034464575,0.0012075031,0.0017126993,0.0033399912,0.0005715532,0.00346151,0.0033217552,0.0027521884,0.0046601677],"category_scores_gemma":[0.27722445,0.000472175,0.0016183463,0.002221865,0.0032409886,0.0044308533,0.0027515695,0.0042719548,0.00071621157],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015859789,0.00029118112,0.044835795,0.0010381363,0.0008152443,0.0007395902,0.0013372422,0.40574703,0.0037691232,0.32978672,0.0058836923,0.20417021],"study_design_scores_gemma":[0.00023820697,0.00042175563,0.016159365,0.0004955714,0.0002616042,0.00084143475,0.00032910405,0.82123595,0.0047394014,0.14897938,0.006062966,0.00023520131],"about_ca_topic_score_codex":0.0018531935,"about_ca_topic_score_gemma":0.00077877473,"teacher_disagreement_score":0.034464575,"about_ca_system_score_codex":0.0012560651,"about_ca_system_score_gemma":0.001402101,"threshold_uncertainty_score":0.18226832},"labels":[],"label_agreement":null},{"id":"W4200543014","doi":"10.1002/pst.2181","title":"A dose‐finding design for dual‐agent trials with patient‐specific doses for one agent with application to an opiate detoxification trial","year":2021,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"NIHR Cambridge Biomedical Research Centre; Medical Research Council Canada; Department of Health and Social Care; Medical Research Council; National Institute for Health and Care Research","keywords":"Clinical trial; Medicine; Maximum tolerated dose; Dosing; Detoxification (alternative medicine); Clinical study design; Medical physics; Pharmacology; Computer science; Internal medicine; Alternative medicine; Pathology","score_opus":0.8261641552162479,"score_gpt":0.618695554779733,"score_spread":0.20746860043651483,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4200543014","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.077097066,0.00077688147,0.9013063,0.0021667313,0.00040944203,0.012954806,0.00041222892,0.0004379884,0.0044385046],"genre_scores_gemma":[0.4681321,0.0003809769,0.5075548,0.0012366041,0.00015749357,0.01845599,0.00018416498,0.00006330056,0.0038345505],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.919904,0.07273091,0.0013755298,0.0033697337,0.001910594,0.0007093175],"domain_scores_gemma":[0.89048845,0.090618975,0.007005713,0.0072380863,0.0029226947,0.0017261817],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08276821,0.0017435355,0.0032998305,0.0013368005,0.000657424,0.0023103622,0.0025959946,0.004387639,0.0072310925],"category_scores_gemma":[0.09873686,0.0011991018,0.002427034,0.0009639591,0.0034772723,0.0021895627,0.002263299,0.0039821453,0.00088998716],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.048066158,0.003231366,0.007950839,0.0024957703,0.0018831249,0.00063925574,0.0009500551,0.43575943,0.00856914,0.31668133,0.004271898,0.1695017],"study_design_scores_gemma":[0.02846923,0.01969563,0.0026162507,0.0003443885,0.0011574674,0.00031019954,0.00014346238,0.7865718,0.0042908145,0.14648274,0.009708484,0.00020961005],"about_ca_topic_score_codex":0.00038972477,"about_ca_topic_score_gemma":0.00036601897,"teacher_disagreement_score":0.9172318,"about_ca_system_score_codex":0.0018665822,"about_ca_system_score_gemma":0.0032362721,"threshold_uncertainty_score":0.43772537},"labels":[],"label_agreement":null},{"id":"W4280542190","doi":"10.1002/pst.2234","title":"Standard and reference‐based conditional mean imputation","year":2022,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":44,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Medical Research Council; Medical Research Council Canada","keywords":"Missing data; Jackknife resampling; Frequentist inference; Imputation (statistics); Pooling; Computer science; Statistics; Inference; Bayesian probability; Econometrics; Bayesian inference; Mathematics; Estimator; Artificial intelligence","score_opus":0.6731853966455711,"score_gpt":0.6174407262752281,"score_spread":0.05574467037034303,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4280542190","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0016316595,0.0008240653,0.99412125,0.00036194234,0.00012624996,0.00016524209,0.0004255042,0.0004001268,0.0019440174],"genre_scores_gemma":[0.08288916,0.0012977129,0.90832585,0.0006015131,0.00027582594,0.0015308618,0.0017595628,0.00036879574,0.002950719],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95139796,0.03449709,0.002357032,0.003967204,0.007098104,0.00068265403],"domain_scores_gemma":[0.94018656,0.033406388,0.0052950233,0.014350309,0.0063435356,0.00041806806],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.043610703,0.0009095363,0.0023562496,0.0028756573,0.00079465884,0.0024990267,0.0055396394,0.002135148,0.0075914254],"category_scores_gemma":[0.15004247,0.00087372225,0.0028682537,0.0056259264,0.0014250259,0.002380582,0.0029923338,0.0030015009,0.0028203991],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00056851323,0.0001487527,0.005745168,0.0012412302,0.001369301,0.00020371153,0.00048036122,0.050888807,0.0010768281,0.40038365,0.02715166,0.510742],"study_design_scores_gemma":[0.00033979872,0.00046158838,0.0046538683,0.0008358127,0.00054456136,0.000860198,0.00009112005,0.305104,0.004591535,0.62907666,0.053217497,0.00022331496],"about_ca_topic_score_codex":0.0015868797,"about_ca_topic_score_gemma":0.0021087404,"teacher_disagreement_score":0.043610703,"about_ca_system_score_codex":0.001176618,"about_ca_system_score_gemma":0.0046398747,"threshold_uncertainty_score":0.2306382},"labels":[],"label_agreement":null},{"id":"W4280562270","doi":"10.1002/pst.2231","title":"Two‐stage subgroup‐specific time‐to‐event (2S‐Sub‐TITE): An adaptive two‐stage time‐to‐toxicity design for subgroup‐specific dose finding in phase I oncology trials","year":2022,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto; Centre for Addiction and Mental Health","funders":"","keywords":"Subgroup analysis; Mathematics; Statistics; Sample size determination; Cluster analysis; Event (particle physics); Confidence interval","score_opus":0.7099022599683399,"score_gpt":0.6160788559671313,"score_spread":0.09382340400120859,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4280562270","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.13728614,0.0007238813,0.83249265,0.00066858495,0.00041963937,0.023877848,0.0005157175,0.0013307482,0.0026847918],"genre_scores_gemma":[0.53418744,0.00020802347,0.4311681,0.00081640953,0.000099821984,0.030444242,0.00051241065,0.00014669284,0.0024167558],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95101225,0.041498404,0.0017938617,0.002999401,0.002107243,0.0005888517],"domain_scores_gemma":[0.95803195,0.028607631,0.0038183723,0.0060729934,0.0021916206,0.001277467],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.052801814,0.0013357296,0.0021358216,0.0010701484,0.00056491053,0.0010584904,0.002193168,0.0025267487,0.006678443],"category_scores_gemma":[0.06737253,0.0010646305,0.0032046242,0.0007745274,0.0018113382,0.0016352859,0.0020696505,0.0031009777,0.0008191031],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.110255815,0.005525324,0.024546083,0.0022384552,0.0042099617,0.00054023793,0.0016995441,0.34748662,0.016669018,0.061206017,0.005904111,0.41971883],"study_design_scores_gemma":[0.04250701,0.05328895,0.009687403,0.00027754554,0.0016691791,0.0005349269,0.00015132487,0.81321925,0.0100547,0.05237297,0.015858741,0.00037793865],"about_ca_topic_score_codex":0.00052975863,"about_ca_topic_score_gemma":0.00078863854,"teacher_disagreement_score":0.052801814,"about_ca_system_score_codex":0.0011054482,"about_ca_system_score_gemma":0.002765065,"threshold_uncertainty_score":0.27924603},"labels":[],"label_agreement":null},{"id":"W4285732331","doi":"10.1002/pst.2257","title":"Left truncation in linked data: A practical guide to understanding left truncation and applying it using <scp>SAS</scp> and R","year":2022,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Bioinformatics and Genomic Networks","field":"Biochemistry, Genetics and Molecular Biology","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Roche (Canada)","funders":"Genentech","keywords":"Truncation (statistics); Mathematics; Computer science; Left behind; Statistics; Medicine","score_opus":0.13040517423521852,"score_gpt":0.4019486621779917,"score_spread":0.27154348794277317,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4285732331","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00030158335,0.0007817853,0.9392306,0.0022401696,0.0008510099,0.0007336003,0.013171344,0.038415022,0.004274754],"genre_scores_gemma":[0.0031933212,0.0012139214,0.9675666,0.0024568024,0.0004670121,0.003651521,0.0047441814,0.013424365,0.0032822657],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.97526914,0.016173698,0.0033359572,0.0014428071,0.0033914489,0.00038688924],"domain_scores_gemma":[0.79479533,0.17285536,0.005840506,0.01436085,0.010949794,0.0011981591],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04384577,0.002703533,0.0022089006,0.0049336175,0.00096831075,0.0056417882,0.004728367,0.0026230386,0.088108845],"category_scores_gemma":[0.15512612,0.0037302987,0.0038604266,0.00592816,0.0024081876,0.004430437,0.0036581848,0.010015684,0.060256366],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00022010559,0.00012524267,0.0020022532,0.00405366,0.00044168028,0.0009858476,0.0012987909,0.007581964,0.0032009422,0.09661014,0.6788041,0.20467535],"study_design_scores_gemma":[0.00028731834,0.000106796666,0.0019951873,0.0022125007,0.00010914571,0.0010753573,0.0004733881,0.029722461,0.003741317,0.16783006,0.7921695,0.00027687952],"about_ca_topic_score_codex":0.004295874,"about_ca_topic_score_gemma":0.005303301,"teacher_disagreement_score":0.088108845,"about_ca_system_score_codex":0.0012124066,"about_ca_system_score_gemma":0.00598521,"threshold_uncertainty_score":0.29475343},"labels":[],"label_agreement":null},{"id":"W4289544708","doi":"10.1002/pst.2258","title":"Key considerations for choosing a statistical method to deal with incomplete treatment adherence in pragmatic trials","year":2022,"lang":"en","type":"review","venue":"Pharmaceutical Statistics","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Causal inference; Inverse probability weighting; Covariate; Computer science; Weighting; Statistical inference; Inference; Key (lock); Clinical trial; Medicine; Machine learning; Econometrics; Artificial intelligence; Statistics; Mathematics; Estimator","score_opus":0.7492029465102941,"score_gpt":0.639858577016374,"score_spread":0.10934436949392001,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4289544708","genre_codex":"methods","genre_gemma":"review","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0013579415,0.013433882,0.9065529,0.06146482,0.002957521,0.0072500953,0.0006604337,0.00063539855,0.0056870324],"genre_scores_gemma":[0.016360072,0.0045052073,0.95229924,0.007091046,0.0011382293,0.017579516,0.00015954215,0.00031308384,0.00055409945],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.27049842,0.6496689,0.050825585,0.006102561,0.021825098,0.0010794449],"domain_scores_gemma":[0.11563481,0.831257,0.018251937,0.017885875,0.015787328,0.0011831224],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5355977,0.002632151,0.0055977046,0.006031406,0.0028251666,0.013405961,0.0068293605,0.01037003,0.010936435],"category_scores_gemma":[0.83608305,0.0030932645,0.0061320476,0.010651581,0.009738573,0.014371659,0.0059738085,0.018537926,0.0036125206],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011585872,0.00019427863,0.0030260037,0.030287473,0.0022023001,0.0006557943,0.006910423,0.009221804,0.0006655393,0.50298214,0.05237919,0.39031655],"study_design_scores_gemma":[0.0012551806,0.000414618,0.0014722971,0.034741506,0.0012742083,0.00087611226,0.0008205278,0.02622882,0.0016485929,0.81276935,0.11801005,0.00048879575],"about_ca_topic_score_codex":0.0025299606,"about_ca_topic_score_gemma":0.0040032887,"teacher_disagreement_score":0.46440232,"about_ca_system_score_codex":0.00570561,"about_ca_system_score_gemma":0.022273134,"threshold_uncertainty_score":0.5726908},"labels":[],"label_agreement":null},{"id":"W4293724130","doi":"10.1002/pst.2263","title":"A comparison of statistical methods for animal oncology studies","year":2022,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Mathematical Biology Tumor Growth","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"AstraZeneca (Canada)","funders":"","keywords":"Statistics; Missing data; Data set; Statistical power; Medicine; Volume (thermodynamics); Oncology; Medical physics; Computer science; Mathematics","score_opus":0.4606261912846036,"score_gpt":0.6514265736967599,"score_spread":0.19080038241215636,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4293724130","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015836922,0.0036866514,0.97471267,0.00057582813,0.0005068264,0.00075893133,0.00051868067,0.0007993217,0.0026041074],"genre_scores_gemma":[0.0920978,0.002613077,0.8993151,0.0003398465,0.00023333084,0.0029616284,0.0006490963,0.00063509034,0.0011549343],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.90225846,0.08111538,0.0034396984,0.0024876243,0.010201992,0.00049694965],"domain_scores_gemma":[0.6394719,0.3190769,0.0071104993,0.018018661,0.015435079,0.00088699727],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08618454,0.0012483858,0.001642524,0.0035475502,0.0007504656,0.0016972044,0.0017878601,0.0018490674,0.0040266984],"category_scores_gemma":[0.23968717,0.00052598224,0.001957983,0.0034897714,0.0015562804,0.0023419643,0.0018042438,0.002813647,0.00078665704],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0040555,0.00084928406,0.014849626,0.0041676247,0.0030595595,0.00041010926,0.0010835134,0.08443992,0.009641784,0.17671353,0.015701992,0.6850276],"study_design_scores_gemma":[0.00091150345,0.007737029,0.024993686,0.002315119,0.0012647094,0.0017077894,0.00078315305,0.7040781,0.0166705,0.17142422,0.06765071,0.00046343633],"about_ca_topic_score_codex":0.0008915034,"about_ca_topic_score_gemma":0.0008961188,"teacher_disagreement_score":0.91381544,"about_ca_system_score_codex":0.0014662022,"about_ca_system_score_gemma":0.002500934,"threshold_uncertainty_score":0.45579284},"labels":[],"label_agreement":null},{"id":"W4297811834","doi":"10.1002/pst.2336","title":"Evaluating hybrid controls methodology in early‐phase oncology trials: A simulation study based on the <scp>MORPHEUS‐UC</scp> trial","year":2023,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Roche (Canada)","funders":"F. Hoffmann-La Roche","keywords":"Frequentist inference; Bayesian probability; Sample size determination; Prior probability; Statistics; Clinical trial; Computer science; Econometrics; Medicine; Oncology; Bayesian inference; Mathematics; Internal medicine","score_opus":0.9531832501912486,"score_gpt":0.7584745799826099,"score_spread":0.1947086702086387,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4297811834","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5662974,0.0026584938,0.4213058,0.0012782566,0.00017070211,0.0025760643,0.00081051566,0.00035695577,0.0045458595],"genre_scores_gemma":[0.90235084,0.00032095678,0.09419889,0.00028173823,0.000040025192,0.0014936351,0.00045053643,0.000038493352,0.00082498876],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9460118,0.050119597,0.0007911783,0.0013719496,0.0012285357,0.0004768604],"domain_scores_gemma":[0.6119252,0.36318162,0.010183303,0.008749546,0.0047452603,0.0012150478],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10277665,0.0013920473,0.0019131901,0.0011460918,0.0004921725,0.0015301803,0.0024615948,0.0017611152,0.0023635535],"category_scores_gemma":[0.14453003,0.0006502523,0.0022758013,0.0011862246,0.0017550304,0.001691758,0.0015351189,0.0025219088,0.00017512096],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0064688413,0.0006180036,0.011576446,0.00046288056,0.0013353894,0.00028552563,0.00030327594,0.91894764,0.00059894676,0.031516656,0.0010814932,0.026804892],"study_design_scores_gemma":[0.0024509418,0.0014052814,0.0017098408,0.00008358061,0.00039730937,0.000060708964,0.00005598081,0.9784453,0.00070828095,0.013899111,0.0007349663,0.00004867205],"about_ca_topic_score_codex":0.0056561017,"about_ca_topic_score_gemma":0.0025955343,"teacher_disagreement_score":0.89722335,"about_ca_system_score_codex":0.0020473055,"about_ca_system_score_gemma":0.0021165602,"threshold_uncertainty_score":0.5435413},"labels":[],"label_agreement":null},{"id":"W4311687382","doi":"10.1002/pst.2280","title":"Parametric and nonparametric methods for confidence intervals and sample size planning for win probability in parallel‐group randomized trials with Likert item and Likert scale data","year":2022,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; Western University","funders":"Natural Sciences and Engineering Research Council of Canada; National Natural Science Foundation of China","keywords":"Likert scale; Nonparametric statistics; Statistics; Confidence interval; Sample size determination; Context (archaeology); Parametric statistics; Computer science; Mathematics","score_opus":0.863855780924001,"score_gpt":0.6504602847221514,"score_spread":0.2133954962018495,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4311687382","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000667048,0.0014130648,0.99180716,0.00054161024,0.00046827734,0.0025069357,0.00049821177,0.00074692833,0.001350749],"genre_scores_gemma":[0.033120584,0.0013357791,0.933483,0.0008302596,0.00046944126,0.028827991,0.0005768386,0.000524376,0.0008316162],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.48736113,0.45266706,0.015849188,0.015314698,0.027326789,0.0014810971],"domain_scores_gemma":[0.19566675,0.73399866,0.020960035,0.038278196,0.010355364,0.00074090867],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.39980677,0.0050468408,0.009096893,0.01307156,0.0015467848,0.0064642034,0.00971953,0.00874321,0.016724829],"category_scores_gemma":[0.7582874,0.0032926302,0.012079017,0.011154544,0.008910226,0.008771469,0.0063214917,0.016732413,0.0031406945],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00294213,0.00035022653,0.0038450505,0.010673958,0.005401601,0.00074921706,0.0026127761,0.07545401,0.00083825586,0.55230826,0.024162915,0.32066163],"study_design_scores_gemma":[0.0018197565,0.001483556,0.0027406567,0.0048142658,0.0019958364,0.0007606251,0.00037687266,0.26892644,0.0032111988,0.6683447,0.045097154,0.00042894084],"about_ca_topic_score_codex":0.0025168783,"about_ca_topic_score_gemma":0.0013408406,"teacher_disagreement_score":0.60019326,"about_ca_system_score_codex":0.005229407,"about_ca_system_score_gemma":0.0067970105,"threshold_uncertainty_score":0.7401451},"labels":[],"label_agreement":null},{"id":"W4315619070","doi":"10.1002/pst.2285","title":"Natural cubic splines for the analysis of Alzheimer's clinical trials","year":2023,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":33,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Institute of Biomedical Imaging and Bioengineering; Canadian Institutes of Health Research; National Institutes of Health; Genentech; IXICO; H. Lundbeck A/S; Servier; Eisai; Northern California Institute for Research and Education; F. Hoffmann-La Roche; National Center for Advancing Translational Sciences; University of Southern California; Biogen; Eli Lilly and Company; Bristol-Myers Squibb; BioClinica; Meso Scale Diagnostics; Alzheimer's Disease Neuroimaging Initiative; Novartis Pharmaceuticals Corporation; Pfizer; National Institute on Aging; Alzheimer's Association; Foundation for the National Institutes of Health","keywords":"Categorical variable; Repeated measures design; Clinical trial; Type I and type II errors; Statistics; Mixed model; Mathematics; Covariate; Computer science; Medicine; Econometrics","score_opus":0.7673059866440124,"score_gpt":0.6597566135267552,"score_spread":0.10754937311725721,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4315619070","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0012899205,0.0021893552,0.99296653,0.00078362337,0.00029368003,0.00026376222,0.00051355903,0.0008212138,0.0008782608],"genre_scores_gemma":[0.07663894,0.005032865,0.9047101,0.0009316431,0.0008449652,0.0051937196,0.0021473318,0.0007487513,0.003751757],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9237923,0.06565269,0.0023590645,0.0027882515,0.0047915517,0.00061608024],"domain_scores_gemma":[0.8267593,0.1446299,0.009282083,0.013269488,0.004775173,0.0012840932],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08228739,0.0016512752,0.003504534,0.004371029,0.00078794936,0.0027556876,0.0037390173,0.0026317842,0.016174514],"category_scores_gemma":[0.22306238,0.0014154731,0.0050771,0.0065888762,0.0030779324,0.0025580395,0.0035330984,0.007298306,0.0036573661],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009959704,0.00020646158,0.0054004053,0.0026839077,0.0019364767,0.00040813372,0.0006508817,0.121395834,0.00076608756,0.5825679,0.025074942,0.25791305],"study_design_scores_gemma":[0.00035371672,0.00046452062,0.0018485329,0.000730859,0.00022940656,0.00015454329,0.00007148392,0.42551962,0.00041356715,0.53685224,0.033257883,0.000103620594],"about_ca_topic_score_codex":0.005956075,"about_ca_topic_score_gemma":0.0041882675,"teacher_disagreement_score":0.08228739,"about_ca_system_score_codex":0.0025239056,"about_ca_system_score_gemma":0.0076350346,"threshold_uncertainty_score":0.4351825},"labels":[],"label_agreement":null},{"id":"W4319657136","doi":"10.1002/pst.2292","title":"Bayesian single‐to‐double arm transition design using both short‐term and long‐term endpoints","year":2023,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"National Cancer Institute","keywords":"Clinical endpoint; Frequentist inference; Sample size determination; Bayesian probability; Computer science; Statistics; Randomized controlled trial; Bayesian inference; Medicine; Mathematics; Internal medicine","score_opus":0.7559694574546456,"score_gpt":0.6006447139559284,"score_spread":0.15532474349871717,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4319657136","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01268689,0.00052449194,0.98268753,0.0003911823,0.00015281289,0.0019475587,0.00015715584,0.00021985693,0.0012325508],"genre_scores_gemma":[0.3214929,0.00086510275,0.6629063,0.0007958524,0.00017371579,0.010708027,0.00042392442,0.00008747942,0.0025466986],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9602263,0.032851424,0.0011716828,0.002737819,0.002541625,0.00047111453],"domain_scores_gemma":[0.9623825,0.027901333,0.0032957008,0.0033923623,0.0021268944,0.0009012093],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.041140698,0.0015457842,0.0029516446,0.001138329,0.00059486675,0.0017810031,0.0021225773,0.0022336917,0.0052415733],"category_scores_gemma":[0.05134093,0.00086120464,0.0022055705,0.00087903143,0.0023669514,0.0020030064,0.0018326141,0.0035002972,0.0008983304],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.018641217,0.0017121509,0.0060759075,0.0023855339,0.0012632009,0.00027501484,0.0005770249,0.17976622,0.010679485,0.4146831,0.0043501575,0.35959098],"study_design_scores_gemma":[0.0067719254,0.015287893,0.004156943,0.00045314652,0.0010898624,0.00038562567,0.00008105455,0.6442493,0.008976324,0.30073282,0.01749476,0.00032021853],"about_ca_topic_score_codex":0.0005373596,"about_ca_topic_score_gemma":0.00052409555,"teacher_disagreement_score":0.041140698,"about_ca_system_score_codex":0.001263708,"about_ca_system_score_gemma":0.0036734915,"threshold_uncertainty_score":0.21757537},"labels":[],"label_agreement":null},{"id":"W4378803918","doi":"10.1002/pst.2317","title":"Variance estimation of the risk difference when using propensity‐score matching and weighting with time‐to‐event outcomes","year":2023,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Sunnybrook Hospital; University of Toronto","funders":"Ministry of Long-Term Care; Canadian Institutes of Health Research; Institute for Clinical Evaluative Sciences; Ministry of Health, Ontario","keywords":"Propensity score matching; Weighting; Statistics; Matching (statistics); Variance (accounting); Event (particle physics); Estimation; Mathematics; Event data; Econometrics; Inverse probability weighting; Medicine; Covariate; Economics","score_opus":0.20647756529539354,"score_gpt":0.4342598912393333,"score_spread":0.22778232594393974,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4378803918","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0028531929,0.00043898996,0.99441695,0.00034244324,0.00013989302,0.00034841528,0.0002032109,0.00012025251,0.001136702],"genre_scores_gemma":[0.13474497,0.0012762293,0.85741305,0.00068461837,0.00035363343,0.0026263306,0.00086257525,0.00018591878,0.001852691],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.93335986,0.049968448,0.0034677081,0.004925806,0.0075775455,0.0007007565],"domain_scores_gemma":[0.8828481,0.090223424,0.008439002,0.014091821,0.004150092,0.0002475994],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06385167,0.0013619531,0.0019306366,0.0032243473,0.00064612506,0.0028202715,0.0026978834,0.0023304105,0.0048742495],"category_scores_gemma":[0.26792857,0.00078390026,0.003461265,0.0045581968,0.002149839,0.0035372877,0.0030310997,0.003493589,0.000709668],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003567664,0.00020727771,0.02218682,0.0012278579,0.0019237278,0.00029947757,0.0007424274,0.10638967,0.0009479965,0.55393565,0.008874602,0.30290776],"study_design_scores_gemma":[0.00027035578,0.0003151527,0.010175441,0.00059967325,0.00058917253,0.00039983023,0.00019204289,0.2546582,0.002363687,0.7092428,0.021024732,0.00016889753],"about_ca_topic_score_codex":0.004045733,"about_ca_topic_score_gemma":0.0023126241,"teacher_disagreement_score":0.06385167,"about_ca_system_score_codex":0.0014391827,"about_ca_system_score_gemma":0.0028696873,"threshold_uncertainty_score":0.33768392},"labels":[],"label_agreement":null},{"id":"W4380728286","doi":"10.1002/pst.2318","title":"Alone, together: On the benefits of Bayesian borrowing in a meta‐analytic setting","year":2023,"lang":"en","type":"review","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"College of Veterinarians of British Columbia; NeuroDevNet","funders":"","keywords":"Bayes' theorem; Bayesian probability; Randomized controlled trial; Inference; Sample size determination; Bayesian inference; Econometrics; Fraction (chemistry); Bayesian hierarchical modeling; Randomized response; Meta-analysis; Bayes factor; Statistics; Computer science; Mathematics; Medicine; Artificial intelligence; Internal medicine","score_opus":0.8273449729662699,"score_gpt":0.6381815326863675,"score_spread":0.1891634402799024,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4380728286","genre_codex":"methods","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0084633855,0.070188485,0.90081894,0.011159552,0.00047682764,0.0005745927,0.00045774452,0.00035377388,0.0075066984],"genre_scores_gemma":[0.38481766,0.039760426,0.56234205,0.0063989307,0.001421979,0.002270214,0.00037891028,0.0003800894,0.0022297313],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.70709854,0.27283412,0.0065825875,0.0039732545,0.009078393,0.00043307428],"domain_scores_gemma":[0.4493711,0.51644397,0.012521117,0.017084602,0.0039739674,0.00060521293],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.20757043,0.0016917057,0.0053871884,0.003644456,0.0006086189,0.004325536,0.0033655118,0.0030037716,0.005906498],"category_scores_gemma":[0.43531477,0.0014781366,0.0044513266,0.005386831,0.003356746,0.008102517,0.00501759,0.006306447,0.0006885746],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00240337,0.000155005,0.0038054609,0.012883488,0.016562467,0.00032387147,0.00097542786,0.061992545,0.0005482158,0.3724293,0.00629505,0.52162576],"study_design_scores_gemma":[0.000853898,0.00058657443,0.0024022022,0.0077000083,0.006549986,0.00038852188,0.00014578897,0.09390492,0.0009800451,0.8687106,0.017629914,0.00014759685],"about_ca_topic_score_codex":0.0016599804,"about_ca_topic_score_gemma":0.0013569684,"teacher_disagreement_score":0.79242957,"about_ca_system_score_codex":0.0017872705,"about_ca_system_score_gemma":0.0030346394,"threshold_uncertainty_score":0.9772067},"labels":[],"label_agreement":null},{"id":"W4386810972","doi":"10.1002/pst.2338","title":"A marginalized two‐part joint model for a longitudinal biomarker and a terminal event with application to advanced head and neck cancers","year":2023,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Lunenfeld-Tanenbaum Research Institute; University of Toronto; Mount Sinai Hospital","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Institut National Du Cancer","keywords":"Biomarker; Covariate; Medicine; Oncology; Head and neck cancer; Clinical trial; Event (particle physics); Statistics; Internal medicine; Cancer; Mathematics; Biology","score_opus":0.5750052390768035,"score_gpt":0.5987650466759101,"score_spread":0.023759807599106675,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386810972","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.043993413,0.0012127747,0.9506138,0.001402186,0.00018862312,0.00019906978,0.0008900079,0.0002818034,0.0012182542],"genre_scores_gemma":[0.7795379,0.001967505,0.19950995,0.00086014735,0.00054297154,0.0016249,0.0020400044,0.00018723194,0.013729382],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9923707,0.005417509,0.00024088757,0.0010074008,0.00044817672,0.0005153705],"domain_scores_gemma":[0.95887834,0.0348375,0.0024620628,0.0015573727,0.0016187414,0.0006459857],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.018468557,0.0019441388,0.0030027626,0.0013183496,0.0005210442,0.0019173777,0.0034665682,0.002421681,0.0059660156],"category_scores_gemma":[0.035664696,0.001031764,0.0031490268,0.0016344525,0.002713252,0.0021032768,0.0021316055,0.0034856927,0.0007279217],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017041719,0.00026812404,0.014113846,0.00049821177,0.00082206324,0.0011774878,0.00056507526,0.6944546,0.0017253136,0.23699725,0.003429848,0.044244036],"study_design_scores_gemma":[0.00016545346,0.0003099644,0.0021404808,0.00004596063,0.00023114987,0.00014953848,0.000059123224,0.93829113,0.00031377852,0.0565145,0.0017316189,0.000047300884],"about_ca_topic_score_codex":0.009101326,"about_ca_topic_score_gemma":0.0044041965,"teacher_disagreement_score":0.018468557,"about_ca_system_score_codex":0.0016298448,"about_ca_system_score_gemma":0.0025598125,"threshold_uncertainty_score":0.097672224},"labels":[],"label_agreement":null},{"id":"W4390236152","doi":"10.1002/pst.2357","title":"Evaluation of a flexible piecewise linear mixed‐effects model in the analysis of randomized cross‐over trials","year":2023,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"International Development Research Centre; Government of Canada","keywords":"Univariate; Mixed model; Random effects model; Generalized linear mixed model; Piecewise; Linear model; Statistics; Computer science; Sample size determination; Residual; Piecewise linear function; Econometrics; Meta-analysis; Mathematics; Medicine; Algorithm; Multivariate statistics; Internal medicine","score_opus":0.8296799381412743,"score_gpt":0.7032816473815637,"score_spread":0.12639829075971054,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4390236152","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014660689,0.0015157006,0.9807926,0.0005387297,0.00018587548,0.0011713458,0.0002195192,0.00044873919,0.00046672975],"genre_scores_gemma":[0.28138345,0.0012996987,0.7077405,0.0008618162,0.00013792176,0.006111355,0.00050099276,0.00014014823,0.0018241396],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.82939917,0.15830511,0.0025871578,0.0052791554,0.0034315481,0.000997814],"domain_scores_gemma":[0.6396152,0.33735827,0.008909297,0.0077406936,0.0053143296,0.0010622377],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.15694432,0.0025300186,0.0047652707,0.0018864213,0.0006613252,0.0024933645,0.0042280043,0.0041346494,0.0056922804],"category_scores_gemma":[0.2442074,0.0014901246,0.005759534,0.0020825397,0.001970224,0.0033104462,0.00269295,0.0052090124,0.0007694949],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009567262,0.00049659587,0.009920956,0.0028250676,0.005288381,0.00066716754,0.0009068375,0.71403337,0.0016066406,0.078936264,0.0020758002,0.1736757],"study_design_scores_gemma":[0.0009234347,0.0033999835,0.0017739206,0.00025786797,0.001082607,0.00013029124,0.00010491905,0.96380204,0.00090945547,0.025231414,0.0022890032,0.00009504508],"about_ca_topic_score_codex":0.0029706655,"about_ca_topic_score_gemma":0.001779827,"teacher_disagreement_score":0.84305567,"about_ca_system_score_codex":0.0023326399,"about_ca_system_score_gemma":0.0034180651,"threshold_uncertainty_score":0.83001083},"labels":[],"label_agreement":null},{"id":"W4391650553","doi":"10.1002/pst.2365","title":"Assessing the performance of group‐based trajectory modeling method to discover different patterns of medication adherence","year":2024,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact; University of Victoria; Public Health Ontario; University of Toronto; Cascades (Canada); Université Laval; NeuroDevNet","funders":"","keywords":"Covariate; Nonparametric statistics; Variance (accounting); Trajectory; Statistics; Computer science; Medicine; Repeated measures design; Quality (philosophy); Medication adherence; Random effects model; Mathematics; Econometrics; Internal medicine; Meta-analysis","score_opus":0.620705524597161,"score_gpt":0.6302486008024275,"score_spread":0.00954307620526651,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391650553","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.64702696,0.0011412634,0.34720045,0.00069979264,0.00009794059,0.00023824055,0.0010097139,0.000727909,0.0018576776],"genre_scores_gemma":[0.898971,0.00030185172,0.098385595,0.00007799586,0.000027603017,0.0001740918,0.0014350787,0.000051535062,0.00057510677],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9978715,0.0013053459,0.00012045204,0.00036381237,0.00021495414,0.00012384576],"domain_scores_gemma":[0.98303026,0.0136068305,0.00097298075,0.0010276602,0.0010498852,0.0003124182],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008938635,0.00086754694,0.001156354,0.0016798682,0.00058259396,0.0011287197,0.00094658806,0.0011423451,0.0012335147],"category_scores_gemma":[0.02424642,0.00027247553,0.0016449544,0.0012371782,0.00035659462,0.0010358755,0.0009331591,0.001186627,0.00027668342],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012204583,0.00049219077,0.16812609,0.0003324166,0.0011912568,0.00019733788,0.0006225453,0.67977387,0.0015186236,0.005833005,0.00201338,0.1386788],"study_design_scores_gemma":[0.000019449855,0.0000963257,0.005615332,0.000021564414,0.00005313455,0.000024176115,0.00007395754,0.9917551,0.00038554872,0.001673423,0.00026547242,0.000016556305],"about_ca_topic_score_codex":0.028487423,"about_ca_topic_score_gemma":0.011256411,"teacher_disagreement_score":0.9910614,"about_ca_system_score_codex":0.00089203025,"about_ca_system_score_gemma":0.0020809567,"threshold_uncertainty_score":0.056643188},"labels":[],"label_agreement":null},{"id":"W4400056122","doi":"10.1002/pst.2414","title":"<scp>T3</scp> + 3: 3 + 3 Design With Delayed Outcomes","year":2024,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"National Institute of General Medical Sciences; National Cancer Institute; National Institutes of Health","keywords":"Isotonic regression; Maximum tolerated dose; Toxicity; Clinical trial; Outcome (game theory); Research design; Duration (music); Clinical study design; Computer science; Logistic regression; Medicine; Reliability engineering; Statistics; Mathematics; Internal medicine; Engineering; Machine learning","score_opus":0.6329948285871682,"score_gpt":0.5963440292239538,"score_spread":0.03665079936321436,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4400056122","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0416545,0.00048774155,0.93583983,0.0006642698,0.00052274234,0.008655961,0.0010239073,0.0014534032,0.00969761],"genre_scores_gemma":[0.31573868,0.00040244518,0.6535512,0.0017181075,0.00020111717,0.020594789,0.0007872095,0.0004039365,0.0066025616],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.986681,0.010403195,0.0005383607,0.0009434015,0.0011050039,0.00032898615],"domain_scores_gemma":[0.98029673,0.012318308,0.0022207536,0.0028005196,0.0015662203,0.0007975613],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01565156,0.0010661063,0.0012320922,0.0006158785,0.00046283627,0.0009465219,0.001255775,0.0011880441,0.017351452],"category_scores_gemma":[0.023310574,0.00043608612,0.0019239098,0.00066370144,0.0010563764,0.0006757611,0.0010107235,0.0017541718,0.0018409673],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.03415636,0.0027129946,0.011303563,0.0028272262,0.0015173698,0.00054590154,0.00041597045,0.13481271,0.01727719,0.14394924,0.023735644,0.6267457],"study_design_scores_gemma":[0.022290753,0.06351337,0.014329105,0.0005224585,0.002146503,0.0013136321,0.00015627097,0.5650862,0.029181331,0.16422644,0.13682468,0.00040936065],"about_ca_topic_score_codex":0.000672218,"about_ca_topic_score_gemma":0.0009493315,"teacher_disagreement_score":0.017351452,"about_ca_system_score_codex":0.00091624603,"about_ca_system_score_gemma":0.0034806926,"threshold_uncertainty_score":0.08277434},"labels":[],"label_agreement":null},{"id":"W4402323319","doi":"10.1002/pst.2436","title":"Propensity Score Analysis With Baseline and Follow‐Up Measurements of the Outcome Variable","year":2024,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sunnybrook Hospital; University of Toronto","funders":"Canadian Institutes of Health Research","keywords":"Propensity score matching; Baseline (sea); Statistics; Weighting; Estimator; Observational study; Variable (mathematics); Matching (statistics); Regression analysis; Mathematics; Medicine","score_opus":0.4838276404501285,"score_gpt":0.4782297378790365,"score_spread":0.005597902571091995,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402323319","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0048025697,0.00043194016,0.9879552,0.0004304772,0.00020194145,0.002759252,0.0011179319,0.0005238212,0.0017768858],"genre_scores_gemma":[0.13694899,0.0011510988,0.8441992,0.00067457225,0.00026425524,0.010343807,0.0026320922,0.00041309895,0.0033729451],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9028565,0.07859636,0.005808572,0.0053934776,0.0064862953,0.00085876643],"domain_scores_gemma":[0.805305,0.13531613,0.014992374,0.034759697,0.008900907,0.00072582054],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.12366636,0.0019206573,0.0026601662,0.0046359017,0.0012276962,0.0036697236,0.0036574963,0.002388369,0.01785894],"category_scores_gemma":[0.3298706,0.0012157116,0.0066776173,0.008079628,0.0018703687,0.0046186917,0.0035801274,0.0044209673,0.002470668],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0023777455,0.0007736342,0.068119355,0.0033279513,0.012097734,0.00047314225,0.0014471306,0.08870691,0.0011737597,0.3253165,0.03856699,0.45761913],"study_design_scores_gemma":[0.0018497276,0.0018269663,0.027400242,0.0017030318,0.004345447,0.0006276006,0.00040074336,0.29351678,0.0039568255,0.5642409,0.09975441,0.00037724996],"about_ca_topic_score_codex":0.0061101224,"about_ca_topic_score_gemma":0.004463773,"teacher_disagreement_score":0.12366636,"about_ca_system_score_codex":0.0018615827,"about_ca_system_score_gemma":0.006587247,"threshold_uncertainty_score":0.65401804},"labels":[],"label_agreement":null},{"id":"W4409232946","doi":"10.1002/pst.70014","title":"Chauhan Weighted Trajectory Analysis of Combined Efficacy and Safety Outcomes for Risk–Benefit Analysis","year":2025,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta; University of Calgary","funders":"","keywords":"Sample size determination; Medicine; Clinical trial; Confidence interval; Statistics; Clinical endpoint; Computer science; Mathematics; Internal medicine","score_opus":0.23610505534910645,"score_gpt":0.4808698903269627,"score_spread":0.24476483497785625,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409232946","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.019007508,0.0005379331,0.97667533,0.0005378916,0.00007667181,0.0011707942,0.00045068734,0.0003154845,0.0012276905],"genre_scores_gemma":[0.508565,0.0005668917,0.4790926,0.000733143,0.00011662379,0.00673766,0.001182401,0.00022145009,0.0027842142],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.96908647,0.02579178,0.00079659274,0.0016441137,0.0022364294,0.00044466936],"domain_scores_gemma":[0.8650435,0.12152113,0.0039027724,0.004977665,0.0037740632,0.000780948],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04745817,0.0020243756,0.0025808292,0.0036609627,0.000788893,0.0024432414,0.0018623928,0.0017345705,0.008598186],"category_scores_gemma":[0.13757627,0.0008168241,0.0038151266,0.0021401616,0.0017689738,0.002385847,0.0028626185,0.004391521,0.0006435709],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014557799,0.00033690478,0.010396511,0.0007467905,0.0020122158,0.00036787454,0.00029964975,0.78028005,0.0020039573,0.118658304,0.004075351,0.07936655],"study_design_scores_gemma":[0.000106829044,0.0002651564,0.00068563886,0.00005344471,0.00014078617,0.0000392444,0.000022767721,0.9532805,0.00043624983,0.043409687,0.0015277452,0.00003197727],"about_ca_topic_score_codex":0.005945408,"about_ca_topic_score_gemma":0.0032900071,"teacher_disagreement_score":0.9525418,"about_ca_system_score_codex":0.0024378968,"about_ca_system_score_gemma":0.005194782,"threshold_uncertainty_score":0.2509858},"labels":[],"label_agreement":null},{"id":"W4413024598","doi":"10.1002/pst.70029","title":"Target Aggregate Data Adjustment Method for Transportability Analysis Utilizing Summary‐Level Data From the Target Population","year":2025,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; University of British Columbia; York University; Centre for Advancing Health Outcomes; Impact; Simon Fraser University","funders":"Canadian Statistical Sciences Institute; Natural Sciences and Engineering Research Council of Canada; Mitacs","keywords":"Censoring (clinical trials); Interpretability; Computer science; Weighting; Causal inference; Population; Statistics; Aggregate data; Statistical inference; Inference; Inverse probability weighting; Econometrics; Data mining; Estimator; Artificial intelligence; Mathematics; Medicine","score_opus":0.513051299481403,"score_gpt":0.5578152948638562,"score_spread":0.044763995382453126,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413024598","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.002129619,0.000105231884,0.9966673,0.00011622587,0.000041060524,0.00019594598,0.00019063849,0.0001960557,0.00035789216],"genre_scores_gemma":[0.20056033,0.00045558618,0.7916574,0.0004374067,0.00023737612,0.002605142,0.0016912526,0.00035660728,0.001998937],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9829488,0.011116539,0.0012178604,0.0022365085,0.0021852187,0.00029506],"domain_scores_gemma":[0.9627758,0.023850178,0.0037411181,0.0062945094,0.0030379489,0.00030052342],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.029208409,0.0011658231,0.0015812577,0.0035108863,0.0008398624,0.0015367848,0.0022035984,0.0012675885,0.006404094],"category_scores_gemma":[0.088023536,0.0005416622,0.0032111318,0.003880801,0.0011465232,0.0025952219,0.0028870527,0.0031830478,0.00076135335],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00071817293,0.00028208442,0.02504283,0.0010838116,0.002340155,0.00033581033,0.00082506635,0.09903526,0.0037283346,0.2345642,0.010248155,0.62179613],"study_design_scores_gemma":[0.00038288083,0.0006064433,0.0109311165,0.00021606438,0.000895532,0.00030053922,0.00017011927,0.6582608,0.0049086064,0.28836796,0.034826018,0.00013393397],"about_ca_topic_score_codex":0.0030450805,"about_ca_topic_score_gemma":0.0019838118,"teacher_disagreement_score":0.029208409,"about_ca_system_score_codex":0.0012427156,"about_ca_system_score_gemma":0.0025926298,"threshold_uncertainty_score":0.15447068},"labels":[],"label_agreement":null},{"id":"W4413114376","doi":"10.1002/pst.70037","title":"Drift Parameter Based Sample Size Determination in Multi‐Stage Bayesian Randomized Clinical Trials","year":2025,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Sample size determination; Bayesian probability; Statistics; Stage (stratigraphy); Randomized controlled trial; Clinical trial; Mathematics; Computer science; Econometrics; Medicine; Internal medicine; Biology","score_opus":0.7611773147160606,"score_gpt":0.6919872169274078,"score_spread":0.06919009778865282,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413114376","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0014009431,0.00058719754,0.99616647,0.0003296068,0.000066808556,0.00049458304,0.00007870635,0.0003268999,0.0005487663],"genre_scores_gemma":[0.066028774,0.000778732,0.9272608,0.00066521455,0.000154293,0.0037638461,0.00023706537,0.00032402299,0.0007872712],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9376535,0.053684108,0.0018997352,0.0024545968,0.0039127073,0.0003953451],"domain_scores_gemma":[0.87272894,0.1133007,0.004746465,0.0051324996,0.0033016305,0.0007896898],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07962111,0.0014976491,0.003470918,0.002548344,0.0008022006,0.002255333,0.003635491,0.0028137472,0.0066505317],"category_scores_gemma":[0.20588537,0.0016618455,0.0019851641,0.0021691076,0.0027522498,0.0030912212,0.003504581,0.004876919,0.0015182297],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002293114,0.00025025965,0.0040909476,0.0025370927,0.00071939355,0.00028272343,0.00055566704,0.18401788,0.003003249,0.27702162,0.011024152,0.51420397],"study_design_scores_gemma":[0.0012553431,0.000611838,0.0009812396,0.0005503463,0.00023314977,0.00023167803,0.000049518032,0.6232255,0.0036505158,0.35730582,0.011804969,0.00010007469],"about_ca_topic_score_codex":0.0014438514,"about_ca_topic_score_gemma":0.0014644398,"teacher_disagreement_score":0.9203789,"about_ca_system_score_codex":0.00205624,"about_ca_system_score_gemma":0.0058845486,"threshold_uncertainty_score":0.42108166},"labels":[],"label_agreement":null},{"id":"W4414140358","doi":"10.1002/pst.70022","title":"Finding the Optimal Number of Splits and Repetitions in Double Cross‐Fitting Targeted Maximum Likelihood Estimators","year":2025,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada; Michael Smith Health Research BC","keywords":"Estimator; Range (aeronautics); Sample size determination; Selection (genetic algorithm); Maximum likelihood; Sample (material); Design of experiments","score_opus":0.03203732618303272,"score_gpt":0.3995152458703245,"score_spread":0.36747791968729177,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414140358","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.10840454,0.00076933595,0.8875655,0.00059583946,0.000089305366,0.000416591,0.00017795384,0.0005238621,0.0014570013],"genre_scores_gemma":[0.48198172,0.00024261333,0.514586,0.00040834412,0.00005593058,0.00096954394,0.00062779913,0.00030598938,0.0008221181],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98575264,0.010029277,0.00089414994,0.001832171,0.0010914559,0.00040030372],"domain_scores_gemma":[0.90283096,0.07766763,0.004007854,0.0093610985,0.0049839183,0.0011484732],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.043749023,0.0011434169,0.0018439491,0.0012632982,0.0016269394,0.0017493016,0.0030821166,0.0027417669,0.003971849],"category_scores_gemma":[0.14779003,0.0009784246,0.001906386,0.0010082931,0.0020782175,0.0046510287,0.003104606,0.0035572725,0.00096281234],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0025570358,0.0012958414,0.06949214,0.0009761069,0.0011321387,0.0009581342,0.0036108005,0.3807889,0.009372749,0.08510199,0.007160607,0.43755347],"study_design_scores_gemma":[0.00033318024,0.00069484644,0.0102117015,0.0005336845,0.0003130521,0.0004189715,0.0010216103,0.88109714,0.008078783,0.09129507,0.00583868,0.00016317326],"about_ca_topic_score_codex":0.0019338063,"about_ca_topic_score_gemma":0.0024524797,"teacher_disagreement_score":0.043749023,"about_ca_system_score_codex":0.00083033333,"about_ca_system_score_gemma":0.0018953544,"threshold_uncertainty_score":0.23136967},"labels":[],"label_agreement":null},{"id":"W4414534411","doi":"10.1002/pst.70031","title":"Adjustment for Inconsistency in Adaptive Phase 2/3 Designs With Dose Optimization","year":2025,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"Canada Research Chairs; Michael Smith Health Research BC","keywords":"Phase (matter); Selection (genetic algorithm); Maximum tolerated dose; Computerized adaptive testing; Adaptive design; Cutoff","score_opus":0.34858204789850666,"score_gpt":0.5624556527276947,"score_spread":0.21387360482918805,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414534411","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.022763526,0.00060679007,0.96911734,0.0004804371,0.0002534366,0.004570592,0.0001262189,0.00033979362,0.0017418301],"genre_scores_gemma":[0.24035518,0.00024449534,0.7452295,0.0010589799,0.00011268208,0.011675184,0.00016317047,0.0001339287,0.0010268019],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.82941663,0.14591081,0.006685346,0.007567842,0.00916469,0.0012546683],"domain_scores_gemma":[0.74678385,0.1996362,0.019312814,0.025855826,0.007371148,0.0010402284],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1775388,0.0022179657,0.0028097965,0.0016264204,0.0011139314,0.0024865933,0.0034814372,0.0035486342,0.005700283],"category_scores_gemma":[0.28753266,0.0012958662,0.0036198052,0.0015220918,0.0032573103,0.0029580859,0.0027431848,0.004639824,0.0006159369],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.025346039,0.0023733685,0.027854191,0.0029147505,0.0060427007,0.0005198927,0.0026956892,0.18499821,0.015355828,0.27201027,0.005659786,0.4542292],"study_design_scores_gemma":[0.008395298,0.017925795,0.014196308,0.00093522476,0.0030233334,0.0003646462,0.00033778982,0.4597661,0.019159691,0.45212927,0.023255998,0.0005105754],"about_ca_topic_score_codex":0.00052636693,"about_ca_topic_score_gemma":0.00073184015,"teacher_disagreement_score":0.8224612,"about_ca_system_score_codex":0.0020028509,"about_ca_system_score_gemma":0.0037306764,"threshold_uncertainty_score":0.9389261},"labels":[],"label_agreement":null},{"id":"W4415152696","doi":"10.1002/pst.70043","title":"Nonparametric Inference for the Covariate‐Adjusted Youden Index and Associated Cut‐Off Points for Three Ordinal Diagnostic Groups","year":2025,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"DoD Alzheimer's Disease Neuroimaging Initiative; National Institute of Biomedical Imaging and Bioengineering; National Institute on Aging; Canadian Institutes of Health Research; National Institutes of Health; Alzheimer's Disease Neuroimaging Initiative; Meso Scale Diagnostics; Takeda Pharmaceutical Company; Bristol-Myers Squibb; Eli Lilly and Company; Northern California Institute for Research and Education; Alzheimer's Drug Discovery Foundation; Simons Foundation; Foundation for the National Institutes of Health","keywords":"Estimator; Confidence interval; Heteroscedasticity; Youden's J statistic; Context (archaeology); Nonparametric statistics; Inference; Point estimation; Statistical inference","score_opus":0.19714102671591952,"score_gpt":0.45609796827218,"score_spread":0.2589569415562605,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4415152696","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0665073,0.0011577092,0.93017423,0.00035136455,0.000035464975,0.000114792296,0.00045837872,0.0002518399,0.0009488861],"genre_scores_gemma":[0.5570386,0.0008644189,0.4388946,0.00020524405,0.000094872805,0.00069078204,0.0016036452,0.00007726879,0.0005305127],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99300283,0.004195151,0.00045105716,0.0011931685,0.0008870478,0.00027073457],"domain_scores_gemma":[0.9234759,0.063157015,0.0053292764,0.0049357438,0.0025731355,0.0005288093],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.022467973,0.0006868833,0.0014994114,0.0036685958,0.0005029102,0.0018606394,0.0019447496,0.0018864035,0.0018831788],"category_scores_gemma":[0.10376765,0.00035052595,0.0010784175,0.0021876507,0.0016861607,0.002186269,0.0017874916,0.0028206916,0.0003598716],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015369405,0.00046050295,0.12908646,0.0009316883,0.00086977665,0.00046716636,0.0012945738,0.105168834,0.0037237653,0.26120025,0.00431374,0.49094635],"study_design_scores_gemma":[0.00029353457,0.0005450838,0.06874285,0.0004965731,0.0003466067,0.00088444067,0.00058915885,0.46088666,0.005695435,0.45472622,0.0065560187,0.00023745529],"about_ca_topic_score_codex":0.0010202874,"about_ca_topic_score_gemma":0.0006166711,"teacher_disagreement_score":0.022467973,"about_ca_system_score_codex":0.0008388938,"about_ca_system_score_gemma":0.0011571578,"threshold_uncertainty_score":0.11882341},"labels":[],"label_agreement":null},{"id":"W4416257149","doi":"10.1002/pst.70049","title":"Great Wall: A Generalized Dose Optimization Design for Drug Combination Trials Maximizing Survival Benefit","year":2025,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"National Center for Advancing Translational Sciences; Ralph W. and Grace M. Showalter Research Trust Fund; National Institute of General Medical Sciences; National Institutes of Health; National Cancer Institute; Indiana Clinical and Translational Sciences Institute","keywords":"Clinical trial; Modular design; Clinical study design; Maximum tolerated dose; Set (abstract data type); Randomization; Optimal design","score_opus":0.7105486185993154,"score_gpt":0.606923267544355,"score_spread":0.1036253510549604,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4416257149","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010307834,0.00042299047,0.98484486,0.00034104384,0.00008561392,0.001287709,0.00021124116,0.0004412,0.0020575668],"genre_scores_gemma":[0.19363125,0.0005021987,0.79398334,0.0006391911,0.0000844646,0.007255601,0.00029415914,0.00016835528,0.003441347],"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.984641,0.01347259,0.00026642508,0.0006577247,0.00067360536,0.0002885543],"domain_scores_gemma":[0.991568,0.006493336,0.0007048298,0.0005078994,0.00046739893,0.00025852423],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01777457,0.0020766882,0.0024321538,0.0015751955,0.00043580917,0.0013045601,0.001537333,0.0017241357,0.0063475594],"category_scores_gemma":[0.019237714,0.0011761956,0.0019834605,0.0010630576,0.0016579971,0.0011806728,0.0020243332,0.0019509725,0.0007442314],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.003965384,0.00030105427,0.0015425185,0.0007451314,0.0006467162,0.00020544788,0.00015298431,0.7357089,0.004386679,0.10807665,0.005701862,0.13856672],"study_design_scores_gemma":[0.001800773,0.0029527212,0.00064887636,0.0001392723,0.00031307832,0.00011295353,0.000028699435,0.9030796,0.002399179,0.07891576,0.009524794,0.00008420517],"about_ca_topic_score_codex":0.0006446046,"about_ca_topic_score_gemma":0.0008253672,"teacher_disagreement_score":0.01777457,"about_ca_system_score_codex":0.0011789392,"about_ca_system_score_gemma":0.0026003208,"threshold_uncertainty_score":0.09400201},"labels":[],"label_agreement":null}]}