{"id":"W4389814783","doi":"10.1101/2023.12.14.23299971","title":"GPT for RCTs?: Using AI to determine adherence to reporting guidelines","year":2023,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ottawa Hospital; University of British Columbia","funders":"Canadian Institutes of Health Research","keywords":"Guideline; Medicine; Hyperparameter; Test (biology); Computer science; Artificial intelligence; Machine learning; Medical physics","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7224005,0.002346373,0.005088916,0.01059725,0.002594948,0.01217806,0.006810137,0.004530496,0.005159511],"category_scores_gemma":[0.9086467,0.00205377,0.01167405,0.01015612,0.004602476,0.01089107,0.007807027,0.006874985,0.001683386],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006474216,"about_ca_system_score_gemma":0.01701078,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002961863,"about_ca_topic_score_gemma":0.004052332,"domain_scores_codex":[0.1474824,0.6313292,0.1575216,0.02020552,0.04200673,0.001454567],"domain_scores_gemma":[0.03337819,0.7857858,0.09012871,0.05846274,0.03092832,0.001316334],"domain_codex":"methods","domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.006170626,0.0006400591,0.2079261,0.07654784,0.03802447,0.000553794,0.02142841,0.01153065,0.002167348,0.01800163,0.06397256,0.5530365],"study_design_scores_gemma":[0.01121232,0.01065847,0.2173,0.1104635,0.03274921,0.002530223,0.008858559,0.2281276,0.01914312,0.1419937,0.2145229,0.002440447],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1496121,0.04999142,0.5764986,0.07579838,0.005473885,0.08358408,0.02608018,0.01050409,0.02245725],"genre_scores_gemma":[0.5556255,0.003168329,0.3556089,0.01059549,0.001030916,0.06680512,0.005429443,0.0008626903,0.0008736356],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2775995,"threshold_uncertainty_score":0.3423296,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9556734511813036,"score_gpt":0.6544568837914889,"score_spread":0.3012165673898147,"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."}}