{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002106142,0.0002322899,0.001021676,0.000549054,0.00006441905,0.00003011065,0.00006533571,0.00007903099,0.0001296334],"category_scores_gemma":[0.001574963,0.0002033843,0.0001472295,0.001596129,0.0001433425,0.00008936351,0.00008472615,0.0004403981,0.00000338418],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003635721,"about_ca_system_score_gemma":0.000053536,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002343268,"about_ca_topic_score_gemma":0.00003129835,"domain_scores_codex":[0.9965398,0.0009598727,0.001107253,0.0005602462,0.0005017266,0.0003311179],"domain_scores_gemma":[0.9945549,0.004641308,0.0002441827,0.000287551,0.00007494303,0.0001970511],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001195926,0.001558725,0.9811652,0.00004220921,0.0000237843,0.00007558674,0.001160998,0.004125137,0.001077257,0.0002926077,0.000006523455,0.009276027],"study_design_scores_gemma":[0.00464766,0.0006357766,0.9125652,0.00005163028,0.00004158873,0.000008710224,0.0009409803,0.080255,0.0002948769,0.00006181082,0.0003007441,0.0001960207],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9981057,0.00005591034,0.00003037873,0.00016496,0.000232725,0.0006617053,0.000003890125,0.00008396797,0.0006607565],"genre_scores_gemma":[0.9994135,0.000003631355,0.00001320019,0.0003502062,0.0000402264,0.000005513991,0.00001104888,0.00002875079,0.000133921],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07612987,"threshold_uncertainty_score":0.8293773,"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."}}