{"id":"W3022658701","doi":"10.2196/15444","title":"An Objective Structured Clinical Examination for Medical Student Radiology Clerkships: Reproducibility Study","year":2020,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Intraclass correlation; Objective structured clinical examination; Grading (engineering); Medicine; Reproducibility; Medical physics; Physical examination; Radiology; Medical education; Psychometrics","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":[0.08333082,0.0006453171,0.0006474264,0.001647081,0.0006235178,0.001294568,0.0009577528,0.00095598,0.001083289],"category_scores_gemma":[0.1635589,0.0004345944,0.001710093,0.0009819176,0.001506252,0.001034924,0.001890894,0.0006631513,0.0004788691],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007511814,"about_ca_system_score_gemma":0.001393586,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000551102,"about_ca_topic_score_gemma":0.001003422,"domain_scores_codex":[0.9236719,0.04277641,0.009216277,0.005799964,0.01764674,0.0008887749],"domain_scores_gemma":[0.746071,0.1401862,0.03097375,0.03294226,0.04775915,0.002067522],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001986147,0.00124952,0.9069757,0.0006060624,0.001309565,0.0001979364,0.004194489,0.001309914,0.004907124,0.0006452929,0.0008242669,0.07579378],"study_design_scores_gemma":[0.0003040683,0.009238838,0.9635441,0.0002787585,0.0008857252,0.001202133,0.001616017,0.009933061,0.006788772,0.0005328883,0.005545111,0.0001305409],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.970941,0.0007189743,0.0230791,0.0001160263,0.0001735772,0.001194232,0.000195472,0.00009547234,0.003486019],"genre_scores_gemma":[0.9901536,0.00009104912,0.008541862,0.00007705867,0.00006893071,0.0005720985,0.0001824897,0.00004713909,0.0002657717],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08333082,"threshold_uncertainty_score":0.4407008,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04895001446605569,"score_gpt":0.4964158936368512,"score_spread":0.4474658791707956,"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."}}