{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009146445,0.0001025858,0.0003779628,0.00004447885,0.0001783454,0.000001464513,0.0001935968,0.0001258499,0.000009208175],"category_scores_gemma":[0.004399898,0.00006046818,0.00009035323,0.0004214439,0.0001007655,0.00009375178,0.0000727687,0.0005100968,2.582253e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003800111,"about_ca_system_score_gemma":0.002393232,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003791564,"about_ca_topic_score_gemma":0.001231858,"domain_scores_codex":[0.9969073,0.001170015,0.001005182,0.00020142,0.0004193377,0.0002967806],"domain_scores_gemma":[0.9924764,0.006338143,0.0005441286,0.0003328398,0.0002557071,0.0000527509],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000301474,0.0001390018,0.9863392,0.0001020075,0.000007558475,4.429053e-7,0.009414697,0.001803081,0.00005370489,0.0007423481,0.0001749465,0.001192887],"study_design_scores_gemma":[0.0005693404,0.00004453844,0.948642,0.000390477,0.00001168267,5.212588e-7,0.04492901,0.004325017,0.00001542163,0.0008570307,0.0001546544,0.0000603336],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9801221,0.00003886791,0.01328773,0.002432605,0.0006541432,0.00160585,0.00005336994,0.00001471978,0.001790585],"genre_scores_gemma":[0.994467,0.000006985763,0.00471038,0.000296848,0.00006372817,0.0004238398,0.00001915414,0.00000631196,0.000005814207],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03769721,"threshold_uncertainty_score":0.5267406,"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."}}