{"id":"W4225736941","doi":"10.1097/bot.0000000000002110","title":"Validation of the AO Spine Sacral Classification System: Reliability Among Surgeons Worldwide","year":2021,"lang":"en","type":"article","venue":"Journal of Orthopaedic Trauma","topic":"Spinal Fractures and Fixation Techniques","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; Vancouver General Hospital","funders":"","keywords":"Medicine; Kappa; Cohen's kappa; Reproducibility; Reliability (semiconductor); Grading (engineering); Orthopedic surgery; Nuclear medicine; Surgery; Statistics; Mathematics","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":[0.0009202194,0.0001036493,0.0003485997,0.0001566285,0.00006701452,0.00001624621,0.0001059414,0.0001027795,0.00008559628],"category_scores_gemma":[0.0006740444,0.00006614929,0.000387345,0.0005762097,0.00009391952,0.0001437911,0.0000203576,0.0003570335,0.000001063668],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001350972,"about_ca_system_score_gemma":0.0002459626,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002065367,"about_ca_topic_score_gemma":0.000008542911,"domain_scores_codex":[0.9981625,0.0001240093,0.0008962274,0.0001254535,0.0005725938,0.0001191963],"domain_scores_gemma":[0.9975396,0.00005330678,0.000953893,0.0003395085,0.001011911,0.000101807],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002034021,0.0003036815,0.9611933,0.0004070805,0.00009555806,0.00005047476,0.0001316062,0.00004412501,0.01639145,0.001013662,0.002027416,0.01813822],"study_design_scores_gemma":[0.0006751106,0.0001618832,0.928152,0.000723345,0.0001975956,0.0002812789,0.0004104056,0.00008272712,0.06653178,0.0001142526,0.002611848,0.00005775603],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9919273,0.0003427983,0.002558748,0.003378788,0.0004372577,0.0001871591,0.000004649795,0.00002630731,0.001136941],"genre_scores_gemma":[0.9976802,0.00007169649,0.001641991,0.00008024436,0.0002146608,0.000003675406,0.000008118504,0.00001230914,0.0002870795],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05014033,"threshold_uncertainty_score":0.269749,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0202066040585619,"score_gpt":0.2808383372503219,"score_spread":0.26063173319176,"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."}}