{"id":"W4411373846","doi":"10.2196/76128","title":"Comparative Analysis of Generative Artificial Intelligence Systems in Solving Clinical Pharmacy Problems: Mixed Methods Study","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Pharmacy; Intraclass correlation; Medicine; Artificial intelligence; Computer science; Family medicine; Psychometrics; Clinical psychology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06223187,0.000754245,0.001033092,0.004288674,0.001651175,0.003385373,0.002122537,0.001022108,0.002613293],"category_scores_gemma":[0.1489584,0.0006806374,0.00176649,0.00338278,0.002023896,0.00257306,0.003633006,0.001088249,0.0002875093],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003813189,"about_ca_system_score_gemma":0.004919421,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00194857,"about_ca_topic_score_gemma":0.003236263,"domain_scores_codex":[0.9255553,0.05934644,0.005184163,0.002906492,0.005943913,0.001063582],"domain_scores_gemma":[0.7712971,0.1865482,0.01228034,0.0115666,0.01670282,0.001604896],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.004069796,0.01120867,0.2092674,0.009560321,0.002685793,0.0004773268,0.2714094,0.003313615,0.002691741,0.009079343,0.001283109,0.4749536],"study_design_scores_gemma":[0.003944289,0.04843232,0.450031,0.007332508,0.004545951,0.001394349,0.3788122,0.0393653,0.01328427,0.02646691,0.02571662,0.0006742498],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9652595,0.001143813,0.02048293,0.0002624763,0.00003167879,0.0091385,0.0002086746,0.00004834092,0.003424028],"genre_scores_gemma":[0.9248486,0.0008430406,0.05705334,0.0003255676,0.00004457669,0.01591683,0.0002478634,0.00003290238,0.000687347],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06223187,"threshold_uncertainty_score":0.3291175,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3585744896393742,"score_gpt":0.6052675788860328,"score_spread":0.2466930892466586,"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."}}