{"id":"W3033780089","doi":"10.1097/sla.0000000000003897","title":"Development and Evaluation of a Novel Instrument to Measure Severity of Intraoperative Events Using Video Data","year":2020,"lang":"en","type":"article","venue":"Annals of Surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":23,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Medicine; Interquartile range; Intraclass correlation; Construct validity; Confidence interval; Reliability (semiconductor); Body mass index; Surgery; Physical therapy; Internal medicine; Psychometrics","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":[0.001483726,0.00006675179,0.0003142104,0.00006423306,0.00001666317,0.000001934009,0.00003923092,0.00003171991,0.00005138117],"category_scores_gemma":[0.000939556,0.00005882034,0.00003099515,0.0002102035,0.00002356718,0.00009324855,0.00007157407,0.00004180332,4.291914e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000009462309,"about_ca_system_score_gemma":0.0002749874,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000208276,"about_ca_topic_score_gemma":0.000003289471,"domain_scores_codex":[0.998769,0.0000509071,0.0004140031,0.0001557661,0.0005293046,0.00008099719],"domain_scores_gemma":[0.999031,0.0001346918,0.0001397379,0.0001266743,0.0004590301,0.0001088552],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001670082,0.0007767806,0.2703171,0.0005927759,0.0007355556,0.000004214346,0.009377114,0.00156329,0.02631939,0.00006799305,0.0002185455,0.6883572],"study_design_scores_gemma":[0.003435018,0.000105107,0.8284046,0.001430537,0.0001823455,0.000005078416,0.001180807,0.04935558,0.1133943,0.00002027749,0.002259901,0.0002265201],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9983879,0.00009359165,0.0002173992,0.0009333398,0.00002745503,0.0001985487,0.000025853,0.000004278088,0.0001115742],"genre_scores_gemma":[0.9972939,0.00001008865,0.002128002,0.0005061122,0.00001717743,0.000001869763,0.00003710361,0.000004899468,8.113817e-7],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6881307,"threshold_uncertainty_score":0.2398624,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7643549078891856,"score_gpt":0.4565247145430722,"score_spread":0.3078301933461134,"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."}}