{"id":"W2054627871","doi":"10.1016/j.amjsurg.2011.11.006","title":"The script concordance test as a measure of clinical reasoning: a national validation study","year":2012,"lang":"en","type":"article","venue":"The American Journal of Surgery","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":71,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia; University of Ottawa; Western University; Queen's University; McMaster University; University of Alberta; Université de Montréal; Memorial University of Newfoundland; University of Calgary; McGill University","funders":"","keywords":"Concordance; Cronbach's alpha; Test (biology); Reliability (semiconductor); Measure (data warehouse); Medicine; Medical physics; Psychology; Clinical psychology; Psychometrics; Data mining; Internal medicine; Computer science","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"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":[0.02065901,0.0006988166,0.0009681949,0.001884476,0.001968918,0.001437424,0.001691852,0.001361086,0.001112217],"category_scores_gemma":[0.04268237,0.0007261321,0.00138016,0.001561201,0.001911334,0.00236596,0.002047119,0.001715631,0.0005819849],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002082428,"about_ca_system_score_gemma":0.003461648,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01122533,"about_ca_topic_score_gemma":0.01786256,"domain_scores_codex":[0.9871148,0.0052466,0.001525611,0.00123591,0.004525083,0.0003519299],"domain_scores_gemma":[0.948721,0.02087305,0.007218916,0.005586723,0.01521276,0.002387617],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001219511,0.002725509,0.9784502,0.00006832995,0.0003946835,0.0001143743,0.002310782,0.0005709244,0.00146799,0.0003012171,0.0009431948,0.01143324],"study_design_scores_gemma":[0.0003252112,0.002235121,0.9897614,0.0000609491,0.0002027559,0.000486438,0.001234564,0.002775818,0.001403111,0.000197121,0.001277002,0.00004052932],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9962944,0.0001210447,0.001061621,0.00007198074,0.00003361475,0.0002727386,0.0002588638,0.00001484377,0.00187088],"genre_scores_gemma":[0.9942053,0.0001554363,0.003067376,0.0001617716,0.00001826166,0.0004688455,0.001144406,0.00003619043,0.0007423552],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02065901,"threshold_uncertainty_score":0.1092566,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1085774429683783,"score_gpt":0.4208620363548373,"score_spread":0.312284593386459,"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."}}