{"id":"W4386699918","doi":"10.1101/2023.09.12.23295381","title":"Randomized Controlled Trials Evaluating AI in Clinical Practice: A Scoping Evaluation","year":2023,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"Public Health Ontario; University of Toronto","funders":"","keywords":"Clinical trial; Randomized controlled trial; Medicine; Health care; Clinical Practice; Intervention (counseling); Systematic review; Alternative medicine; MEDLINE; Artificial intelligence; Medical physics; Family medicine; Nursing; Computer science; Pathology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.301556,0.0003398737,0.00659526,0.0005857685,0.00007781935,0.0001079979,0.0001664472,0.0007465298,0.000236702],"category_scores_gemma":[0.6674558,0.000251223,0.001428921,0.0003521622,0.0001319638,0.0001186205,0.0002008403,0.001822218,0.0002613049],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004885822,"about_ca_system_score_gemma":0.007731109,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001055266,"about_ca_topic_score_gemma":0.0001827433,"domain_scores_codex":[0.9575219,0.02814468,0.01063807,0.001109591,0.0020588,0.0005269767],"domain_scores_gemma":[0.8436952,0.1406112,0.00775759,0.001838464,0.005613429,0.0004841738],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.6080341,0.0009205822,0.01850471,0.00236974,0.001919546,0.0000635714,0.007174017,0.00286982,0.0001711084,0.0001871138,0.001549099,0.3562366],"study_design_scores_gemma":[0.3048234,0.0004610078,0.004047758,0.02706021,0.01036763,0.00001788012,0.002686118,0.6040184,0.0002267259,0.04554421,0.0000842273,0.0006623348],"study_design_candidate":"randomized_trial","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8943119,0.005889674,0.001811841,0.04622021,0.0171405,0.03249513,0.000003180602,0.0001997229,0.001927845],"genre_scores_gemma":[0.9770037,0.003263315,0.003175774,0.002572054,0.004315024,0.00851002,0.0001902168,0.00008170246,0.0008881809],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6011487,"threshold_uncertainty_score":0.999994,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6999636305230319,"score_gpt":0.6744827929932773,"score_spread":0.02548083752975461,"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."}}