{"id":"W4410380081","doi":"10.2196/71966","title":"Use of a Medical Communication Framework to Assess the Quality of Generative Artificial Intelligence Replies to Primary Care Patient Portal Messages: Content Analysis","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":"Preprint; Quality (philosophy); Psychology; Primary care; Computer science; Generative grammar; Artificial intelligence; World Wide Web; Medicine; Family medicine","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":[0.09759452,0.0008204553,0.001270829,0.02313791,0.002708947,0.003309837,0.002198754,0.001105943,0.001580475],"category_scores_gemma":[0.2125169,0.0006915402,0.001464038,0.01371199,0.005782587,0.003126536,0.005719157,0.00126155,0.0002556284],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0100944,"about_ca_system_score_gemma":0.008347312,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004036778,"about_ca_topic_score_gemma":0.004626434,"domain_scores_codex":[0.9307865,0.04672912,0.007613524,0.002567587,0.010809,0.001494287],"domain_scores_gemma":[0.6388565,0.2787092,0.02666125,0.01033927,0.04407009,0.001363629],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.0007709421,0.0005185077,0.1095394,0.004910307,0.0002263106,0.0005321406,0.6316248,0.001269932,0.008658261,0.006606332,0.00318793,0.2321552],"study_design_scores_gemma":[0.0004493443,0.002091143,0.4227288,0.003422104,0.0005752257,0.0009710448,0.4831405,0.02092239,0.01968684,0.01002474,0.03550299,0.0004848815],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9249367,0.0004207542,0.04864797,0.001466286,0.00008722018,0.01505374,0.001430588,0.000350131,0.007606569],"genre_scores_gemma":[0.8259504,0.0003614705,0.1459897,0.0004128365,0.00005468373,0.02450619,0.001335402,0.0001286065,0.001260839],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.09759452,"threshold_uncertainty_score":0.5161353,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6349133649489475,"score_gpt":0.6068874196677573,"score_spread":0.02802594528119018,"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."}}