{"id":"W4390106639","doi":"10.5489/cuaj.8571","title":"Inter-observer variance of examiner scoring in urology Objective Structured Clinical Examinations (OSCEs)","year":2023,"lang":"en","type":"article","venue":"Canadian Urological Association Journal","topic":"Innovations in Medical Education","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kingston General Hospital; Queen's University","funders":"Canadian Urological Association","keywords":"Objective structured clinical examination; Medicine; Checklist; Summative assessment; Context (archaeology); Urology; Medical education; Psychology; Formative assessment","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07489786,0.0005430807,0.0009475752,0.00170838,0.0007843453,0.001366065,0.0008563583,0.0006723676,0.0008464189],"category_scores_gemma":[0.1315113,0.000553711,0.001315747,0.001205722,0.001410684,0.0008739163,0.002167339,0.0007988494,0.0004552013],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009947859,"about_ca_system_score_gemma":0.0007344024,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001726637,"about_ca_topic_score_gemma":0.002566012,"domain_scores_codex":[0.9034581,0.05935711,0.009979018,0.007810649,0.01800135,0.001393804],"domain_scores_gemma":[0.7726176,0.178762,0.01121266,0.01263569,0.0237947,0.0009774128],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002485987,0.0003055247,0.9011538,0.0005706427,0.002645313,0.0001777131,0.01154368,0.003343192,0.004595362,0.0009848535,0.001731993,0.07046189],"study_design_scores_gemma":[0.0001396862,0.001925146,0.9680743,0.0001753148,0.000735489,0.0005370965,0.002184706,0.01627975,0.004703918,0.001077657,0.004039961,0.0001269976],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9338524,0.00136917,0.05768503,0.0001040214,0.0003260005,0.001042562,0.0005220061,0.0002143092,0.004884506],"genre_scores_gemma":[0.9909979,0.0001016588,0.007340745,0.00003271169,0.00003714902,0.0005070237,0.000361981,0.0000779557,0.0005429919],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07489786,"threshold_uncertainty_score":0.3961024,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03521915002844403,"score_gpt":0.3300977586150169,"score_spread":0.2948786085865729,"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."}}