{"id":"W2922061857","doi":"10.1016/j.wneu.2019.03.059","title":"A Comparison of Visual Rating Scales and Simulated Virtual Reality Metrics in Neurosurgical Training: A Generalizability Theory Study","year":2019,"lang":"en","type":"article","venue":"World Neurosurgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":false,"ca_institutions":"McGill University; Artificial Intelligence in Medicine (Canada)","funders":"Fonds de Recherche du Québec - Santé; National Research Council Canada; Montreal Neurological Institute and Hospital; School of Life Sciences and Biotechnology Division of Life Sciences, Korea University; McGill University","keywords":"Generalizability theory; Rating scale; Medicine; Reliability (semiconductor); Scale (ratio); Neurosurgery; Medical physics; Inter-rater reliability; Intraclass correlation; Psychology; Psychometrics; Clinical psychology; Surgery; Developmental psychology","routes":{"ca_aff":true,"ca_fund":true,"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":[0.153656,0.001050749,0.001839794,0.003606362,0.001605866,0.003116955,0.003199615,0.002391817,0.007588113],"category_scores_gemma":[0.4145684,0.001068886,0.004130108,0.001985159,0.006226268,0.006101395,0.00470907,0.003413092,0.0009648702],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002413766,"about_ca_system_score_gemma":0.001573598,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002541005,"about_ca_topic_score_gemma":0.001813124,"domain_scores_codex":[0.8211489,0.1404407,0.00706145,0.0132739,0.01675138,0.001323602],"domain_scores_gemma":[0.2448163,0.6879598,0.01859307,0.02943799,0.01759999,0.001592801],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.04493212,0.008212732,0.7673171,0.002600366,0.01748538,0.0003274215,0.01843707,0.008111936,0.004393939,0.01618187,0.003112639,0.1088874],"study_design_scores_gemma":[0.01125821,0.05936786,0.8327032,0.001072189,0.009523413,0.0007360179,0.01066603,0.04297813,0.003971896,0.02182544,0.005468158,0.0004293914],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9504437,0.001615069,0.03308295,0.0005795474,0.0006769566,0.00243631,0.0003777043,0.0001142678,0.01067349],"genre_scores_gemma":[0.9927559,0.0001434685,0.004904585,0.0002953483,0.0001359721,0.0008585999,0.0002200746,0.00006133258,0.0006247257],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.153656,"threshold_uncertainty_score":0.8126203,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09848270130194468,"score_gpt":0.3880934796109528,"score_spread":0.2896107783090081,"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."}}