{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":180,"total_is_capped":false,"direct_labels_cover":11,"predictions_cover":180,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"query_hash":"5a820786216e","filters":{"venue":"Clinical Trials"}},"results":[{"id":"W2128251606","doi":"10.1177/1740774514558307","title":"Unsuccessful trial accrual and human subjects protections: An empirical analysis of recently closed trials","year":2014,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":387,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; McGill University","funders":"Canadian Institutes of Health Research","keywords":"Accrual; Clinical trial; Medicine; Placebo; Sample size determination; Randomized controlled trial; Family medicine; Internal medicine; Alternative medicine; Accounting; Pathology; Statistics","retraction":null,"screen_n_in":null,"score":{"opus":0.891537863837492,"gpt":0.7465437281129513,"spread":0.1449941357245407,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.3246425,0.0003472408,0.007248066,0.0007707733,0.0002303583,0.0001186839,0.0005153968,0.001854925,0.0008797914],"category_scores_gemma":[0.7760038,0.0002483941,0.002324508,0.001772244,0.001229611,0.000148756,0.0002485628,0.002986158,0.00002436419],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005831374,"about_ca_system_score_gemma":0.0007899248,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001277271,"about_ca_topic_score_gemma":0.0003422554,"domain_scores_codex":[0.9505555,0.03481382,0.01065935,0.00146584,0.001917537,0.0005879328],"domain_scores_gemma":[0.5170389,0.4708009,0.004798692,0.003213318,0.002185265,0.001962917],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.4622098,0.01793992,0.3074745,0.001061149,0.02286366,0.00005553738,0.0005008884,0.0000291543,0.007337197,0.02317346,0.00283726,0.1545175],"study_design_scores_gemma":[0.4143423,0.05294071,0.3533142,0.0008151627,0.03022445,0.000008026571,0.0003282554,0.001406566,0.004215653,0.1294293,0.01179192,0.00118339],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9881576,0.0001157038,0.0008466761,0.003990897,0.0009882691,0.004389782,0.00004985941,0.0001123483,0.001348893],"genre_scores_gemma":[0.9922218,0.0006446588,0.001450324,0.0009168437,0.003838613,0.0001960536,0.00009322396,0.00004899604,0.0005894921],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4513613,"threshold_uncertainty_score":0.9999968,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2164947695","doi":"10.1191/1740774505cn071oa","title":"Determinants of the intracluster correlation coefficient in cluster randomized trials: the case of implementation research","year":2005,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Behavioral Health and Interventions","field":"Psychology","cited_by":284,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Canada Research Chairs","keywords":"Sample size determination; Randomized controlled trial; Cluster (spacecraft); Cluster randomised controlled trial; Statistics; Medicine; Research design; Psychological intervention; Intraclass correlation; Correlation; Clinical psychology; Psychometrics; Mathematics; Internal medicine; Computer science; Psychiatry","retraction":null,"screen_n_in":null,"score":{"opus":0.6941588368323329,"gpt":0.7033108857755119,"spread":0.009152048943179003,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1961856,0.00009451906,0.001447639,0.000131557,0.0001046636,0.00001358724,0.0002531757,0.0002214795,0.001805159],"category_scores_gemma":[0.03241503,0.0000461717,0.0008936304,0.0003299377,0.0004936495,0.00005520875,0.00008632234,0.0005003582,0.00004710491],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004066815,"about_ca_system_score_gemma":0.0001042512,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008421965,"about_ca_topic_score_gemma":0.002085558,"domain_scores_codex":[0.9302582,0.06173604,0.00718467,0.0002641706,0.0002612795,0.0002956357],"domain_scores_gemma":[0.9466915,0.0505369,0.001930237,0.0005223976,0.0002523214,0.00006662842],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.07331623,0.00171482,0.01725575,0.00003421554,0.00009492825,0.00001432402,0.003156233,0.00005704715,0.00002230928,0.002186568,0.01817794,0.8839696],"study_design_scores_gemma":[0.8244048,0.002142694,0.138685,0.0007544782,0.001419524,0.0001350679,0.01090568,0.009972386,0.001161044,0.003178143,0.006853531,0.0003877393],"study_design_candidate":"randomized_trial","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9904155,0.0003530862,0.0003198847,0.001956413,0.002211236,0.004416819,0.00005858449,0.000004949087,0.0002635799],"genre_scores_gemma":[0.9986621,0.0000302064,0.0000762553,0.0002232103,0.0003584316,0.000261328,0.000005053054,0.000009157015,0.0003743155],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8835819,"threshold_uncertainty_score":0.9991073,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2109054675","doi":"10.1177/1740774507087551","title":"Sensible approaches for reducing clinical trial costs","year":2008,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":181,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Population Health Research Institute","funders":"","keywords":"Clinical trial; Medicine; Randomized controlled trial; Payment; Duration (music); Cost–benefit analysis; Operations management; Operations research; Business; Surgery; Engineering; Finance; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9772817153848696,"gpt":0.7307094998449754,"spread":0.2465722155398943,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2205771,0.0006035994,0.006994665,0.0001335079,0.0003987238,0.00008899564,0.0008476813,0.001620519,0.00037915],"category_scores_gemma":[0.9264608,0.0004691704,0.004040637,0.0004116771,0.001453419,0.0001349282,0.0002975665,0.00146484,0.0002090442],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009086327,"about_ca_system_score_gemma":0.0006424894,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000007391456,"about_ca_topic_score_gemma":0.000002300503,"domain_scores_codex":[0.9445345,0.03285517,0.01839224,0.00213342,0.0009653827,0.00111924],"domain_scores_gemma":[0.2431925,0.7495364,0.003834724,0.001972497,0.0004677245,0.0009961684],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.1381226,0.006530461,0.00106249,0.0002443983,0.001584456,0.00009074555,0.00008179196,0.000005859795,0.00002415015,0.1682257,0.2454717,0.4385557],"study_design_scores_gemma":[0.1571015,0.004514127,0.0004044351,0.0001938742,0.001026669,0.00002608096,0.00005367846,0.001040609,0.0002484113,0.8169404,0.01762435,0.0008258701],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.2894155,0.0003494565,0.6130906,0.004505509,0.04986492,0.02535537,0.0007671203,0.001349132,0.01530234],"genre_scores_gemma":[0.03545307,0.0005034011,0.9384357,0.0008664038,0.02220589,0.0005596886,0.00001559493,0.0001767746,0.001783456],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7166812,"threshold_uncertainty_score":0.999776,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2076160163","doi":"10.1177/1740774507087704","title":"Specific barriers to the conduct of randomized trials","year":2008,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":180,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Boehringer Ingelheim (Canada); Health Canada; McMaster University","funders":"","keywords":"Protocol (science); Risk analysis (engineering); Clinical trial; Randomized controlled trial; Psychological intervention; Quality (philosophy); Informed consent; Process (computing); Medicine; Computer science; Alternative medicine; Nursing","retraction":null,"screen_n_in":null,"score":{"opus":0.9468991071094657,"gpt":0.7278013918630317,"spread":0.219097715246434,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5653369,0.0002334954,0.009461583,0.0001451642,0.0001781726,0.00002439501,0.0006868995,0.0008168729,0.00466055],"category_scores_gemma":[0.9740935,0.0001145428,0.004247778,0.0005061829,0.002723915,0.00003836969,0.0002309728,0.002506241,0.0005932798],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003529556,"about_ca_system_score_gemma":0.001951743,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002122665,"about_ca_topic_score_gemma":0.000004435869,"domain_scores_codex":[0.8889698,0.0799332,0.02463301,0.001517492,0.003984903,0.0009615946],"domain_scores_gemma":[0.0344504,0.957649,0.002303935,0.002247014,0.00157714,0.001772493],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.6298826,0.0008175463,0.00312326,0.0001793177,0.003126637,0.0001583964,0.0008651421,0.00001342101,0.003377517,0.02994947,0.3115691,0.01693753],"study_design_scores_gemma":[0.6898704,0.002568768,0.001829567,0.0007669757,0.001667519,0.00004361242,0.0004397287,0.00004185398,0.01111238,0.06488031,0.2263283,0.0004506442],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6730308,0.006854626,0.003459989,0.2398675,0.01240405,0.02647367,0.0002155844,0.0002161365,0.03747758],"genre_scores_gemma":[0.9483272,0.02197558,0.004944913,0.009479155,0.006440334,0.0002748693,0.000012747,0.00006274001,0.008482456],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8777158,"threshold_uncertainty_score":0.9999901,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2162300921","doi":"10.1177/1740774508096474","title":"A comprehensive comparison of the continual reassessment method to the standard 3 + 3 dose escalation scheme in Phase I dose-finding studies","year":2008,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":164,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences","funders":"National Cancer Institute","keywords":"De-escalation; Phase (matter); Medicine; Medical physics; Computer science; Intensive care medicine; Chemistry","retraction":null,"screen_n_in":null,"score":{"opus":0.9337735777746451,"gpt":0.7655472026703487,"spread":0.1682263751042964,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0734451,0.0003605547,0.004579008,0.0001000131,0.0002524055,0.00003458064,0.0007539479,0.0003456363,0.00009964092],"category_scores_gemma":[0.5870852,0.0001927072,0.000887515,0.0007630432,0.0007348828,0.00006576611,0.0005334856,0.0009996628,0.00001803384],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001343347,"about_ca_system_score_gemma":0.0002473265,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001095761,"about_ca_topic_score_gemma":0.0000146413,"domain_scores_codex":[0.9517269,0.03583797,0.009746145,0.0008517309,0.001298318,0.0005389221],"domain_scores_gemma":[0.454271,0.540097,0.003245986,0.001326534,0.0008113023,0.0002481103],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.02770601,0.01031753,0.05351938,0.000853024,0.004215291,0.00007354922,0.01041197,0.0002887315,0.004835156,0.1606826,0.07969908,0.6473976],"study_design_scores_gemma":[0.05736508,0.00470884,0.05832715,0.001924236,0.00153325,0.00001451684,0.00583055,0.002540961,0.009088401,0.8342251,0.02329837,0.001143513],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.7371926,0.0005417982,0.2425062,0.006942712,0.004503164,0.007366962,0.0004055517,0.0001052337,0.0004357506],"genre_scores_gemma":[0.304987,0.0001286857,0.6933195,0.0004924602,0.0007953374,0.0001566382,0.000001579965,0.00003701807,0.00008175802],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6735425,"threshold_uncertainty_score":0.9540833,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2122523415","doi":"10.1191/1740774505cn085oa","title":"Randomized controlled trials of aprotinin in cardiac surgery: could clinical equipoise have stopped the bleeding?","year":2005,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Blood transfusion and management","field":"Medicine","cited_by":164,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; McGill University Health Centre","funders":"","keywords":"Aprotinin; Medicine; Randomized controlled trial; Clinical equipoise; Clinical trial; Cardiac surgery; MEDLINE; Antithrombins; Surgery; Intensive care medicine; Internal medicine; Heparin","retraction":null,"screen_n_in":null,"score":{"opus":0.3943694536107366,"gpt":0.5276767485687768,"spread":0.1333072949580402,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.5306896,0.0003993301,0.02770207,0.0002937098,0.00006302094,0.00004290104,0.0002892806,0.0006208985,0.0009640559],"category_scores_gemma":[0.3905054,0.0001973197,0.01256701,0.0002945306,0.0007135813,0.00006654873,0.0001074163,0.000974206,0.00006566676],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003179591,"about_ca_system_score_gemma":0.0004757247,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005229582,"about_ca_topic_score_gemma":0.00002370953,"domain_scores_codex":[0.9021235,0.06706441,0.02802017,0.0009413133,0.001180878,0.000669719],"domain_scores_gemma":[0.6098915,0.3819683,0.005885844,0.001283234,0.0004195347,0.0005515381],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.8347263,0.002623563,0.01430602,0.0001058902,0.006313432,0.00002278615,0.0001609482,0.000007583094,0.0001414094,0.001545697,0.03965029,0.1003961],"study_design_scores_gemma":[0.930618,0.0004282811,0.002620736,0.0005070057,0.009141359,0.000001574671,0.0002655153,0.0004587037,0.0003405712,0.0004866652,0.05486419,0.0002673717],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7439955,0.01100915,0.0007765248,0.1437552,0.01218239,0.0683734,0.0001492177,0.0002948942,0.01946371],"genre_scores_gemma":[0.9716297,0.01456493,0.0005098473,0.005234866,0.004406486,0.00110732,0.00002946487,0.00005095035,0.002466375],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3149039,"threshold_uncertainty_score":0.9999492,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2103725571","doi":"10.1177/1740774512450097","title":"The role for pragmatic randomized controlled trials (pRCTs) in comparative effectiveness research","year":2012,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":162,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Health Sciences Centre; Sunnybrook Health Science Centre","funders":"H2020 European Research Council; Blue Cross and Blue Shield Association; Canadian Institutes of Health Research; Patient-Centered Outcomes Research Institute","keywords":"Generalizability theory; Comparative effectiveness research; Randomized controlled trial; Internal validity; Psychological intervention; External validity; Health care; Medicine; Cost effectiveness; Alternative medicine; Management science; Risk analysis (engineering); Psychology; Nursing; Political science; Economics; Social psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.8997568960764607,"gpt":0.7007489774770788,"spread":0.1990079185993819,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.940848,0.0002786579,0.02012279,0.0003986986,0.0004563195,0.0002335052,0.0005100927,0.0003963653,0.0002052413],"category_scores_gemma":[0.8750094,0.0001878496,0.00244721,0.00030611,0.0004853517,0.0003709694,0.00007933707,0.0005769464,0.001326206],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003693635,"about_ca_system_score_gemma":0.0003506453,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001542254,"about_ca_topic_score_gemma":0.00004862354,"domain_scores_codex":[0.5740873,0.361329,0.06108235,0.001218893,0.0004462796,0.001836102],"domain_scores_gemma":[0.07757646,0.9093239,0.01162912,0.0007825448,0.0003261883,0.0003617515],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.1829434,0.0009673058,0.01233967,0.0003452979,0.002261216,2.362219e-7,0.002204366,0.00008033396,0.000006247486,0.7812392,0.01625178,0.001360882],"study_design_scores_gemma":[0.5638541,0.0002355773,0.003803371,0.000273377,0.0001431305,6.069934e-7,0.001838051,0.007814661,0.00001795078,0.3597831,0.06187214,0.0003639881],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5330954,0.1356723,0.02419079,0.07942301,0.02116283,0.1758989,0.001021273,0.000194134,0.02934145],"genre_scores_gemma":[0.9780283,0.000959269,0.001068201,0.001188112,0.003484763,0.01462519,0.00002871922,0.00004032105,0.0005770996],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5479949,"threshold_uncertainty_score":0.9994514,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2139118963","doi":"10.1177/1740774511414444","title":"Altruism among participants in cancer clinical trials","year":2011,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":147,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"","keywords":"Clinical trial; Medicine; Altruism (biology); Cancer; Psychology; Clinical psychology; Social psychology; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9883663841010735,"gpt":0.8144164904305321,"spread":0.1739498936705414,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity","insufficient_payload"],"category_scores_codex":[0.5624104,0.0004084424,0.007315355,0.000251927,0.00008792868,0.00005264118,0.0008010991,0.002898217,0.006883673],"category_scores_gemma":[0.8902321,0.0002884935,0.002709108,0.000609844,0.002020015,0.0001539269,0.0004348816,0.007309954,0.000890874],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001154868,"about_ca_system_score_gemma":0.002379879,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007171375,"about_ca_topic_score_gemma":0.00124451,"domain_scores_codex":[0.9143785,0.05172676,0.02748279,0.002302031,0.002226929,0.001883025],"domain_scores_gemma":[0.1651225,0.8192881,0.005568373,0.004028682,0.001568669,0.004423698],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.006033427,0.002992457,0.9078963,0.0001260766,0.0006064978,0.0002721628,0.0001900591,6.358222e-7,0.00004626263,0.002682188,0.005741362,0.07341251],"study_design_scores_gemma":[0.02054379,0.0022864,0.9070206,0.001321674,0.0008780728,0.000002253934,0.000153605,0.0002116852,0.0005474068,0.06141097,0.005198606,0.000424951],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9713603,0.0009881039,0.0001129524,0.005519891,0.006425514,0.003470721,0.00005113456,0.0001465448,0.01192487],"genre_scores_gemma":[0.9697875,0.01382569,0.001675258,0.004487046,0.005539244,0.0004248069,0.00001078784,0.0000889521,0.004160715],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7675613,"threshold_uncertainty_score":0.9999567,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2077506728","doi":"10.1177/1740774510374091","title":"Assessing contamination and compliance in the prostate component of the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial","year":2010,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Prostate Cancer Diagnosis and Treatment","field":"Medicine","cited_by":141,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Public Health Ontario; University of Toronto","funders":"","keywords":"Medicine; Prostate; Prostate cancer; Prostate cancer screening; Incidence (geometry); Prostate-specific antigen; Internal medicine; Oncology; Cancer screening; Gynecology; Epidemiology; Rectal examination; Cancer","retraction":null,"screen_n_in":null,"score":{"opus":0.2756276787216155,"gpt":0.5110173590625215,"spread":0.2353896803409061,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007654836,0.0001591441,0.0008626909,0.00003822035,0.0001270227,0.00009912233,0.000100308,0.0001279492,0.00001934472],"category_scores_gemma":[0.002993971,0.00008186379,0.0001357726,0.0001650918,0.0004164001,0.000145207,0.00006852055,0.0004586172,3.746307e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003186501,"about_ca_system_score_gemma":0.0001780413,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002014103,"about_ca_topic_score_gemma":0.0001844101,"domain_scores_codex":[0.9968669,0.001077958,0.001251776,0.0003452617,0.0002602268,0.0001979199],"domain_scores_gemma":[0.9949642,0.003878946,0.0006861145,0.0002614751,0.0001193901,0.00008990682],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01891552,0.001154057,0.7880283,0.0002245264,0.0002693209,0.00002245981,0.001173807,0.000003332779,0.003817673,0.0005454352,0.0006962321,0.1851493],"study_design_scores_gemma":[0.06175476,0.0007917883,0.9324118,0.000692061,0.0003018645,0.000009845406,0.0002361245,0.0006014989,0.001234606,0.000254904,0.001602927,0.0001078551],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.987054,0.0008750566,0.00002522899,0.006623407,0.0008126397,0.004409533,0.00005315295,0.00001135978,0.0001355959],"genre_scores_gemma":[0.9973499,0.0009848388,0.0003808538,0.0005164184,0.0003001783,0.0004062106,0.00001584299,0.00001347682,0.00003225875],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1850415,"threshold_uncertainty_score":0.3584279,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2006780292","doi":"10.1177/1740774507081223","title":"The intention-to-treat approach in randomized controlled trials: Are authors saying what they do and doing what they say?","year":2007,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":135,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Royal Victoria Hospital; McGill University; Centre Hospitalier Universitaire Sainte-Justine","funders":"","keywords":"Randomized controlled trial; Missing data; Medicine; Concordance; Sample size determination; Intention-to-treat analysis; Meta-analysis; Statistics; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.8477915287045936,"gpt":0.6338580029690107,"spread":0.2139335257355829,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","scholarly_communication"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.9603344,0.0005974422,0.04220973,0.000675508,0.0004277286,0.01197982,0.001868172,0.0004288966,0.0006678794],"category_scores_gemma":[0.9090601,0.0001890824,0.01529425,0.0009024368,0.0002538116,0.0009740467,0.0002987007,0.0005886022,0.0004380163],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000043788,"about_ca_system_score_gemma":0.00008778431,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001855844,"about_ca_topic_score_gemma":0.0001231696,"domain_scores_codex":[0.1816791,0.6707908,0.1339674,0.003671428,0.00861806,0.001273304],"domain_scores_gemma":[0.03226168,0.9277884,0.03407174,0.003760105,0.0014232,0.0006948105],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.09544805,0.0003279733,0.004350306,0.00006265855,0.002971195,0.00001532852,0.002537627,0.0001123648,0.00004040452,0.006093737,0.003695327,0.8843451],"study_design_scores_gemma":[0.6652561,0.0002256535,0.003182029,0.002979322,0.009408755,0.00002022774,0.1032534,0.009662788,0.00001554781,0.1823649,0.02243785,0.001193351],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5249392,0.1716388,0.1745321,0.02552997,0.01865739,0.06806274,0.00002459529,0.00008047752,0.01653467],"genre_scores_gemma":[0.9678497,0.01888064,0.004979003,0.001223698,0.001064251,0.0007190862,0.000004688364,0.00003884432,0.005240096],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8831517,"threshold_uncertainty_score":0.9945673,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2146514939","doi":"10.1177/1740774507087972","title":"Do we need to adjudicate major clinical events?","year":2008,"lang":"en","type":"review","venue":"Clinical Trials","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":120,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Population Health Research Institute","funders":"","keywords":"Adjudication; Medicine; Clinical trial; Political science; Law; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.949609721965297,"gpt":0.7577542437055319,"spread":0.1918554782597651,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity","insufficient_payload"],"category_scores_codex":[0.1968734,0.00223109,0.04207572,0.0005976148,0.0002883776,0.0001732754,0.00399044,0.006232312,0.003579156],"category_scores_gemma":[0.9282384,0.001580718,0.01797415,0.001810314,0.00112288,0.000110624,0.001852693,0.00605414,0.01143136],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002248961,"about_ca_system_score_gemma":0.001840693,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000004581627,"about_ca_topic_score_gemma":0.000001522102,"domain_scores_codex":[0.8190073,0.1213037,0.05058715,0.004732025,0.002328589,0.002041149],"domain_scores_gemma":[0.3175006,0.6640069,0.01061216,0.004570367,0.0004959747,0.002813987],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000342002,0.001271193,0.00001407195,0.003639254,0.00200418,0.0001218637,0.000008252141,3.104777e-8,1.849599e-8,0.008381782,0.1878152,0.7964022],"study_design_scores_gemma":[0.002902809,0.0007425545,0.000006352809,0.01510738,0.00512924,0.0000361685,0.000005340362,9.759846e-7,1.837794e-7,0.3267646,0.6481919,0.001112493],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000002401706,0.9473287,0.01642584,0.001639252,0.01976259,0.01124441,0.001392435,0.0005956726,0.001608692],"genre_scores_gemma":[1.892655e-7,0.6578504,0.3256548,0.0006650217,0.01298726,0.0007372887,0.00001979131,0.000392318,0.001692863],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7952897,"threshold_uncertainty_score":0.9990429,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2069943709","doi":"10.1177/1740774507083567","title":"Meta-analysis of longitudinal studies","year":2007,"lang":"en","type":"review","venue":"Clinical Trials","topic":"Neurological disorders and treatments","field":"Medicine","cited_by":119,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Royal Victoria Hospital; Montreal General Hospital; Montreal Children's Hospital; McGill University; Canadian Association of Radiation Oncology","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Random effects model; Multivariate statistics; Statistics; Meta-analysis; Multivariate analysis; Econometrics; Contrast (vision); Mixed model; Mathematics; Computer science; Medicine; Artificial intelligence; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9607841748425348,"gpt":0.7121983436394579,"spread":0.2485858312030769,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaepi_broad"],"category_scores_codex":[0.01414739,0.0004571806,0.03588398,0.0004171082,0.00002845589,0.00000692914,0.0001652768,0.0005879246,0.001112096],"category_scores_gemma":[0.01924454,0.0002081412,0.03211141,0.001116971,0.0003031277,0.00001693236,0.0001062198,0.0004879814,0.0000941934],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000184971,"about_ca_system_score_gemma":0.0001012344,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00000238984,"about_ca_topic_score_gemma":0.000002607427,"domain_scores_codex":[0.989875,0.002810658,0.005923813,0.000746269,0.0004018107,0.0002424693],"domain_scores_gemma":[0.9673111,0.02906836,0.002435593,0.0007808763,0.0001996008,0.0002044599],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.00007377841,0.0005613681,0.0001226491,0.002016899,0.6132587,0.0001125373,0.000001118933,6.592546e-8,2.080808e-9,0.00004677679,0.000516039,0.3832901],"study_design_scores_gemma":[0.0004638546,0.0004519583,0.00006053932,0.0001392993,0.6061431,0.000001226018,0.000001404819,9.683725e-8,4.240709e-8,0.0002064079,0.3924509,0.00008124293],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000007464969,0.9974144,0.00001531997,0.0001018817,0.0002880456,0.001303662,0.0001458754,0.00002422368,0.0006991643],"genre_scores_gemma":[0.00001791695,0.9981062,0.0001716413,0.0003990803,0.0002086879,0.00006404032,0.00009825974,0.0000234523,0.0009107213],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.3919349,"threshold_uncertainty_score":0.999801,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2030531780","doi":"10.1177/1740774509105223","title":"Evaluating the benefit of event adjudication of cardiovascular outcomes in large simple RCTs","year":2009,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":114,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; Population Health Research Institute","funders":"","keywords":"Adjudication; Medicine; Randomized controlled trial; Event (particle physics); Simple (philosophy); Cardiovascular event; Medical physics; Intensive care medicine; Internal medicine; Myocardial infarction; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.7228107795496578,"gpt":0.6578663880181861,"spread":0.06494439153147169,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04148861,0.0001204065,0.00157485,0.00005631385,0.00002440674,0.000004616883,0.0002849875,0.0001519986,0.00004339918],"category_scores_gemma":[0.0974957,0.00007593197,0.0007892295,0.0001870607,0.00006439361,0.00006316936,0.00006211902,0.0002084738,0.000002613873],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002717724,"about_ca_system_score_gemma":0.00005258632,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001328236,"about_ca_topic_score_gemma":0.000007431685,"domain_scores_codex":[0.9946315,0.001676383,0.002818096,0.0002225986,0.0004715308,0.0001799442],"domain_scores_gemma":[0.9833719,0.01444456,0.001112627,0.0008466934,0.0001864272,0.00003781762],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004217911,0.002922937,0.03208074,0.0002082142,0.00120642,0.000003920766,0.0006078057,0.001444334,0.00373848,0.3382007,0.0004943178,0.6186704],"study_design_scores_gemma":[0.001540314,0.0006059519,0.03072087,0.0001225611,0.0002595234,3.130607e-7,0.00005517758,0.0005049494,0.005770919,0.9601352,0.000177887,0.0001063398],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9614605,0.0005552793,0.03554805,0.000443931,0.0001048613,0.001482416,0.00002889035,0.00006829565,0.000307764],"genre_scores_gemma":[0.9837435,0.0001816226,0.01584774,0.00008900304,0.00007014132,0.00003421977,0.000004131221,0.00001010227,0.00001953867],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6219345,"threshold_uncertainty_score":0.9869892,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2127023478","doi":"10.1177/1740774514563583","title":"Are outcome-adaptive allocation trials ethical?","year":2015,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":109,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"Canadian Institutes of Health Research; Genome Alberta; Genome Canada","keywords":"Outcome (game theory); Randomization; Scrutiny; Informed consent; Clinical trial; Cornerstone; Randomized controlled trial; Clinical equipoise; Psychology; Medicine; Computer science; Actuarial science; Economics; Political science; Alternative medicine; Law; Microeconomics","retraction":null,"screen_n_in":null,"score":{"opus":0.9820105155223825,"gpt":0.7879909824342786,"spread":0.1940195330881039,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.5268925,0.0003060322,0.005049454,0.0001631492,0.00009388588,0.00008705398,0.0005341002,0.003792495,0.00056774],"category_scores_gemma":[0.9695777,0.0002069787,0.001762296,0.0004596366,0.0009296762,0.00008206509,0.0003168121,0.008853129,0.002074158],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001851153,"about_ca_system_score_gemma":0.00181919,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003108558,"about_ca_topic_score_gemma":0.00004887083,"domain_scores_codex":[0.9523961,0.0296305,0.01261446,0.001334147,0.003302637,0.0007221746],"domain_scores_gemma":[0.09392957,0.8849778,0.007272156,0.003360515,0.006457948,0.004002044],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.04366769,0.005295697,0.3242319,0.0006857984,0.003221165,0.0006470616,0.0005119478,0.00003440693,0.0004706817,0.1808207,0.3297569,0.110656],"study_design_scores_gemma":[0.06132311,0.007813513,0.09627451,0.001732881,0.00256337,0.00003088959,0.001885937,0.0007142398,0.00124235,0.6762661,0.1490073,0.001145727],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.1894174,0.002629882,0.02150458,0.690792,0.01940778,0.01442443,0.0002292433,0.001180297,0.06041437],"genre_scores_gemma":[0.9310611,0.0009643937,0.01005356,0.03692946,0.008549687,0.0001921819,0.00003954685,0.00009542486,0.01211464],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8584665,"threshold_uncertainty_score":0.9987028,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2909678312","doi":"10.1177/1740774518820060","title":"Conducting clinical trials—costs, impacts, and the value of clinical trials networks: A scoping review","year":2019,"lang":"en","type":"review","venue":"Clinical Trials","topic":"Pharmaceutical industry and healthcare","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":107,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University; Ontario Institute for Cancer Research; Simon Fraser University; Canadian Centre for Applied Research in Cancer Control","funders":"Canadian Cancer Society Research Institute; Ontario Institute for Cancer Research","keywords":"Clinical trial; Medicine; Value (mathematics); Intensive care medicine; Computer science; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9815948427230464,"gpt":0.8194658296390914,"spread":0.162129013083955,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity"],"category_scores_codex":[0.8382099,0.001890031,0.06717128,0.0002316879,0.0005921631,0.0001673504,0.00228527,0.01210997,0.003121882],"category_scores_gemma":[0.7886487,0.001066435,0.01989873,0.001030404,0.004362251,0.0002693396,0.001165159,0.02534507,0.0003868689],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001375689,"about_ca_system_score_gemma":0.006187805,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002477224,"about_ca_topic_score_gemma":0.000006834634,"domain_scores_codex":[0.2573027,0.6540617,0.08263853,0.003018053,0.0009583434,0.002020725],"domain_scores_gemma":[0.09082586,0.855844,0.04647192,0.002570311,0.000830233,0.003457702],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.003914006,0.0007088713,0.0002913508,0.09207278,0.006205379,0.00003711641,0.0000144278,0.000003875131,1.968203e-7,0.001658614,0.02911639,0.865977],"study_design_scores_gemma":[0.01841765,0.0005929047,0.00000725039,0.226959,0.04274944,0.00005823029,0.0000303635,0.0003425906,0.000001488952,0.0001942656,0.7097943,0.0008524139],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00003953743,0.9365181,0.00007693225,0.004200329,0.02947531,0.02798245,0.0007809983,0.0001386846,0.0007876308],"genre_scores_gemma":[0.0000572201,0.9553964,0.0001424809,0.01666776,0.02626458,0.0006896456,0.0002530355,0.0001858892,0.0003429862],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8651246,"threshold_uncertainty_score":0.9994462,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2106121516","doi":"10.1177/1740774513504151","title":"Beyond intention to treat: What is the right question?","year":2013,"lang":"en","type":"review","venue":"Clinical Trials","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":106,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; Jewish General Hospital","funders":"","keywords":"Randomized controlled trial; Causal inference; Medicine; Terminology; Treatment effect; Treatment and control groups; Population; Protocol (science); Instrumental variable; Psychology; Alternative medicine; Statistics; Mathematics; Surgery; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.7023504462850969,"gpt":0.6546002616242445,"spread":0.04775018466085246,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.02151103,0.0006152574,0.005549182,0.0001469709,0.0001139013,0.000383292,0.0009220301,0.001044057,0.001438549],"category_scores_gemma":[0.04276428,0.0003068968,0.002332441,0.0003426438,0.0002100074,0.0003980063,0.0002959203,0.001029817,0.002251433],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001377148,"about_ca_system_score_gemma":0.00015932,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000007516914,"about_ca_topic_score_gemma":0.000003730651,"domain_scores_codex":[0.9866417,0.006621045,0.005215742,0.0007736846,0.000381403,0.0003663939],"domain_scores_gemma":[0.9503146,0.04483816,0.002540837,0.001721407,0.0003191659,0.0002658517],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000006458411,0.00009710167,1.924166e-7,0.001579132,0.0001109281,0.000002511177,0.00001643383,3.431236e-9,1.940755e-7,0.05069252,0.04435958,0.9031349],"study_design_scores_gemma":[0.0001135496,0.0001289429,9.960478e-8,0.01094481,0.001058509,0.000004811652,0.00001062667,3.204304e-7,0.000007970327,0.3833423,0.6041582,0.0002299164],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000001178878,0.978678,0.01007159,0.0008911158,0.00163449,0.006173721,0.00006034501,0.0003889028,0.002100589],"genre_scores_gemma":[3.490176e-7,0.9746017,0.0164358,0.0007415072,0.001108606,0.001451899,0.00004898758,0.00009441139,0.005516745],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.902905,"threshold_uncertainty_score":0.9999383,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2313243722","doi":"10.1177/1740774516634316","title":"Substantial risks associated with few clusters in cluster randomized and stepped wedge designs","year":2016,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":97,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Women's College Hospital; University of Toronto; Ottawa Hospital; University of Ottawa","funders":"","keywords":"Generalizability theory; Sample size determination; Cluster (spacecraft); Statistics; Statistical power; Computer science; Research design; Wedge (geometry); Type I and type II errors; Correlation; Cluster size; Contrast (vision); Econometrics; Mathematics; Artificial intelligence","retraction":null,"screen_n_in":null,"score":{"opus":0.8556771333791159,"gpt":0.6453227649648124,"spread":0.2103543684143034,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1765956,0.0004857042,0.006486426,0.0001638444,0.00008609622,0.00009452566,0.0004013958,0.0008843074,0.0005111834],"category_scores_gemma":[0.8608698,0.0002510527,0.0007792021,0.0002964486,0.001593648,0.000141547,0.0001931129,0.0006599027,0.00005665598],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008989144,"about_ca_system_score_gemma":0.000176624,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001314386,"about_ca_topic_score_gemma":0.00007126064,"domain_scores_codex":[0.9333732,0.05697376,0.007281194,0.001023172,0.0006365482,0.0007121292],"domain_scores_gemma":[0.2050551,0.7911507,0.00235307,0.000729624,0.0002286913,0.0004827986],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.7965221,0.002965973,0.01345488,0.0002300483,0.003551735,0.000258401,0.0003132027,0.000003924748,0.0003104976,0.04644231,0.01036482,0.1255821],"study_design_scores_gemma":[0.4828572,0.0005089731,0.001742313,0.0009074343,0.0009445113,0.000003195166,0.00002671492,0.0002165221,0.0001047883,0.5121175,0.00008286176,0.0004879933],"study_design_candidate":"randomized_trial","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3853274,0.0001561375,0.5945581,0.004279857,0.003008818,0.0104568,0.000271885,0.0003246822,0.00161642],"genre_scores_gemma":[0.6898203,0.0006984149,0.3055158,0.001064376,0.001265153,0.0004202817,0.000003951029,0.000165454,0.001046298],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7341769,"threshold_uncertainty_score":0.9999942,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2914452574","doi":"10.1177/1740774518824539","title":"Challenges and solutions in the design and execution of the PROSPECT Phase II/III neoadjuvant rectal cancer trial (NCCTG N1048/Alliance)","year":2019,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Colorectal Cancer Surgical Treatments","field":"Medicine","cited_by":73,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"National Cancer Institute","keywords":"Medicine; Colorectal cancer; FOLFOX; Oxaliplatin; Oncology; Neoadjuvant therapy; Clinical trial; Clinical endpoint; Randomized controlled trial; Internal medicine; Surgery; Cancer","retraction":null,"screen_n_in":null,"score":{"opus":0.4787653320178136,"gpt":0.5176457104897636,"spread":0.03888037847195003,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00892259,0.0001737088,0.001118892,0.00003224507,0.00009069868,0.00001358862,0.0001287142,0.0001947311,0.00009235392],"category_scores_gemma":[0.003760066,0.00008599008,0.0002805143,0.0001776305,0.0002859596,0.00006239757,0.0001066487,0.0003423985,0.000003123226],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006950585,"about_ca_system_score_gemma":0.0002068517,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005341536,"about_ca_topic_score_gemma":0.0000642016,"domain_scores_codex":[0.9959179,0.001917119,0.001169462,0.0004388254,0.0003107751,0.0002458882],"domain_scores_gemma":[0.9940429,0.005002838,0.0004038963,0.000389724,0.00006281453,0.00009781995],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.2664855,0.00598069,0.007013865,0.0001592237,0.0006923648,0.00003840899,0.002463799,0.00001843452,0.002097673,0.001558077,0.001285685,0.7122062],"study_design_scores_gemma":[0.7010859,0.03984835,0.2320502,0.001844607,0.001789962,0.00007166409,0.0008325209,0.001230183,0.002059413,0.00669341,0.01189694,0.0005968119],"study_design_candidate":"randomized_trial","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9741163,0.01272006,0.00001594809,0.005410385,0.001035463,0.005647333,0.00001756576,0.00001776657,0.001019193],"genre_scores_gemma":[0.98405,0.01458733,0.0001085595,0.0001800951,0.0004835352,0.0003254766,0.000002465202,0.00001191614,0.0002506456],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7116094,"threshold_uncertainty_score":0.4501422,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2106029785","doi":"10.1177/1740774508089459","title":"Maintaining confidentiality of interim data to enhance trial integrity and credibility","year":2008,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":71,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Health Research Council of New Zealand; National Institute of Allergy and Infectious Diseases; Canadian Institutes of Health Research","keywords":"Interim; Safeguarding; Interim analysis; Credibility; Confidentiality; Data monitoring committee; Harm; Clinical trial; Medicine; Psychology; Computer science; Computer security; Political science; Social psychology; Nursing; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.9540912610429189,"gpt":0.7680111871722448,"spread":0.186080073870674,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2117597,0.0001854988,0.002470759,0.00008255448,0.00008958994,0.00002637708,0.0008474133,0.0007868806,0.0005672659],"category_scores_gemma":[0.796498,0.0001439068,0.0003916056,0.0002372634,0.001842845,0.0001129621,0.001741062,0.00424773,0.00004976436],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004157194,"about_ca_system_score_gemma":0.001366384,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002972572,"about_ca_topic_score_gemma":0.0001295504,"domain_scores_codex":[0.9832373,0.007013449,0.006416992,0.001536265,0.001355242,0.0004407687],"domain_scores_gemma":[0.4495883,0.5317602,0.002708296,0.010136,0.003034247,0.002772931],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"observational","study_design_scores_codex":[0.7040665,0.008812128,0.1265687,0.001434984,0.001711172,0.0002841157,0.001247025,9.539903e-7,0.009714678,0.01232313,0.02968515,0.1041515],"study_design_scores_gemma":[0.3312933,0.0537416,0.3566649,0.005611424,0.002099153,0.0001626854,0.002123618,0.001149236,0.03298901,0.1700676,0.04202836,0.002068998],"study_design_candidate":"randomized_trial","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9842522,0.0001371399,0.002951343,0.00761163,0.001501599,0.002038819,0.0001573552,0.00004683674,0.001303114],"genre_scores_gemma":[0.9881817,0.001223334,0.006642829,0.001346185,0.002012283,0.00002244527,0.00003734258,0.00001853868,0.0005154078],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5847383,"threshold_uncertainty_score":0.9980495,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"methods","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3037121760","doi":"10.1177/1740774520943846","title":"Anti-Thrombotic Therapy to Ameliorate Complications of COVID-19 (ATTACC): Study design and methodology for an international, adaptive Bayesian randomized controlled trial","year":2020,"lang":"en","type":"article","venue":"Clinical Trials","topic":"COVID-19 Clinical Research Studies","field":"Medicine","cited_by":68,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Research Institute in Oncology and Hematology; Ozmosis Research (Canada); Université Laval; University of British Columbia; Toronto General Hospital; McMaster University; Thrombosis and Atherosclerosis Research Institute; CancerCare Manitoba; Ottawa Hospital; University Health Network; Health Sciences Centre; McGill University; McGill University Health Centre; St. Michael's Hospital; Jewish General Hospital; Sunnybrook Health Science Centre; Hamilton Health Sciences; University of Toronto; University of Manitoba","funders":"Peter Munk Cardiac Centre, University Health Network; Research Manitoba","keywords":"Coronavirus disease 2019 (COVID-19); Randomized controlled trial; Medicine; 2019-20 coronavirus outbreak; Bayesian probability; Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); Intensive care medicine; Internal medicine; Statistics; Virology; Mathematics; Disease","retraction":null,"screen_n_in":null,"score":{"opus":0.8215050255771472,"gpt":0.6769166508533498,"spread":0.1445883747237974,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.09188976,0.0002930896,0.008588765,0.0002040885,0.0001362524,0.00004206234,0.0004221787,0.0002075918,0.0001128801],"category_scores_gemma":[0.8258093,0.000190326,0.001203327,0.0003431246,0.0007279692,0.00007960996,0.0001778915,0.0003641476,0.00000824472],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005065494,"about_ca_system_score_gemma":0.0009129768,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004826563,"about_ca_topic_score_gemma":0.000003566196,"domain_scores_codex":[0.9627957,0.03052295,0.004867502,0.0009733094,0.0005282311,0.0003122881],"domain_scores_gemma":[0.2504927,0.743519,0.002071149,0.0008380018,0.00131823,0.001760924],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.9863217,0.002399154,0.0007765696,0.00004318601,0.003851451,0.000005342578,0.0007087584,0.0003213832,0.0005609252,0.0005133312,0.0007570381,0.003741093],"study_design_scores_gemma":[0.9601802,0.02395388,0.001226339,0.0000237975,0.001272934,2.731838e-7,0.0003664611,0.008733911,0.00004527607,0.003049148,0.0009837685,0.0001639628],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.07788698,0.0001727786,0.7972087,0.06134941,0.0006094855,0.06258193,0.00007656242,0.00008636955,0.00002780666],"genre_scores_gemma":[0.9147928,0.001516474,0.06063879,0.01708072,0.001146412,0.004734185,0.00001801843,0.00003887946,0.00003376234],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8369058,"threshold_uncertainty_score":0.9350906,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3165583239","doi":"10.1177/17407745211020852","title":"Intra-cluster correlations from the CLustered OUtcome Dataset bank to inform the design of longitudinal cluster trials","year":2021,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":59,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"Monash University; European Commission; Medical Research Council; National Institute for Health and Care Research","keywords":"Correlation; Autocorrelation; Cluster (spacecraft); Interquartile range; Statistics; Correlation coefficient; Econometrics; Mathematics; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.96021934455992,"gpt":0.6763256042187579,"spread":0.2838937403411621,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.7116694,0.0004077273,0.01108575,0.0001343562,0.0003523635,0.001367111,0.00365075,0.0003005713,0.02613737],"category_scores_gemma":[0.8608375,0.0001329199,0.005225156,0.001766794,0.0002414453,0.0002890848,0.0008340825,0.0005190054,0.006361179],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002503344,"about_ca_system_score_gemma":0.0003022484,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002182698,"about_ca_topic_score_gemma":0.0001023463,"domain_scores_codex":[0.6522998,0.2643641,0.074233,0.001903849,0.006658547,0.0005407376],"domain_scores_gemma":[0.09322587,0.8669024,0.02402336,0.0129719,0.002386169,0.0004902736],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0003081444,0.000117656,0.01841475,0.000009491066,0.001152388,0.000003665011,0.0003189088,0.001429008,0.00001271455,0.0005007095,0.9563037,0.02142886],"study_design_scores_gemma":[0.005443949,0.0002540916,0.06056382,0.0001560618,0.006042115,0.00001513474,0.001440682,0.0147399,0.00005628636,0.02146767,0.8892606,0.0005596303],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01692105,0.00139932,0.8638139,0.09137122,0.007695614,0.009943294,0.007504933,0.00001452547,0.001336142],"genre_scores_gemma":[0.7809439,0.0004328671,0.07686908,0.09097914,0.01012677,0.00100265,0.003114999,0.0001339731,0.03639667],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7869448,"threshold_uncertainty_score":0.9996696,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2037636378","doi":"10.1177/1740774508098413","title":"Optimal sample size determinations from an industry perspective based on the expected value of information","year":2008,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":58,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"SickKids Foundation","funders":"","keywords":"Sample size determination; Bayesian probability; Value of information; Profit (economics); Econometrics; Time horizon; Statistics; Sample (material); Expected utility hypothesis; Expected value; Robustness (evolution); Actuarial science; Computer science; Economics; Mathematics; Microeconomics; Mathematical optimization; Artificial intelligence","retraction":null,"screen_n_in":null,"score":{"opus":0.7750729232342143,"gpt":0.6449265610705023,"spread":0.130146362163712,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02574246,0.0002585478,0.001689665,0.00008227416,0.0002011536,0.00004392432,0.0005849958,0.000840462,0.002042982],"category_scores_gemma":[0.9292617,0.0001696573,0.0006366025,0.0003653348,0.0006992233,0.0002066339,0.00008696684,0.001163304,0.00004161594],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007577852,"about_ca_system_score_gemma":0.0003418649,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008282694,"about_ca_topic_score_gemma":0.000002325461,"domain_scores_codex":[0.9797942,0.01430416,0.004476505,0.0004174942,0.0007331201,0.0002745453],"domain_scores_gemma":[0.2124456,0.7834963,0.00202328,0.001144871,0.0006268281,0.0002631347],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.009618251,0.009384344,0.009697063,0.0001053584,0.001309715,0.00004083765,0.005208985,0.001607138,0.0003457208,0.9075837,0.02055412,0.03454478],"study_design_scores_gemma":[0.004642711,0.001167472,0.01703243,0.0001172209,0.0003228602,0.000001200348,0.001031971,0.01160643,0.001398257,0.9621604,0.0002108686,0.0003081962],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.7301983,0.000007694994,0.2575409,0.003500432,0.001590958,0.002338386,0.002249781,0.000204495,0.002369011],"genre_scores_gemma":[0.3632128,0.000009963705,0.6347357,0.001107804,0.0007960678,0.00008754458,0.00000918331,0.00002337658,0.00001753624],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9035193,"threshold_uncertainty_score":0.9988693,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2806663270","doi":"10.1177/1740774518771233","title":"Incidence and impact of withdrawal of life-sustaining therapies in clinical trials of severe traumatic brain injury: A systematic review","year":2018,"lang":"en","type":"review","venue":"Clinical Trials","topic":"Traumatic Brain Injury and Neurovascular Disturbances","field":"Medicine","cited_by":58,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Centre Hospitalier Universitaire de Sherbrooke; Université de Sherbrooke; University of Manitoba; Université Laval","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Randomized controlled trial; Clinical trial; Glasgow Coma Scale; Traumatic brain injury; CINAHL; Neurointensive care; Incidence (geometry); MEDLINE; Intensive care medicine; Internal medicine; Surgery; Psychological intervention; Psychiatry","retraction":null,"screen_n_in":null,"score":{"opus":0.4191813226836116,"gpt":0.5713822359644174,"spread":0.1522009132808058,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2336397,0.0007946818,0.04629534,0.00040881,0.00003179077,0.00002119127,0.0006512406,0.001044499,0.0001669213],"category_scores_gemma":[0.5626813,0.0004462693,0.00951836,0.0009355103,0.001289871,0.0001233186,0.0001601436,0.001107648,0.000007069173],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005707756,"about_ca_system_score_gemma":0.002350458,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002587401,"about_ca_topic_score_gemma":0.000004865988,"domain_scores_codex":[0.8997077,0.05933687,0.03862358,0.000995947,0.0009163998,0.000419448],"domain_scores_gemma":[0.694657,0.2596495,0.04105098,0.003148692,0.0007716872,0.0007221835],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0008619957,0.0005753463,0.0005266538,0.8481646,0.004489451,0.000008823544,0.0001961472,5.573295e-8,5.435117e-7,0.00005117907,0.001016145,0.1441091],"study_design_scores_gemma":[0.00450832,0.004633373,0.0009430757,0.968869,0.0188298,0.00005915817,0.0001644412,0.000004904788,9.366944e-7,0.0002686363,0.001327264,0.0003911509],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.008403108,0.9793717,0.00001792163,0.0001225529,0.0003810929,0.0114069,0.0002259872,0.00002281562,0.00004797057],"genre_scores_gemma":[0.007393516,0.9911775,0.0003661435,0.0002396385,0.0004399391,0.0002450481,0.00003282596,0.00007064248,0.00003474387],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.3290416,"threshold_uncertainty_score":0.9997989,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3023732552","doi":"10.1177/1740774519896799","title":"Cluster over individual randomization: are study design choices appropriately justified? Review of a random sample of trials","year":2020,"lang":"en","type":"review","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":56,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Western University; Ottawa Hospital; University of Ottawa","funders":"Agence Nationale de la Recherche","keywords":"Randomization; Randomized controlled trial; Psychological intervention; Informed consent; Medicine; Cluster randomised controlled trial; Cluster (spacecraft); Sample size determination; Clinical trial; Family medicine; MEDLINE; External validity; Psychology; Alternative medicine; Nursing; Statistics; Computer science; Social psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.9331299248576629,"gpt":0.7196946620532215,"spread":0.2134352628044414,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad","research_integrity"],"category_scores_codex":[0.5438799,0.0009826353,0.05353865,0.0004034507,0.00008477038,0.00007448644,0.001538562,0.002874183,0.002018318],"category_scores_gemma":[0.961174,0.0005900654,0.0102335,0.00152061,0.0008503907,0.00009541172,0.0008794486,0.005554996,0.00008325414],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007952822,"about_ca_system_score_gemma":0.005497474,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002474295,"about_ca_topic_score_gemma":0.000006682102,"domain_scores_codex":[0.7285202,0.1978324,0.06217275,0.003231339,0.007254008,0.0009892919],"domain_scores_gemma":[0.02719039,0.9398805,0.02666281,0.00280822,0.002500555,0.0009575039],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.02234036,0.00253082,0.00008657564,0.3655842,0.008290268,0.00005214101,0.00006082233,9.419804e-7,2.12695e-7,0.00009016673,0.009587438,0.591376],"study_design_scores_gemma":[0.0918356,0.002716531,0.00001937437,0.3010393,0.04997836,0.000007784557,0.00003722922,0.000005157377,0.000002046665,0.001188764,0.5526966,0.0004731914],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000002834104,0.9547453,0.003459005,0.000826359,0.001569836,0.03813935,0.00104722,0.00005720922,0.0001528886],"genre_scores_gemma":[0.000009522211,0.9885153,0.004267166,0.001153107,0.004081072,0.001199859,0.0003580262,0.0001565564,0.0002593471],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7420481,"threshold_uncertainty_score":0.9996788,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2120773292","doi":"10.1177/1740774515571140","title":"Ethical and regulatory issues of pragmatic cluster randomized trials in contemporary health systems","year":2015,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":52,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"National Center for Complementary and Integrative Health; National Institute of Mental Health; National Institutes of Health; Fresenius Medical Care North America","keywords":"Collaboratory; Randomized controlled trial; Cluster randomised controlled trial; Health care; Research ethics; Informed consent; Medicine; Health informatics; Clinical trial; Public relations; Nursing; Psychological intervention; Medical education; Alternative medicine; Political science; Public health; Computer science; Psychiatry","retraction":null,"screen_n_in":null,"score":{"opus":0.910000639947712,"gpt":0.7324794847624002,"spread":0.1775211551853118,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.8191075,0.0002522969,0.01387545,0.0002563249,0.00003354966,0.00005017631,0.0002626204,0.002163197,0.00004186539],"category_scores_gemma":[0.9494281,0.0001602964,0.001115564,0.0002685978,0.001913902,0.00007079836,0.0002381472,0.004411432,0.00002967244],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008935865,"about_ca_system_score_gemma":0.00403208,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003477274,"about_ca_topic_score_gemma":0.00002778717,"domain_scores_codex":[0.592895,0.3520775,0.04691455,0.001939835,0.005079478,0.001093686],"domain_scores_gemma":[0.04616174,0.9443843,0.00512288,0.0014137,0.001324675,0.001592728],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.7162291,0.002325903,0.01308588,0.008350538,0.001833971,0.0001255863,0.001916997,0.00002011639,0.00008158229,0.08838992,0.1609094,0.006730981],"study_design_scores_gemma":[0.8381326,0.003381903,0.001616791,0.007865072,0.0003892299,0.00002069962,0.001109082,0.001646376,0.00005758907,0.1363839,0.009067331,0.0003294776],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.3577045,0.08316943,0.001746929,0.5071878,0.00687667,0.03524046,0.0001147693,0.000224938,0.007734564],"genre_scores_gemma":[0.9793553,0.006309589,0.0028469,0.007680314,0.00177184,0.0002176515,0.00002410986,0.00005095508,0.001743309],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6216509,"threshold_uncertainty_score":0.9991322,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2035324238","doi":"10.1177/1740774512469312","title":"Central statistical monitoring: Detecting fraud in clinical trials","year":2013,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":52,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Hamilton Health Sciences; McMaster University; Population Health Research Institute","funders":"Canadian Institutes of Health Research; British Heart Foundation; Medical Research Council; Australian Government","keywords":"Medicine; Clinical trial; Multicenter study; Risk assessment; Statistics; Cutoff; Missing data; Statistical model; Computer science; Data mining; Randomized controlled trial; Internal medicine; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.9255815327318341,"gpt":0.7339168505663992,"spread":0.1916646821654349,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity","insufficient_payload"],"category_scores_codex":[0.4196652,0.0008520088,0.01183752,0.0002622334,0.0001802432,0.0003901251,0.00123616,0.002135351,0.005753752],"category_scores_gemma":[0.979388,0.0006491916,0.002925391,0.0006956147,0.001014224,0.0002961688,0.0005068926,0.003884286,0.001028489],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001684415,"about_ca_system_score_gemma":0.0004643434,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009051193,"about_ca_topic_score_gemma":0.00000960968,"domain_scores_codex":[0.7544338,0.1889246,0.04899273,0.003008852,0.001923447,0.002716647],"domain_scores_gemma":[0.03129462,0.9578483,0.00623011,0.002018574,0.0005248141,0.002083553],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001646044,0.002795413,0.1125241,0.0002036122,0.0007543957,0.0001359302,0.00008145305,0.000005129828,0.0001554587,0.04077259,0.02216443,0.8187615],"study_design_scores_gemma":[0.01082489,0.0008913197,0.09012695,0.000410163,0.0006047466,0.000005939297,0.0001747639,0.0007278666,0.0003234227,0.8933734,0.001732253,0.0008043075],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.414927,0.0004269688,0.514247,0.003529974,0.04635067,0.01427694,0.0005590525,0.00108296,0.004599383],"genre_scores_gemma":[0.2621369,0.0004711722,0.7217322,0.0005482969,0.01412771,0.0004919837,0.000006219561,0.0001774604,0.0003080442],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8526008,"threshold_uncertainty_score":0.9997493,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2172067102","doi":"10.1191/1740774505cn111oa","title":"Minority recruitment to the Selenium and Vitamin E Cancer Prevention Trial (SELECT)","year":2005,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":50,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"National Cancer Institute","keywords":"Cancer prevention; Medicine; Cancer; Selenium; Gerontology; Oncology; Environmental health; Internal medicine; Chemistry","retraction":null,"screen_n_in":null,"score":{"opus":0.885305631011485,"gpt":0.7263981773616662,"spread":0.1589074536498187,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.09113205,0.0001700417,0.001154349,0.00006007443,0.0001376443,0.00006031145,0.0002679901,0.0006261418,0.000648127],"category_scores_gemma":[0.228516,0.0001017663,0.0004776818,0.0002665023,0.0002540031,0.00005301316,0.0002158078,0.001944873,0.0002299859],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001260625,"about_ca_system_score_gemma":0.0008212958,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000491822,"about_ca_topic_score_gemma":0.0007340591,"domain_scores_codex":[0.9908727,0.004098954,0.002823971,0.000747273,0.001020276,0.0004368333],"domain_scores_gemma":[0.9293147,0.06845619,0.0004056249,0.0007820977,0.0004344019,0.0006069441],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.08613257,0.002201469,0.006596226,0.0001568911,0.0005681461,0.000008131151,0.000201616,0.000005754677,0.0007100475,0.003097257,0.08133052,0.8189914],"study_design_scores_gemma":[0.140646,0.01137583,0.03317353,0.0006036282,0.001079777,0.000008517683,0.00009042922,0.0002085028,0.002784122,0.01199354,0.79761,0.0004261622],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8055698,0.001208316,0.0002995116,0.1770545,0.002410842,0.01064089,0.00002984334,0.00007837025,0.002707989],"genre_scores_gemma":[0.9006016,0.01189997,0.005012723,0.01184393,0.02669104,0.001960038,0.00001532394,0.00004713448,0.04192824],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8185652,"threshold_uncertainty_score":0.9358708,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2888017152","doi":"10.1177/1740774518795637","title":"Moving forward toward standardizing analysis of quality of life data in randomized cancer clinical trials","year":2018,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Cancer survivorship and care","field":"Medicine","cited_by":49,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Health Canada; University of British Columbia","funders":"National Cancer Institute; Health Canada; Boehringer Ingelheim","keywords":"Randomized controlled trial; Clinical trial; Medicine; Quality of life (healthcare); Medical physics; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.6922122348330196,"gpt":0.6432187469384346,"spread":0.04899348789458502,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4198647,0.0002693599,0.01875273,0.0004260046,0.00003166174,0.00002269754,0.0005902166,0.0005571684,0.001298338],"category_scores_gemma":[0.5915821,0.0001881441,0.004770357,0.001310698,0.001009342,0.0001214865,0.0003692303,0.0005389487,0.000004506835],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006490617,"about_ca_system_score_gemma":0.001525169,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004597773,"about_ca_topic_score_gemma":0.00506805,"domain_scores_codex":[0.9332084,0.04246411,0.0216742,0.001111068,0.001130795,0.0004114131],"domain_scores_gemma":[0.6819613,0.2948889,0.01482452,0.005102401,0.002094795,0.001128146],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.5110404,0.0007577322,0.3682884,0.0004623605,0.01905989,0.00000505488,0.0007637949,0.00001230403,0.0002417049,0.0003753379,0.002328055,0.09666496],"study_design_scores_gemma":[0.7602321,0.001402819,0.1697047,0.001569454,0.04951548,4.344282e-7,0.00197001,0.004158264,0.0009223953,0.001200126,0.008534141,0.0007900873],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9722881,0.004637568,0.01054804,0.002568217,0.004382923,0.002536432,0.002209447,0.00004582869,0.0007834512],"genre_scores_gemma":[0.9909111,0.004249638,0.0006400354,0.001031111,0.002878796,0.00005106585,0.0001475415,0.00002567167,0.00006499344],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2524247,"threshold_uncertainty_score":0.9996146,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3125944419","doi":"10.1177/1740774520984870","title":"Blinding, sham, and treatment effects in randomized controlled trials for back pain in 2000–2019: A review and meta-analytic approach","year":2021,"lang":"en","type":"review","venue":"Clinical Trials","topic":"Musculoskeletal pain and rehabilitation","field":"Medicine","cited_by":49,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Work & Health; Public Health Ontario; University of Toronto","funders":"National Center for Advancing Translational Sciences; National Institute of Arthritis and Musculoskeletal and Skin Diseases; National Institutes of Health","keywords":"Blinding; Medicine; Randomized controlled trial; Meta-analysis; Physical therapy; N of 1 trial; Alternative medicine; Placebo; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.482903966861432,"gpt":0.5632263736170604,"spread":0.0803224067556284,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.4112827,0.0009133906,0.1164365,0.0005220668,0.00003153194,0.0000633314,0.0001078963,0.001009296,0.0002655707],"category_scores_gemma":[0.6341327,0.0004074736,0.02044232,0.0004976519,0.0002556811,0.00004223731,0.00004479242,0.0005316165,0.000009262041],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001782238,"about_ca_system_score_gemma":0.0008083264,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001796057,"about_ca_topic_score_gemma":0.00000781843,"domain_scores_codex":[0.8298475,0.1492355,0.01853014,0.001560943,0.0003662469,0.0004596779],"domain_scores_gemma":[0.5408452,0.4530439,0.004830534,0.0007460116,0.000146415,0.0003879651],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.02380051,0.001186631,0.000004881937,0.2321456,0.03844513,0.00001067802,0.00003111135,1.807015e-7,1.412488e-7,0.0001678367,0.001566753,0.7026405],"study_design_scores_gemma":[0.5173133,0.001210983,0.000003037897,0.03728289,0.3244471,0.000004501405,0.000009026306,0.00007797166,3.969854e-8,0.0002022503,0.1190913,0.000357607],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000007946699,0.9179586,0.000205,0.0009045642,0.0002038205,0.08050577,0.00004904627,0.00001346412,0.0001517814],"genre_scores_gemma":[0.000004498308,0.9793416,0.00249129,0.0005970696,0.0004522613,0.01444788,0.0003980748,0.00005979638,0.002207542],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7022829,"threshold_uncertainty_score":0.9998377,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2104499805","doi":"10.1177/1740774514561242","title":"The impact of registration of clinical trials units: The UK experience","year":2014,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":49,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Clinical trial; Medicine; Staffing; Accreditation; Medical education; Family medicine; Medical physics; Nursing; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.9555622389471763,"gpt":0.7996324837973618,"spread":0.1559297551498144,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.615947,0.0002234948,0.003893511,0.00005344936,0.0001804072,0.00005465053,0.001089845,0.001112702,0.0003438196],"category_scores_gemma":[0.9731249,0.00009198313,0.002671334,0.0005962087,0.004088107,0.00005587944,0.000237851,0.003043666,0.00003540627],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002901417,"about_ca_system_score_gemma":0.002340224,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001959844,"about_ca_topic_score_gemma":0.00006707804,"domain_scores_codex":[0.8424705,0.1069226,0.04325852,0.001607865,0.004630212,0.001110245],"domain_scores_gemma":[0.0184672,0.9675477,0.007371934,0.003179921,0.002851332,0.000581914],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.04119755,0.002892223,0.3326049,0.0003309454,0.003049612,0.00001313019,0.0007890459,0.0000469857,0.003039424,0.1265711,0.06428297,0.4251821],"study_design_scores_gemma":[0.04085925,0.02957976,0.5922912,0.002328291,0.001799673,0.00001948361,0.0022997,0.002867563,0.005922911,0.2823925,0.03892738,0.0007121987],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9722062,0.0005956545,0.001244866,0.01099931,0.002342144,0.002807007,0.00005269399,0.00003820278,0.009713865],"genre_scores_gemma":[0.9897674,0.004827465,0.0003278532,0.0005001417,0.003033307,0.00004989864,0.000009472587,0.00002390955,0.00146056],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8606251,"threshold_uncertainty_score":0.9992564,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2013383304","doi":"10.1177/1740774508093981","title":"Determining optimal sample sizes for multi-stage randomized clinical trials using value of information methods","year":2008,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":48,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Randomized controlled trial; Statistics; Sample size determination; Value (mathematics); Clinical trial; Stage (stratigraphy); Medicine; Computer science; Mathematics; Internal medicine; Biology","retraction":null,"screen_n_in":null,"score":{"opus":0.9543317968023223,"gpt":0.7614272996126612,"spread":0.1929044971896611,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6663435,0.0007179563,0.01922518,0.0003016691,0.0003127356,0.0001030367,0.0009265519,0.001642092,0.0005421399],"category_scores_gemma":[0.992816,0.0005244972,0.008058362,0.0004395029,0.002098974,0.0004660015,0.0003750786,0.001097824,0.00001941008],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006660228,"about_ca_system_score_gemma":0.0007832056,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002647109,"about_ca_topic_score_gemma":8.745818e-7,"domain_scores_codex":[0.5738138,0.3464895,0.07451887,0.001919357,0.00166019,0.001598306],"domain_scores_gemma":[0.02928778,0.9512555,0.01630109,0.001408758,0.001050398,0.0006964463],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.2449022,0.003278669,0.001553058,0.001387574,0.005701969,0.00001882169,0.0007694986,0.0005137495,0.0006189508,0.1071533,0.002511107,0.6315911],"study_design_scores_gemma":[0.4046019,0.001227676,0.0002121643,0.000460822,0.004614114,0.00001193448,0.0001959032,0.2426094,0.0029894,0.3396125,0.002468391,0.0009957651],"study_design_candidate":"randomized_trial","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.023837,0.00009340208,0.9604595,0.0001165999,0.006043451,0.008032414,0.001120931,0.0001896662,0.0001070045],"genre_scores_gemma":[0.002708448,0.0005019255,0.9931689,0.0004759755,0.002555521,0.0003817311,0.00002443684,0.0001012871,0.00008176158],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6305953,"threshold_uncertainty_score":0.9997206,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2315664626","doi":"10.1177/1740774516637871","title":"The design and implementation of a randomized controlled trial of a risk reduction and human immunodeficiency virus prevention videogame intervention in minority adolescents: <i>PlayForward: Elm City Stories</i>","year":2016,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Educational Games and Gamification","field":"Psychology","cited_by":46,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"Eunice Kennedy Shriver National Institute of Child Health and Human Development; National Center for Advancing Translational Sciences","keywords":"Randomized controlled trial; Psychological intervention; Psychosocial; Intervention (counseling); Behavior change; Applied psychology; Psychology; Medicine; Medical education; Social psychology; Nursing; Psychiatry","retraction":null,"screen_n_in":null,"score":{"opus":0.1219716352158673,"gpt":0.4972523359522329,"spread":0.3752807007363657,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05293486,0.0001279746,0.001420684,0.0001045,0.00008434585,0.00002176756,0.0001047327,0.0001568868,0.00007028831],"category_scores_gemma":[0.01081846,0.00007315061,0.0005131021,0.0001166537,0.0003889003,0.0001149085,0.00002324106,0.0001318294,0.000001299374],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003620015,"about_ca_system_score_gemma":0.00004930292,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006548254,"about_ca_topic_score_gemma":0.00008390906,"domain_scores_codex":[0.9825366,0.01294939,0.003833445,0.0003305674,0.0002078551,0.0001421849],"domain_scores_gemma":[0.9864984,0.00965819,0.003344544,0.0002734422,0.0001847888,0.00004062103],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.8667812,0.001359617,0.006633703,0.00002532725,0.0003306456,7.86715e-8,0.000973777,5.449126e-7,0.002704018,0.006601515,0.0000983407,0.1144912],"study_design_scores_gemma":[0.9133432,0.001070627,0.06916278,0.0001511172,0.0005207172,9.070907e-7,0.001158274,0.000006320157,0.0003899287,0.01403183,0.0000727439,0.00009156209],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9840077,0.001125397,0.007419154,0.0002167535,0.00144254,0.005716541,0.000022426,0.00000892244,0.00004059805],"genre_scores_gemma":[0.9974262,0.001461819,0.0001330223,0.000004916433,0.000214845,0.000603841,0.000007789879,0.000006408992,0.0001411505],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1143997,"threshold_uncertainty_score":0.9975138,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2084652988","doi":"10.1177/1740774507079439","title":"Monitoring the randomized trials of the Women's Health Initiative: the experience of the Data and Safety Monitoring Board","year":2007,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Menopause: Health Impacts and Treatments","field":"Medicine","cited_by":46,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Safety monitoring; Medicine; Clinical trial; Randomized controlled trial; Data monitoring committee; Adverse effect; Women's Health Initiative; Postmenopausal women; Bioinformatics; Pharmacology; Surgery; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.5609611983456146,"gpt":0.5880129174506635,"spread":0.02705171910504889,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1506893,0.0002213776,0.003170908,0.00003144685,0.0004441686,0.00002685473,0.0007417315,0.0001375275,0.00003304013],"category_scores_gemma":[0.1561189,0.00006910295,0.0002175939,0.0003063682,0.0009221812,0.00008707227,0.0004463573,0.0005136769,0.000001271583],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008988882,"about_ca_system_score_gemma":0.0005777652,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001710562,"about_ca_topic_score_gemma":0.000003577836,"domain_scores_codex":[0.9775089,0.0146475,0.006094511,0.0004236332,0.0007905082,0.0005349708],"domain_scores_gemma":[0.7922809,0.1930853,0.00899348,0.004609836,0.0003374192,0.0006931131],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.3968183,0.001664478,0.2651139,0.0008403034,0.004488173,0.00001207213,0.0371805,0.000007916104,0.002924195,0.0005538128,0.001360734,0.2890357],"study_design_scores_gemma":[0.5043828,0.002654977,0.451229,0.003631224,0.002129063,0.00001845392,0.01337218,0.00004639718,0.01087247,0.001569737,0.009783576,0.0003101376],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9753916,0.004922003,0.00008026353,0.01009523,0.003810862,0.005394346,0.00009855004,0.00001426274,0.0001928578],"genre_scores_gemma":[0.9914562,0.005750306,0.0001796597,0.0008363936,0.001608861,0.00005441407,0.000001820603,0.00001830968,0.00009408624],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2887256,"threshold_uncertainty_score":0.8745441,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2569995479","doi":"10.1177/1740774516685298","title":"PRagmatic trial Of Video Education in Nursing homes: The design and rationale for a pragmatic cluster randomized trial in the nursing home setting","year":2017,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Geriatric Care and Nursing Homes","field":"Health Professions","cited_by":44,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"National Institute on Aging; National Institutes of Health; Schulich School of Medicine and Dentistry, Western University","keywords":"Nursing; Randomized controlled trial; Nursing homes; Cluster (spacecraft); Medicine; Cluster randomised controlled trial; Psychology; Computer science; Intervention (counseling)","retraction":null,"screen_n_in":null,"score":{"opus":0.2175819249906993,"gpt":0.5743555864927983,"spread":0.356773661502099,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1291717,0.0002033913,0.002041658,0.0001742916,0.0008592242,0.0001219786,0.0004433474,0.0003519598,0.00002240938],"category_scores_gemma":[0.1692203,0.0001110261,0.0004608004,0.000150016,0.0004996264,0.0002138201,0.00003299315,0.000616517,0.000004295162],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001120955,"about_ca_system_score_gemma":0.001633938,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005298311,"about_ca_topic_score_gemma":0.00002415945,"domain_scores_codex":[0.960648,0.03353874,0.004656679,0.0003850606,0.0003653414,0.000406146],"domain_scores_gemma":[0.8027282,0.1928262,0.003456823,0.0007381691,0.0001808417,0.00006975445],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.8935121,0.0008070781,0.00005961144,0.0001843216,0.00004048755,2.926333e-7,0.01383472,0.00001135079,0.00000722098,0.001304235,0.00930344,0.08093517],"study_design_scores_gemma":[0.8807389,0.0004813795,0.0008258795,0.002770154,0.0003683747,0.000001165859,0.009452696,0.00127529,0.000001830945,0.1037893,0.0001648482,0.0001301538],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8520445,0.00111847,0.01236893,0.05196853,0.01388778,0.06744064,0.00002009823,0.00002927021,0.001121824],"genre_scores_gemma":[0.9785823,0.0003090522,0.008802603,0.0006900256,0.005821837,0.005414271,0.00004589782,0.00004525804,0.0002887287],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1592875,"threshold_uncertainty_score":0.896701,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1973629098","doi":"10.1177/1740774509105222","title":"Scientific and political challenges in North America's first randomized controlled trial of heroin-assisted treatment for severe heroin addiction: Rationale and design of the NAOMI study","year":2009,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Opioid Use Disorder Treatment","field":"Medicine","cited_by":43,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"HIV Legal Network; Université de Montréal; Vancouver Coastal Health; Centre for Advancing Health Outcomes; University of British Columbia; Providence Health Care","funders":"Canadian Institutes of Health Research","keywords":"Heroin; Randomized controlled trial; Medicine; Methadone maintenance; Hydromorphone; Addiction; Methadone; Psychiatry; Clinical trial; Opioid; Drug; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.2408842585114785,"gpt":0.4340127078411926,"spread":0.1931284493297141,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006279035,0.0002061629,0.003608264,0.00009987639,0.00008624299,0.00002069449,0.00006715909,0.0001073188,0.00001644616],"category_scores_gemma":[0.01488234,0.0001098542,0.0006268941,0.0001609476,0.0004221133,0.00003480243,0.00001991877,0.00008669785,4.982153e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005831521,"about_ca_system_score_gemma":0.0003012537,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001660839,"about_ca_topic_score_gemma":0.0001033763,"domain_scores_codex":[0.9933569,0.00331032,0.002401013,0.0004381154,0.000297415,0.0001962215],"domain_scores_gemma":[0.9786935,0.01993486,0.000704872,0.0003676756,0.0001664355,0.000132676],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.9697742,0.01207749,0.002124491,0.00002827569,0.0009933091,0.000002814592,0.000680201,0.00001429255,0.00001339852,0.0001740333,0.00006886184,0.01404864],"study_design_scores_gemma":[0.9488564,0.009388106,0.03799244,0.00007769663,0.001773689,0.000002811718,0.001076451,0.0003020004,0.0000158344,0.0003239949,0.0001081166,0.00008248193],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9629227,0.00271583,0.00004809651,0.005759356,0.0003615737,0.0280405,0.00007872196,0.00001333099,0.00005992586],"genre_scores_gemma":[0.9967386,0.001262982,0.0005944448,0.00006404197,0.0001643082,0.0009996369,0.00002517179,0.00001057598,0.0001402202],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03586794,"threshold_uncertainty_score":0.9934157,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2604399020","doi":"10.1177/1740774517703057","title":"Inadequacy of ethical conduct and reporting of stepped wedge cluster randomized trials: Results from a systematic review","year":2017,"lang":"en","type":"review","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":42,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Western University; McGill University; Ottawa Hospital; University of Ottawa","funders":"","keywords":"Informed consent; Consolidated Standards of Reporting Trials; Research ethics; Randomized controlled trial; Research design; Family medicine; Cluster randomised controlled trial; Clinical trial; Medicine; Psychological intervention; Cluster (spacecraft); Alternative medicine; Psychology; Medical education; Nursing; Psychiatry; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.9771237733653158,"gpt":0.7924728271372427,"spread":0.1846509462280731,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","sts","research_integrity"],"consensus_categories":["metaresearch","metaepi_broad","research_integrity"],"category_scores_codex":[0.8641493,0.0009399524,0.1336404,0.0003189296,0.0001128587,0.00008990759,0.00134851,0.007650762,0.00016597],"category_scores_gemma":[0.9984752,0.0004996361,0.01603397,0.000338853,0.003507755,0.0000741269,0.001037287,0.01298793,0.00005549465],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008487791,"about_ca_system_score_gemma":0.005121417,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001187344,"about_ca_topic_score_gemma":0.00001443388,"domain_scores_codex":[0.2891681,0.4044367,0.293984,0.004543461,0.006546573,0.001321126],"domain_scores_gemma":[0.002327826,0.8503136,0.1389812,0.005088273,0.002505591,0.0007834657],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.04476264,0.0002531036,0.000001166268,0.8812289,0.005449169,0.00007164088,0.00003186931,5.214619e-9,5.850472e-7,0.0004235065,0.00140709,0.06637028],"study_design_scores_gemma":[0.1414936,0.0004674543,5.198416e-7,0.7768824,0.05668022,0.0000310114,0.0000126163,0.000005260109,0.000002187236,0.004320142,0.01979963,0.000304984],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00001211,0.9548008,0.00005005195,0.004176328,0.00142162,0.03802249,0.0005754394,0.00003671007,0.0009044055],"genre_scores_gemma":[0.00004707371,0.9934258,0.001783099,0.0007866017,0.001543545,0.0008588512,0.000266618,0.000115074,0.001173349],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4458769,"threshold_uncertainty_score":0.9997455,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2510936828","doi":"10.1177/1740774516665597","title":"Adult patient perspectives on clinical trial result reporting: A survey of cancer patients","year":2016,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":41,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa; Dalhousie University; University of Calgary; University of Alberta","funders":"","keywords":"Medicine; Clinical trial; Family medicine; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9095515158207234,"gpt":0.7475305313935051,"spread":0.1620209844272182,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.2076227,0.0003309454,0.003934618,0.0001567328,0.00008368291,0.00002707807,0.0004655737,0.001553266,0.0005989205],"category_scores_gemma":[0.972362,0.0001811662,0.001890712,0.000427709,0.001570682,0.0000672383,0.0003152266,0.00262893,0.0001487304],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001636475,"about_ca_system_score_gemma":0.002181622,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004505869,"about_ca_topic_score_gemma":0.0003054234,"domain_scores_codex":[0.9505852,0.01638811,0.02725342,0.001990036,0.003045409,0.0007377863],"domain_scores_gemma":[0.1046645,0.8414851,0.02837422,0.004679127,0.01853295,0.002264131],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.2583284,0.00767411,0.6279719,0.00005444179,0.0006141852,0.00002494403,0.000112694,1.648804e-7,0.00003999984,0.001007066,0.007023403,0.09714859],"study_design_scores_gemma":[0.1671057,0.03528137,0.7902656,0.00172706,0.0002689743,5.741741e-7,0.0001209793,0.000004061461,0.0004740682,0.00309306,0.001357758,0.0003008393],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9848061,0.0001511234,0.00002208125,0.005608879,0.004057869,0.002855416,0.0003239442,0.00006305472,0.002111599],"genre_scores_gemma":[0.9890528,0.005605374,0.0003832027,0.001072714,0.002266005,0.0001290977,0.00002305634,0.00006092262,0.001406847],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8459207,"threshold_uncertainty_score":0.9997429,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","scholarly_communication"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2165814638","doi":"10.1177/1740774515597702","title":"Clinician trialist rounds: 28. When RCT participants are lost to follow-up. Part 1: Why even a few can matter","year":2015,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":40,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Hamilton Health Sciences; McMaster University; Population Health Research Institute","funders":"","keywords":"Randomized controlled trial; Medicine; Psychology; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9161797198778433,"gpt":0.6095619279054681,"spread":0.3066177919723752,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.2263795,0.0004858405,0.005832803,0.0003431809,0.0002086523,0.0004100221,0.0008838439,0.0006579417,0.003024256],"category_scores_gemma":[0.219606,0.0005314592,0.001079476,0.0003353062,0.0001752825,0.0004589021,0.0002257394,0.000522848,0.04052458],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007283707,"about_ca_system_score_gemma":0.0007725149,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001717589,"about_ca_topic_score_gemma":0.002529702,"domain_scores_codex":[0.9667441,0.008213877,0.02192875,0.001602293,0.0003051012,0.001205831],"domain_scores_gemma":[0.9708323,0.01634881,0.008679209,0.001794536,0.0002783781,0.002066712],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0005064354,0.0002584398,0.1617115,0.0000751243,0.0002688876,0.000004343785,0.001019423,0.0000248596,1.894231e-7,0.001914526,0.8338835,0.000332814],"study_design_scores_gemma":[0.01068626,0.0005025688,0.01962318,0.0002987593,0.0001080519,0.000002193753,0.001227801,0.0001799463,0.00000241583,0.03285325,0.9337199,0.000795713],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.3957215,0.00200813,0.00188847,0.547491,0.03285543,0.006555228,0.003823626,0.0002200294,0.009436521],"genre_scores_gemma":[0.5177823,0.0002751061,0.003213092,0.4222646,0.02427052,0.001081347,0.0002692171,0.0003121731,0.0305316],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.1420883,"threshold_uncertainty_score":0.9997137,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4205896703","doi":"10.1177/17407745211069153","title":"The representation of Indigenous peoples in chronic disease clinical trials in Australia, Canada, New Zealand, and the United States","year":2022,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Indigenous Health, Education, and Rights","field":"Social Sciences","cited_by":38,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Calgary","funders":"Canadian Institutes of Health Research","keywords":"Indigenous; Clinical trial; Representation (politics); Medicine; Geography; Political science; Family medicine; Ethnology; History; Law; Pathology; Politics","retraction":null,"screen_n_in":null,"score":{"opus":0.2608600660078836,"gpt":0.5287024864905064,"spread":0.2678424204826229,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1440807,0.0001167076,0.001219387,0.0001050995,0.0021313,0.00006615104,0.0004466308,0.0001201764,0.0001527041],"category_scores_gemma":[0.02114482,0.00006475366,0.0002677261,0.0007336851,0.0007386961,0.00006051047,0.00001317543,0.0005344943,0.000001969979],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003992523,"about_ca_system_score_gemma":0.05698408,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9882676,"about_ca_topic_score_gemma":0.9990036,"domain_scores_codex":[0.9237807,0.06945062,0.005176923,0.000399825,0.0006639803,0.0005279288],"domain_scores_gemma":[0.8127964,0.183335,0.002634051,0.0005631081,0.0001494036,0.0005220312],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.009373667,0.000907258,0.4741779,0.00007588818,0.0002674666,0.00002915511,0.4468136,0.00130682,4.413333e-7,0.03814507,0.01754735,0.01135532],"study_design_scores_gemma":[0.01118208,0.0003021152,0.2464882,0.00005675155,0.0001721734,5.943643e-7,0.03150927,0.00008656681,0.000002952387,0.02946126,0.6805034,0.0002347349],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9865848,0.0009131785,0.000001802075,0.0078253,0.001905135,0.002599006,0.00007217864,0.000008491034,0.00009015563],"genre_scores_gemma":[0.9241856,0.06315655,0.00001309485,0.0002658088,0.002022092,0.00005452085,0.0001135346,0.00001332639,0.0101755],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.662956,"threshold_uncertainty_score":0.9991678,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2161096693","doi":"10.1177/1740774513491337","title":"Data collection in cancer clinical trials: Too much of a good thing?","year":2013,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":37,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Ontario Clinical Oncology Group; McMaster University","funders":"","keywords":"Clinical trial; Medicine; Data collection; Cancer; Family medicine; Medical physics; Internal medicine; Statistics","retraction":null,"screen_n_in":null,"score":{"opus":0.9731630913816647,"gpt":0.8030679916317143,"spread":0.1700950997499504,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.4931279,0.0002944041,0.005864209,0.0002595996,0.00007766978,0.00007823975,0.001302927,0.00232808,0.003464402],"category_scores_gemma":[0.9120395,0.0002098465,0.001310062,0.0009357585,0.001017674,0.0002736275,0.0009627863,0.00589758,0.0003511665],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001207051,"about_ca_system_score_gemma":0.003616271,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001481283,"about_ca_topic_score_gemma":0.0008215318,"domain_scores_codex":[0.930127,0.04020818,0.02414806,0.002188031,0.002423454,0.0009052762],"domain_scores_gemma":[0.05285433,0.9322873,0.005325016,0.005630456,0.002443791,0.001459071],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.009584862,0.007508953,0.4572005,0.0007476837,0.001626416,0.00005537766,0.0001626004,0.000002980757,0.0007515458,0.003821545,0.2485931,0.2699445],"study_design_scores_gemma":[0.1125472,0.0130218,0.5297891,0.006364306,0.003073115,0.00002124048,0.001081924,0.01467531,0.001244542,0.2127522,0.1039529,0.001476311],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8652301,0.003573856,0.001018714,0.08808798,0.01268439,0.01340814,0.0004559946,0.0002297946,0.01531102],"genre_scores_gemma":[0.908329,0.04380123,0.01335908,0.008451616,0.01034149,0.0006768845,0.0002009838,0.0001511068,0.01468863],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8920792,"threshold_uncertainty_score":0.9989671,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2171984706","doi":"10.1177/1740774510373497","title":"Designing and implementing sample and data collection for an international genetics study: the Type 1 Diabetes Genetics Consortium (T1DGC)","year":2010,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Diabetes and associated disorders","field":"Biochemistry, Genetics and Molecular Biology","cited_by":35,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"National Institute of Child Health and Human Development; National Human Genome Research Institute; National Institute of Diabetes and Digestive and Kidney Diseases; Norwegian Institute of Public Health; University of Texas Health Science Center at San Antonio; Medical School, University of Michigan; National Institutes of Health; Universität Ulm; State University of New York Upstate Medical University; University of Washington; Università degli Studi di Sassari; Universitetet i Oslo; Karolinska Institutet; University of Miami; University of South Carolina; Queen's University; Murdoch University; Universiteit Leiden; Massachusetts General Hospital; Queen's University Belfast; McGill University Health Centre; National Institute of Allergy and Infectious Diseases; Bunning Food Allergy Institute, Ann and Robert H. Lurie Children's Hospital of Chicago; Vrije Universiteit Brussel; University of Alaska Anchorage; University of Bristol; Hospital for Sick Children; Broad Institute; Helsingin Yliopisto; Calgary Laboratory Services; Semmelweis Egyetem; University of South Florida; University of Pittsburgh; Leids Universitair Medisch Centrum; Vanderbilt University; Seattle Children's Research Institute; Joslin Diabetes Center; Children's National Hospital; Juvenile Diabetes Research Foundation International; Children's Mercy Hospital; Economic and Social Research Council; Kaiser Permanente; State University of New York; University of Minnesota; University of Toronto; University of Alberta; Cincinnati Children's Hospital Medical Center; King's College London; Children's Hospital of Philadelphia; Ohio State University; Wake Forest University; University of Rochester; Wellcome Trust; University of Southern California; BC Children's Hospital; Medical Center, University of Rochester; Children's Hospital Los Angeles; McGill University; Cancer Research Institute","keywords":"Genotyping; Type 2 diabetes; Medicine; Library science; Family medicine; Diabetes mellitus; Genetics; Biology; Computer science; Genotype; Gene","retraction":null,"screen_n_in":null,"score":{"opus":0.2444899452040281,"gpt":0.4859426851334839,"spread":0.2414527399294557,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006805646,0.00009792129,0.0002335809,0.00001561796,0.0001850799,0.0001280118,0.0002705614,0.0001253611,0.00001037322],"category_scores_gemma":[0.009787794,0.00007504097,0.00004494061,0.00004182283,0.0001029229,0.000007054221,0.000300914,0.0001069741,3.056204e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000001645526,"about_ca_system_score_gemma":0.00005431246,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00000854407,"about_ca_topic_score_gemma":0.0001479035,"domain_scores_codex":[0.998406,0.0003769214,0.0005577619,0.0003941206,0.00007822102,0.0001869818],"domain_scores_gemma":[0.9979359,0.001258577,0.0002497841,0.0003696282,0.0001163879,0.00006976009],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001067075,0.0002291139,0.8453847,0.000005960431,0.0004770556,6.84964e-8,0.00005209839,0.00000321297,0.08701676,0.00002344841,0.003560445,0.0631405],"study_design_scores_gemma":[0.01955807,0.008883935,0.3673061,0.00002588418,0.001791251,0.000002508297,0.003635924,0.01907086,0.03606078,0.003609068,0.538743,0.001312572],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9971468,0.0003763597,0.000556937,0.0001853208,0.0008985572,0.0006497183,0.0001476586,0.000007150999,0.00003148123],"genre_scores_gemma":[0.9922693,0.0007824091,0.005319605,0.0002686368,0.0008854854,0.00002737322,0.0003999026,0.0000174001,0.00002982822],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5351825,"threshold_uncertainty_score":0.9985532,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2304549429","doi":"10.1177/1740774516638345","title":"Did death certificates and a death review process agree on lung cancer cause of death in the National Lung Screening Trial?","year":2016,"lang":"en","type":"review","venue":"Clinical Trials","topic":"Lung Cancer Diagnosis and Treatment","field":"Medicine","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Public Health Ontario; University of Toronto","funders":"National Cancer Institute; National Institutes of Health","keywords":"Medicine; Death certificate; Cause of death; Lung cancer; Lung cancer screening; Cancer; Randomized controlled trial; Intensive care medicine; Internal medicine; Disease","retraction":null,"screen_n_in":null,"score":{"opus":0.6126625609003139,"gpt":0.6191813903230365,"spread":0.006518829422722638,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.01612442,0.0005985195,0.006700702,0.0001847917,0.00007639312,0.00004011444,0.0003625512,0.0004648052,0.0001847681],"category_scores_gemma":[0.01725526,0.0002679055,0.00146689,0.0003402002,0.00009042245,0.00006582257,0.00006838318,0.0006127061,0.000009807181],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003587547,"about_ca_system_score_gemma":0.001209739,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005123748,"about_ca_topic_score_gemma":0.00001146591,"domain_scores_codex":[0.9901773,0.0030312,0.004520827,0.0009870259,0.0009138184,0.000369862],"domain_scores_gemma":[0.9768078,0.01960278,0.00243888,0.0006232157,0.0003077641,0.0002195662],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.004081671,0.001596259,0.01277171,0.08351268,0.003819239,0.00007276718,0.00004516547,3.662663e-7,4.755362e-8,0.0007011911,0.0115864,0.8818125],"study_design_scores_gemma":[0.03273687,0.001072728,0.002445928,0.5369391,0.01814567,0.00002891241,0.000003938786,0.000007439186,0.000002214343,0.0002144513,0.4079111,0.0004916186],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0001817645,0.9859774,0.000006624694,0.002989859,0.0003836129,0.009655724,0.0003491789,0.00002242672,0.0004334189],"genre_scores_gemma":[0.000815471,0.9921441,0.00006020046,0.00121362,0.00168638,0.003770998,0.00009927418,0.00006098734,0.0001489552],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8813209,"threshold_uncertainty_score":0.9999773,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4210497484","doi":"10.1177/17407745221077691","title":"The PROTEUS-Trials Consortium: Optimizing the use of patient-reported outcomes in clinical trials","year":2022,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Cancer survivorship and care","field":"Medicine","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Queen's University","funders":"National Institute for Health and Care Research; Genentech; Wellcome Trust; Patient-Centered Outcomes Research Institute","keywords":"Clinical trial; Medicine; Protocol (science); Patient-reported outcome; MEDLINE; Patient recruitment; Outcome (game theory); Dissemination; Outcomes research; Family medicine; Quality of life (healthcare); Alternative medicine; Nursing","retraction":null,"screen_n_in":null,"score":{"opus":0.5986131313549837,"gpt":0.5473205174119634,"spread":0.05129261394302032,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3063453,0.0003415508,0.006886642,0.0001295095,0.0003431234,0.0001028786,0.0005104524,0.0003480395,0.000668051],"category_scores_gemma":[0.5865119,0.0001668122,0.00397957,0.0005790524,0.0006373616,0.00008848152,0.0004617372,0.001687653,0.00001065443],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008949689,"about_ca_system_score_gemma":0.001017255,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004703369,"about_ca_topic_score_gemma":0.0002094814,"domain_scores_codex":[0.843606,0.1199249,0.03250585,0.001212819,0.001933903,0.0008165482],"domain_scores_gemma":[0.3987855,0.5786055,0.01747774,0.003749943,0.0007420862,0.000639183],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.03849668,0.003006328,0.5497279,0.00009611836,0.004549311,0.0002294211,0.001128541,0.000316705,0.0002573838,0.0009502271,0.04068469,0.3605567],"study_design_scores_gemma":[0.0503573,0.005931515,0.08546609,0.0004218276,0.005411248,0.00005782269,0.00631687,0.000388398,0.0008167796,0.002308273,0.8416213,0.0009025487],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.957175,0.001834609,0.0001020538,0.01691868,0.01233074,0.01103918,0.0002422884,0.00009162387,0.0002658951],"genre_scores_gemma":[0.9886655,0.001811387,0.0004113746,0.00464566,0.001646704,0.00157442,0.00004834971,0.00007267762,0.001123942],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8009366,"threshold_uncertainty_score":0.7332107,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2150387089","doi":"10.1177/1740774507079237","title":"Clinical decision making and the expected value of information","year":2007,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"SickKids Foundation; University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Sample size determination; Ramipril; Medicine; Value of information; Clinical trial; Randomized controlled trial; Sample (material); Actuarial science; Outcome (game theory); Event (particle physics); Value (mathematics); Statistics; Computer science; Surgery; Internal medicine; Economics; Artificial intelligence; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.7330432828010638,"gpt":0.6306699529826952,"spread":0.1023733298183686,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4605269,0.000129364,0.002414949,0.0001930681,0.0001208545,0.00006675218,0.0002829075,0.0003385989,0.0001722852],"category_scores_gemma":[0.353015,0.0001057108,0.0004856605,0.0001840416,0.0003834979,0.000420558,0.00009961509,0.0003075532,0.0004754306],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005884466,"about_ca_system_score_gemma":0.00009704863,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005878767,"about_ca_topic_score_gemma":0.00001358447,"domain_scores_codex":[0.9655337,0.004366868,0.02921137,0.0004149657,0.0001557681,0.0003173098],"domain_scores_gemma":[0.8158072,0.1702784,0.01272018,0.0008087675,0.0001628769,0.00022268],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.001898934,0.0001817852,0.234435,0.0001410566,0.0002486211,5.798304e-7,0.001268368,0.0000392052,2.478366e-7,0.6525919,0.01922255,0.08997168],"study_design_scores_gemma":[0.01558181,0.0002767351,0.6430249,0.0003015401,0.00005485921,0.000004254751,0.001246121,0.006087728,0.000002394781,0.2489387,0.0841346,0.0003463696],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7911985,0.002288649,0.1827188,0.01136563,0.004353983,0.002037644,0.0001095233,0.00005698147,0.005870352],"genre_scores_gemma":[0.9694849,0.0006684831,0.01718019,0.01133079,0.001253606,0.0000207904,0.000009481081,0.00001353297,0.00003825341],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4085899,"threshold_uncertainty_score":0.6524348,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2153384744","doi":"10.1177/1740774506075667","title":"One relative risk versus two odds ratios: implications for meta-analyses involving paired and unpaired binary data","year":2007,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":30,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Robarts Clinical Trials; Western University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Odds ratio; Relative risk; Meta-analysis; Odds; Statistics; Pooling; Contrast (vision); Diagnostic odds ratio; Confidence interval; Medicine; Logistic regression; Mathematics; Internal medicine; Computer science; Artificial intelligence","retraction":null,"screen_n_in":null,"score":{"opus":0.9940942495411971,"gpt":0.7671393654451276,"spread":0.2269548840960695,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7018186,0.0003811184,0.01091346,0.0003692564,0.0004710232,0.0007312871,0.00249069,0.0002315013,0.001966858],"category_scores_gemma":[0.8196585,0.0001888317,0.005600981,0.001520846,0.0002722482,0.0007107396,0.0005606653,0.0003057369,0.0003572104],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000265751,"about_ca_system_score_gemma":0.0001583674,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002629422,"about_ca_topic_score_gemma":0.0001888482,"domain_scores_codex":[0.8709118,0.07963872,0.0426168,0.003248488,0.003038297,0.0005458764],"domain_scores_gemma":[0.2075313,0.7429265,0.03074235,0.01510562,0.00289553,0.0007987067],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00211641,0.001356087,0.03933496,0.0001065764,0.1158151,0.000006165643,0.0006198799,0.0001743392,0.0008861662,0.0520075,0.3318895,0.4556874],"study_design_scores_gemma":[0.01567216,0.001350435,0.146226,0.00007971212,0.2539761,0.000004011113,0.001887257,0.03249985,0.000113191,0.364731,0.1819711,0.001489134],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.05829004,0.02918484,0.8823537,0.01299167,0.002274408,0.008936162,0.00242772,0.00006799332,0.003473483],"genre_scores_gemma":[0.8325515,0.001188663,0.1577068,0.0008778042,0.001611713,0.0004238919,0.0003935864,0.00005308913,0.005192929],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7742615,"threshold_uncertainty_score":0.9989455,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3154504422","doi":"10.1177/17407745211001504","title":"Strategies for facilitating the delivery of cluster randomized trials in hospitals: A study informed by the CFIR-ERIC matching tool","year":2021,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":30,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Randomized controlled trial; Blueprint; Cluster randomised controlled trial; Implementation research; Process management; Matching (statistics); Identification (biology); Medical education; Knowledge management; Medicine; Computer science; Nursing; Business; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.836595647939426,"gpt":0.7459030377319846,"spread":0.09069261020744135,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4216517,0.0002014971,0.003609644,0.00007557211,0.0006642566,0.00008551295,0.0005125157,0.0001959216,0.0002913375],"category_scores_gemma":[0.6488962,0.00009819942,0.0007645563,0.0005333483,0.0003093877,0.0003276949,0.000231768,0.0006900336,0.00003151851],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000084262,"about_ca_system_score_gemma":0.003585613,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000464214,"about_ca_topic_score_gemma":0.0005273695,"domain_scores_codex":[0.8147227,0.1629058,0.02005467,0.0005780608,0.0008167968,0.0009220135],"domain_scores_gemma":[0.2036228,0.790471,0.004597575,0.000636509,0.0005507543,0.0001213629],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.1314164,0.001582535,0.01916321,0.002887615,0.001791955,0.000008284078,0.4876007,0.0003247869,0.0008929829,0.01001507,0.2329897,0.1113268],"study_design_scores_gemma":[0.3542776,0.0005051463,0.00448059,0.0004108589,0.0003159514,5.408272e-7,0.605297,0.0004712309,0.0000304486,0.01906619,0.01486294,0.0002815117],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.955813,0.0001596886,0.004826854,0.01618847,0.001446212,0.02092244,0.0003045071,0.00002742639,0.000311418],"genre_scores_gemma":[0.9693474,0.0002277821,0.002624335,0.01931465,0.0006063113,0.007459516,0.00001866274,0.00002156167,0.0003798466],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6275652,"threshold_uncertainty_score":0.6360724,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2106972824","doi":"10.1177/1740774513517063","title":"Simulating sequential multiple assignment randomized trials to generate optimal personalized warfarin dosing strategies","year":2014,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Pharmacogenetics and Drug Metabolism","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":30,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; McGill University Health Centre","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Dosing; Warfarin; Pharmacodynamics; Medicine; Pharmacokinetics; Computer science; Intensive care medicine; Pharmacology; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.6609817779989111,"gpt":0.6237253747213521,"spread":0.03725640327755897,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1764513,0.000598011,0.005654797,0.0001837869,0.0004719673,0.0002833848,0.0005558379,0.0007591391,0.002723144],"category_scores_gemma":[0.08837075,0.0004599662,0.002220573,0.0002495336,0.0005126284,0.0001574948,0.000285502,0.001160677,0.0003081437],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005692675,"about_ca_system_score_gemma":0.0003124592,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002857015,"about_ca_topic_score_gemma":0.000002850679,"domain_scores_codex":[0.9211854,0.06989657,0.00642697,0.001022426,0.0005210407,0.0009476153],"domain_scores_gemma":[0.9079105,0.08803499,0.002140132,0.0004813036,0.0002941075,0.001138923],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.2028599,0.001270287,0.0007273549,0.00008594028,0.004119,0.00003986302,0.001301898,0.2188796,0.3264737,0.007566032,0.005710573,0.2309658],"study_design_scores_gemma":[0.4256658,0.0002996667,0.00008672919,0.00005242133,0.003639535,0.000005367802,0.0002036446,0.2650559,0.06490953,0.003130211,0.2359518,0.0009993712],"study_design_candidate":"randomized_trial","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8651765,0.001571821,0.1115171,0.001823098,0.01251036,0.004543598,0.0001842089,0.0002370073,0.002436347],"genre_scores_gemma":[0.9760075,0.0006676341,0.009555801,0.005262007,0.007470133,0.000319233,0.0000542825,0.00007212972,0.0005913061],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2615642,"threshold_uncertainty_score":0.9997852,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2053269013","doi":"10.1191/1740774505cn62oa","title":"A web-based system for managing and co-ordinating multiple multisite studies","year":2005,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Scientific Computing and Data Management","field":"Decision Sciences","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Alberta Cancer Foundation","funders":"National Cancer Institute","keywords":"Computer science; Data collection; Process (computing); Track (disk drive); Data science; Web application; Resource (disambiguation); Data quality; Quality assurance; World Wide Web; Process management; Database; Information retrieval; Operations management; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.7270303352608196,"gpt":0.623113560823416,"spread":0.1039167744374037,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1343374,0.000156846,0.001091754,0.0002397407,0.0004016215,0.0005709192,0.0005895107,0.00006997566,0.00001890056],"category_scores_gemma":[0.1954978,0.0001055085,0.0003556372,0.000341656,0.0001832445,0.0001788276,0.0003285108,0.0001090184,0.000140365],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003917346,"about_ca_system_score_gemma":0.00003188965,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000004904339,"about_ca_topic_score_gemma":0.00002099832,"domain_scores_codex":[0.9930794,0.001861108,0.00295082,0.001071476,0.0007331707,0.0003040425],"domain_scores_gemma":[0.895841,0.1016905,0.001029208,0.000928492,0.0003371784,0.0001736061],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001553496,0.00008055661,0.01377996,0.00006075399,0.0000926067,0.000004002977,0.0001074831,0.000787034,0.00004661555,0.0006182768,0.06674992,0.9175174],"study_design_scores_gemma":[0.009039732,0.0001436758,0.00314563,0.000243019,0.0001114871,0.000001455203,0.003123733,0.7254941,0.0001856773,0.001399077,0.2568279,0.0002845124],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7469977,0.002725451,0.2197027,0.01099406,0.01118751,0.004247002,0.0003960054,0.0006765199,0.00307312],"genre_scores_gemma":[0.9657532,0.00001742238,0.03193844,0.0004308672,0.0008930865,0.00004630489,0.000007688873,0.00001061242,0.0009023963],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9172329,"threshold_uncertainty_score":0.8913819,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2027196621","doi":"10.1177/1740774513475530","title":"Challenges in the research ethics review of cluster randomized trials: International survey of investigators","year":2013,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":27,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Health Sciences Centre; Sunnybrook Health Science Centre; University of Calgary; Robarts Clinical Trials; Western University; Foothills Medical Centre; Ottawa Hospital; University of Ottawa","funders":"Canadian Institutes of Health Research","keywords":"Research ethics; CRTS; Informed consent; Medicine; Research design; Family medicine; Ethics committee; Clinical trial; Alternative medicine; Medical education; Psychology; Political science; Psychiatry; Sociology; Social science; Computer science; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.9892507882343728,"gpt":0.8021300603586891,"spread":0.1871207278756838,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.9267082,0.0001760927,0.006333811,0.0002967873,0.00003970928,0.0000274278,0.001261613,0.001285892,0.0009608806],"category_scores_gemma":[0.9946862,0.00009010985,0.001411572,0.0007043712,0.003523107,0.0000779824,0.0004053272,0.008248818,0.0001006194],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003971466,"about_ca_system_score_gemma":0.002164918,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007634964,"about_ca_topic_score_gemma":0.0003139951,"domain_scores_codex":[0.2374423,0.7172578,0.03211665,0.001587552,0.01062707,0.0009686819],"domain_scores_gemma":[0.001802323,0.9874453,0.002194374,0.001436228,0.006831602,0.0002902081],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"randomized_trial","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.3155134,0.008902002,0.04524231,0.06389979,0.0054972,0.00005116388,0.004020125,0.000003068316,0.0002746015,0.172619,0.1583479,0.2256294],"study_design_scores_gemma":[0.3956088,0.002218417,0.09532828,0.07650532,0.000715138,0.000008308679,0.0009849657,0.0002425142,0.0003578613,0.4188405,0.00879725,0.0003925935],"study_design_candidate":"randomized_trial","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"review","genre_scores_codex":[0.05936015,0.1183215,0.000087868,0.7721818,0.002814926,0.02149563,0.00008560916,0.00003070874,0.0256218],"genre_scores_gemma":[0.2532888,0.7378477,0.0009518837,0.006376913,0.000757255,0.0004367859,0.00003213872,0.00002602523,0.0002825175],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.7658049,"threshold_uncertainty_score":0.9999524,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2057565945","doi":"10.1177/1740774509356461","title":"Does the source of death information affect cancer screening efficacy results? A study of the use of mortality review versus death certificates in four randomized trials","year":2010,"lang":"en","type":"article","venue":"Clinical Trials","topic":"Global Cancer Incidence and Screening","field":"Medicine","cited_by":26,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Public Health Ontario; University of Toronto","funders":"National Cancer Institute; National Center for Research Resources; U.S. Public Health Service; Centers for Disease Control and Prevention","keywords":"Medicine; Death certificate; Cause of death; Mortality rate; Randomized controlled trial; Confidence interval; Internal medicine; Disease","retraction":null,"screen_n_in":null,"score":{"opus":0.7564123322173147,"gpt":0.5769507584911765,"spread":0.1794615737261382,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0927739,0.0002046497,0.004397222,0.0000717439,0.0000582368,0.00002283397,0.0003923434,0.0001832745,0.00004645559],"category_scores_gemma":[0.4321097,0.00007696005,0.00134991,0.0004276268,0.0002840946,0.0001995753,0.0001231091,0.0005790684,0.000001038748],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002111321,"about_ca_system_score_gemma":0.0002410754,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004508098,"about_ca_topic_score_gemma":0.0006329188,"domain_scores_codex":[0.9785615,0.01177144,0.008244341,0.0003067251,0.0008803569,0.0002356542],"domain_scores_gemma":[0.8751583,0.1135451,0.008572207,0.001777188,0.0007910462,0.0001562518],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"observational","study_design_scores_codex":[0.6080602,0.0009786115,0.2830593,0.0009669472,0.002577148,0.000003478373,0.001307683,0.0006301424,0.0006652218,0.0001130136,0.001595212,0.1000431],"study_design_scores_gemma":[0.4553485,0.001235847,0.521372,0.007332476,0.007875127,0.000001874151,0.0005469241,0.000617664,0.002283655,0.00008542588,0.003060672,0.0002398656],"study_design_candidate":"randomized_trial","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9893383,0.0008697021,0.0001200734,0.001705594,0.0009018149,0.006933846,0.00006643532,0.00001408761,0.00005009927],"genre_scores_gemma":[0.996446,0.002504025,0.0003070651,0.0002970706,0.000260115,0.0001506961,0.000007589023,0.000009093896,0.00001832344],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3393358,"threshold_uncertainty_score":0.9341802,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}