{"id":"W2424610854","doi":"10.1080/10401334.2016.1190730","title":"In Brief: Validity of Case Summaries in Written Examinations of Clinical Reasoning","year":2016,"lang":"en","type":"article","venue":"Teaching and Learning in Medicine","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; McGill University","funders":"","keywords":"Generalizability theory; Test (biology); Inter-rater reliability; Rubric; Objective structured clinical examination; Construct validity; Medicine; MEDLINE; Curriculum; Psychology; Clinical psychology; Psychometrics; Medical education; Mathematics education; Rating scale; Developmental psychology","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01177032,0.0001189012,0.000845905,0.0003087844,0.00004283598,0.000002308456,0.000055081,0.0001526297,0.00005720991],"category_scores_gemma":[0.3275527,0.00007903238,0.00004842546,0.0002129099,0.0004134365,0.00006008891,0.00004591296,0.001237959,8.729735e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003220725,"about_ca_system_score_gemma":0.00008466674,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006305557,"about_ca_topic_score_gemma":0.0004809429,"domain_scores_codex":[0.9969403,0.001177809,0.001156334,0.0002830918,0.0002313889,0.000211103],"domain_scores_gemma":[0.9593728,0.03999376,0.0002677753,0.0001917707,0.00005417437,0.0001197294],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001154648,0.0002352855,0.913525,0.0000558423,0.000009667776,0.0006511078,0.002544398,0.00001224464,0.0001272989,0.0005546076,0.0001721366,0.08199694],"study_design_scores_gemma":[0.006348607,0.0009059682,0.9757882,0.01285075,0.00005242239,0.0002525165,0.001917987,0.0006374808,0.00002296734,0.0001882931,0.000941503,0.00009332965],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9940107,0.0003257433,0.0002121667,0.003417917,0.0001060002,0.0001279577,0.000001780472,0.00001926594,0.001778483],"genre_scores_gemma":[0.997555,0.0002916768,0.001332828,0.0001521518,0.0001401812,0.000005043938,0.000006511489,0.00001230261,0.0005043517],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3157824,"threshold_uncertainty_score":0.9532158,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07290377866669441,"score_gpt":0.4191633828742299,"score_spread":0.3462596042075355,"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."}}