{"id":"W4416572991","doi":"10.2196/76618","title":"Teaching Clinical Reasoning in Health Care Professions Learners Using AI-Generated Script Concordance Tests: Mixed Methods Formative Evaluation","year":2025,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Institut Universitaire en Santé Mentale de Québec; Université de Montréal; Institut universitaire en santé mentale de Montréal; Centre Hospitalier Universitaire Sainte-Justine; Institut national de psychiatrie légale Philippe-Pinel; Université du Québec à Montréal","funders":"","keywords":"Formative assessment; Concordance; Health care; Generative grammar; Expert system; Teaching method; Educational measurement; Health professions","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09584215,0.001931819,0.001946148,0.003259714,0.001394613,0.00345218,0.002987582,0.001425375,0.002922142],"category_scores_gemma":[0.1521405,0.0009025959,0.002511497,0.001872849,0.001592168,0.001632348,0.00326538,0.00124697,0.0008142739],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004421553,"about_ca_system_score_gemma":0.004830389,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001111693,"about_ca_topic_score_gemma":0.001986867,"domain_scores_codex":[0.9355768,0.04427157,0.007598687,0.002763155,0.008539857,0.001249936],"domain_scores_gemma":[0.7959815,0.1428642,0.01349834,0.009284097,0.03444806,0.003923771],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.02143692,0.04611655,0.1106588,0.005317255,0.001510481,0.0006643683,0.06333099,0.0099303,0.007874155,0.001804835,0.003127398,0.728228],"study_design_scores_gemma":[0.03043281,0.296364,0.3354301,0.007112162,0.003766937,0.001735358,0.0506535,0.0947001,0.1208165,0.01277116,0.04482689,0.001390455],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9486176,0.0004691201,0.02301989,0.000182243,0.0001220549,0.02305809,0.0005712695,0.0002917283,0.003668053],"genre_scores_gemma":[0.7937462,0.0007674275,0.1461596,0.000362341,0.0001639789,0.05537926,0.001041719,0.0001403621,0.002239189],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.09584215,"threshold_uncertainty_score":0.5068678,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.212115048452466,"score_gpt":0.6378854026687963,"score_spread":0.4257703542163304,"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."}}