{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008687357,0.0007293174,0.0006583649,0.0006690275,0.0002815094,0.0007655305,0.0006795864,0.0005927602,0.001262646],"category_scores_gemma":[0.02657036,0.0004711665,0.0005986611,0.0002862691,0.0004570385,0.0008607809,0.001498569,0.0004771354,0.0005141424],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003398512,"about_ca_system_score_gemma":0.0006998199,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008124479,"about_ca_topic_score_gemma":0.00160233,"domain_scores_codex":[0.995657,0.002414048,0.0005402938,0.0006034551,0.0005876194,0.0001975665],"domain_scores_gemma":[0.9798918,0.01167345,0.001173858,0.001736298,0.00340978,0.002114773],"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":[0.01307923,0.007530574,0.3079334,0.002466525,0.0005962895,0.002416002,0.0470987,0.005637233,0.1095977,0.0003961503,0.004479675,0.4987684],"study_design_scores_gemma":[0.001305041,0.06534313,0.7861315,0.0006740645,0.0009338679,0.008177764,0.0196935,0.05302296,0.04789459,0.0004725255,0.01574189,0.0006091251],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9962953,0.00005964465,0.002659858,0.00002843557,0.00001235317,0.0001841747,0.00006273267,0.0002390685,0.0004583958],"genre_scores_gemma":[0.9721196,0.0001167723,0.02601268,0.00007542609,0.0000179916,0.0003120528,0.000273048,0.00006161331,0.001010704],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9913126,"threshold_uncertainty_score":0.04594374,"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."}}