{"id":"W2913928619","doi":"10.1177/0310057x1704500410","title":"Validity Evidence of Non-Technical Skills Assessment Instruments in Simulated Anaesthesia Crisis Management","year":2017,"lang":"en","type":"article","venue":"Anaesthesia and Intensive Care","topic":"Patient Safety and Medication Errors","field":"Health Professions","cited_by":40,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Intraclass correlation; Cronbach's alpha; Medicine; Content validity; Internal consistency; Reliability (semiconductor); Medical education; Clinical psychology; Psychometrics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1057359,0.0006356827,0.0005899784,0.001767391,0.0008633059,0.001800275,0.001517027,0.001340136,0.001430447],"category_scores_gemma":[0.2631589,0.0006547131,0.001332971,0.001479985,0.003674748,0.001452946,0.002564142,0.001066567,0.0005539371],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001429059,"about_ca_system_score_gemma":0.002261918,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001882844,"about_ca_topic_score_gemma":0.002414777,"domain_scores_codex":[0.8794724,0.07145691,0.01114924,0.00545236,0.03089853,0.001570631],"domain_scores_gemma":[0.6109214,0.3025885,0.02668001,0.01796928,0.0396469,0.002193857],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001719329,0.0006948737,0.9045798,0.0008794396,0.0007076605,0.0001437974,0.006656052,0.001732762,0.00188884,0.001341532,0.0005809228,0.07907493],"study_design_scores_gemma":[0.0002912047,0.005208223,0.9629024,0.001393982,0.0003053316,0.000618685,0.003616856,0.01341569,0.0054188,0.002045563,0.004677162,0.0001063097],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.978096,0.001860457,0.01171078,0.0004669292,0.0002257102,0.0006808647,0.0002355953,0.00004044334,0.006683199],"genre_scores_gemma":[0.9934741,0.0002758052,0.005077334,0.000132443,0.00005389748,0.0003959668,0.0002321751,0.00002561839,0.0003326428],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1057359,"threshold_uncertainty_score":0.5591918,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07757743559712646,"score_gpt":0.4462408457032147,"score_spread":0.3686634101060882,"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."}}