{"id":"W4404917666","doi":"10.2196/59902","title":"Performance Comparison of Junior Residents and ChatGPT in the Objective Structured Clinical Examination (OSCE) for Medical History Taking and Documentation of Medical Records: Development and Usability Study","year":2024,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Documentation; Medical record; Checklist; Objective structured clinical examination; Medical history; Medical education; Usability; Psychology; Medicine; Computer science; Surgery","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006008161,0.000102269,0.0003098449,0.0001670704,0.0000599155,0.00001302267,0.00009788846,0.0002540464,0.0001830247],"category_scores_gemma":[0.005412224,0.0000756036,0.00002015319,0.0001712303,0.0003425729,0.0001430642,0.00003415298,0.0003913758,4.898337e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002828652,"about_ca_system_score_gemma":0.003024797,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006866333,"about_ca_topic_score_gemma":0.0009086246,"domain_scores_codex":[0.9969963,0.0003508855,0.001064047,0.0003158114,0.001132644,0.0001402918],"domain_scores_gemma":[0.9980993,0.001159494,0.0002152253,0.0001294198,0.0001664795,0.000230054],"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":[0.00009620086,0.000594405,0.3611597,0.0006649576,0.00001277863,5.440321e-7,0.070118,3.549339e-8,0.000003614878,0.00008312035,0.0001869971,0.5670797],"study_design_scores_gemma":[0.0002762261,0.0006337688,0.9452027,0.0008706806,0.00003139891,0.00001517129,0.04772996,0.004539151,0.00008092033,0.0001494162,0.0004072826,0.00006329077],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9938406,0.001074754,0.00008585001,0.002735191,0.0009570144,0.001261074,5.982359e-7,0.000010243,0.00003472938],"genre_scores_gemma":[0.9985467,0.0003588205,0.0002154356,0.0002560414,0.0002531993,0.0003172458,0.00002294564,0.000007829905,0.00002180217],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5840431,"threshold_uncertainty_score":0.6479329,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1379794360929445,"score_gpt":0.5268376338722495,"score_spread":0.388858197779305,"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."}}