{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01059301,0.0002934456,0.0007267767,0.0001890716,0.0001961744,0.00003570975,0.0005626604,0.0007672576,0.001648313],"category_scores_gemma":[0.07683024,0.0002534278,0.0001536305,0.0008919556,0.0004523654,0.0002229749,0.0000846467,0.001332685,0.00003649388],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003548583,"about_ca_system_score_gemma":0.005472191,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002246658,"about_ca_topic_score_gemma":0.00001854299,"domain_scores_codex":[0.9925638,0.001161157,0.001742684,0.001944216,0.002175146,0.0004129748],"domain_scores_gemma":[0.9952651,0.0004635999,0.0004067738,0.001547309,0.001234329,0.001082891],"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":[0.0005445923,0.01327788,0.2022082,0.0002048038,0.0001570597,0.000009409772,0.0131557,8.256656e-7,0.0000574062,0.0007778279,0.02799089,0.7416155],"study_design_scores_gemma":[0.004385849,0.004328229,0.9617563,0.0001075162,0.0001801646,0.0000704245,0.01993137,0.003947842,0.0000528988,0.0003336534,0.004629628,0.0002760777],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9460122,0.00007380851,0.003574627,0.0418587,0.00324054,0.004779262,0.000004125548,0.0002169674,0.0002397795],"genre_scores_gemma":[0.9646858,0.00001964803,0.002491583,0.02445387,0.006120868,0.001531626,0.0005834707,0.00004799135,0.00006516134],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7595482,"threshold_uncertainty_score":0.9999918,"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."}}