{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02500961,0.0009644764,0.0009065256,0.002594852,0.000384421,0.001288581,0.00138188,0.0008386004,0.001394649],"category_scores_gemma":[0.04330796,0.0004025652,0.001487276,0.001061014,0.00060718,0.001827921,0.001688283,0.000925892,0.0005499112],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001082432,"about_ca_system_score_gemma":0.002991675,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001129981,"about_ca_topic_score_gemma":0.001117412,"domain_scores_codex":[0.9838096,0.008396339,0.002543733,0.0007899769,0.004037303,0.0004230079],"domain_scores_gemma":[0.958635,0.01488398,0.006419438,0.001536563,0.01754318,0.0009817216],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.003414977,0.004134196,0.3546175,0.00179683,0.0006212706,0.0002604881,0.001575212,0.004198483,0.02166073,0.001109138,0.005447527,0.6011636],"study_design_scores_gemma":[0.002492318,0.03866544,0.7629622,0.001922604,0.001501055,0.001804773,0.003194845,0.1131102,0.04510577,0.002409628,0.02637211,0.0004589943],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6983786,0.001420335,0.271578,0.001334944,0.00050493,0.0183167,0.002168516,0.00103566,0.0052624],"genre_scores_gemma":[0.4934897,0.0009094145,0.486164,0.0005350964,0.0001915238,0.01461478,0.002647505,0.00009236806,0.001355509],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02500961,"threshold_uncertainty_score":0.1322651,"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."}}