{"id":"W4415212837","doi":"10.2196/80752","title":"Quality Assessment of Large Language Model–Generated Medical Dialogue for Clinical Vignettes: Evaluation Study","year":2025,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Interview; Quality (philosophy); Quality assessment; Clinical Practice; Quality assurance; Evaluation methods; MEDLINE; Risk assessment","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05833929,0.0009308574,0.001274863,0.003665265,0.0008502529,0.002704418,0.001775584,0.001301591,0.00178258],"category_scores_gemma":[0.2712951,0.0004641449,0.001642306,0.002186401,0.001139066,0.001925595,0.00259994,0.0008321009,0.0004158718],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003237206,"about_ca_system_score_gemma":0.001367705,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001848736,"about_ca_topic_score_gemma":0.001790619,"domain_scores_codex":[0.9240373,0.06113729,0.005400736,0.002679959,0.006263506,0.0004812014],"domain_scores_gemma":[0.5864722,0.3288122,0.02616743,0.0125794,0.04100393,0.004964814],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.02478382,0.01201674,0.3048331,0.007230127,0.002327491,0.001575831,0.05271338,0.03634868,0.01048972,0.001702999,0.005299965,0.5406782],"study_design_scores_gemma":[0.00542426,0.03373212,0.4856974,0.001761591,0.002381058,0.00297854,0.0254601,0.4061789,0.01903857,0.002728797,0.01368225,0.0009363731],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9653889,0.0007175915,0.02703749,0.0002045569,0.00007218914,0.003529828,0.0006921686,0.0006033392,0.001754012],"genre_scores_gemma":[0.9572631,0.0003719908,0.03783853,0.0001193264,0.00005377302,0.002382899,0.001475118,0.0001009555,0.0003943616],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9416607,"threshold_uncertainty_score":0.3085313,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6501991757279266,"score_gpt":0.7363966636300987,"score_spread":0.08619748790217208,"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."}}