{"id":"W2071262436","doi":"10.1016/j.surg.2014.12.023","title":"Objective structured assessment of nontechnical skills: Reliability of a global rating scale for the in-training assessment in the operating room","year":2015,"lang":"en","type":"article","venue":"Surgery","topic":"Patient Safety and Medication Errors","field":"Health Professions","cited_by":41,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Cronbach's alpha; Inter-rater reliability; Scale (ratio); Accreditation; Medicine; Rating scale; Reliability (semiconductor); Certification; Intraclass correlation; Medical education; Descriptive statistics; Psychometrics; Statistics; Clinical psychology; Mathematics","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":[0.006252487,0.0009359292,0.0007683094,0.0009570194,0.00033934,0.0005325947,0.0005337544,0.0006207023,0.001327818],"category_scores_gemma":[0.0180801,0.0002818669,0.0009364202,0.0003794551,0.0006466219,0.0007137542,0.001099372,0.0007297501,0.0006672114],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004063052,"about_ca_system_score_gemma":0.001284338,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001032089,"about_ca_topic_score_gemma":0.003116282,"domain_scores_codex":[0.9947119,0.00166431,0.0007441266,0.0006029214,0.00203267,0.0002441998],"domain_scores_gemma":[0.9758018,0.007400351,0.006518159,0.001360675,0.007540544,0.001378567],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001485607,0.001938788,0.9247941,0.0003560313,0.0003985477,0.0001398233,0.00170415,0.001324241,0.01619326,0.0001643653,0.002131335,0.04936969],"study_design_scores_gemma":[0.0001043634,0.003206276,0.9891796,0.00004934088,0.00007170236,0.0003217635,0.0005553693,0.001843462,0.003706147,0.0001268368,0.0007947045,0.00004043109],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9851325,0.0001795354,0.007778166,0.00007996232,0.00007443065,0.0008795494,0.0007596466,0.0001163862,0.004999936],"genre_scores_gemma":[0.9860973,0.0002099609,0.01075397,0.0001264465,0.00004269375,0.0004647996,0.0010224,0.00004119085,0.001241244],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.006252487,"threshold_uncertainty_score":0.03306669,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1058024274036022,"score_gpt":0.47325009443716,"score_spread":0.3674476670335577,"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."}}