{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04206537,0.0005113449,0.0004453842,0.003517432,0.0006247536,0.002523383,0.00106025,0.0008601254,0.002223003],"category_scores_gemma":[0.2800997,0.0002995892,0.0007755887,0.001430836,0.001708837,0.00277284,0.003617429,0.001293287,0.0004564895],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001215513,"about_ca_system_score_gemma":0.002058878,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007661296,"about_ca_topic_score_gemma":0.001448294,"domain_scores_codex":[0.9708945,0.01265082,0.006309195,0.001697049,0.007981707,0.0004667402],"domain_scores_gemma":[0.6873826,0.1913863,0.06076772,0.01566591,0.04075008,0.004047369],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001098363,0.0008975705,0.727582,0.001285302,0.0002673623,0.0004557211,0.008233975,0.00135151,0.004109738,0.001528142,0.002499357,0.2506909],"study_design_scores_gemma":[0.0002444792,0.004546001,0.9472743,0.001663964,0.0002829074,0.003401437,0.005446278,0.009334364,0.01352827,0.003889684,0.01025136,0.0001369869],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.973737,0.001507973,0.01472937,0.001133597,0.0002277964,0.001235212,0.0004214781,0.0001826934,0.006824961],"genre_scores_gemma":[0.9803274,0.0002889862,0.01726479,0.0001581306,0.00007454999,0.0005814268,0.0005439605,0.00002767572,0.0007330309],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9579346,"threshold_uncertainty_score":0.2224656,"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."}}