{"id":"W4399706167","doi":"10.1016/j.jval.2024.03.2436","title":"PT22 Consistency Between Comparative Efficacy Estimates from Target Trial Emulations and Their Origin Target Trials: A Meta-Epidemiological Study","year":2024,"lang":"en","type":"article","venue":"Value in Health","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Cascades (Canada); University of Toronto; McMaster University","funders":"","keywords":"Consistency (knowledge bases); Epidemiology; Meta-analysis; Medicine; Computer science; Artificial intelligence; Internal medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07689719,0.0004594841,0.00592529,0.000561094,0.0003268121,0.0001818372,0.0002737243,0.0002139423,0.000560178],"category_scores_gemma":[0.0179298,0.0003909259,0.000423169,0.0004511617,0.0001403565,0.0003612282,0.00008113222,0.0005410622,0.0003754278],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005914538,"about_ca_system_score_gemma":0.0005807551,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0140176,"about_ca_topic_score_gemma":0.0001486154,"domain_scores_codex":[0.980971,0.007441257,0.009615462,0.001223798,0.0001207199,0.0006277608],"domain_scores_gemma":[0.9430918,0.0535888,0.002325824,0.0005660908,0.00004593969,0.0003815511],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002098518,0.00301805,0.5566837,0.001489778,0.008946856,0.00001714651,0.07890541,0.04376577,0.000001446241,0.2769625,0.02742732,0.0006834951],"study_design_scores_gemma":[0.0277661,0.002487274,0.1593479,0.0004442508,0.0003155377,0.0000043453,0.007610914,0.3141011,0.000002665015,0.4365639,0.05006183,0.001294207],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.895419,0.05088916,0.008739827,0.03279048,0.001464427,0.006842552,0.003317432,0.0002142659,0.0003228446],"genre_scores_gemma":[0.9823763,0.0002106233,0.01416347,0.00153127,0.0008761144,0.0005252817,0.0002329876,0.00003711964,0.00004678574],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3973357,"threshold_uncertainty_score":0.9998543,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8587779294536831,"score_gpt":0.5553820052404167,"score_spread":0.3033959242132664,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}