{"id":"W4408450563","doi":"10.1227/neu.0000000000003360_1247","title":"1247 Expert Surgeon-Level OSATS Rating Using Machine Learning for Surgical Performance Assessment","year":2025,"lang":"en","type":"article","venue":"Neurosurgery","topic":"Cardiac, Anesthesia and Surgical Outcomes","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medicine; Rating system; Medical physics; Physical therapy; Surgery","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00666451,0.0005468923,0.0004034115,0.001723329,0.0002280179,0.0008727072,0.0004854604,0.0004065634,0.003212257],"category_scores_gemma":[0.02796972,0.0001305737,0.0003760678,0.0005908695,0.000333105,0.0005645754,0.0007665196,0.0003833252,0.001071027],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004944955,"about_ca_system_score_gemma":0.0005152684,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001444808,"about_ca_topic_score_gemma":0.002800888,"domain_scores_codex":[0.9956833,0.001715986,0.0005300838,0.0006519461,0.001287323,0.0001313454],"domain_scores_gemma":[0.9780955,0.01078131,0.003523374,0.0015857,0.005377188,0.0006369934],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001011814,0.0004450513,0.5057212,0.0003097961,0.0002800216,0.0001224839,0.0003798896,0.01915487,0.006899607,0.0002421535,0.005499837,0.4599333],"study_design_scores_gemma":[0.00008428335,0.001411763,0.5339708,0.0002014985,0.000135814,0.000598253,0.0003976496,0.4365849,0.02070367,0.001487124,0.004289675,0.000134604],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8967749,0.0004860598,0.09114575,0.000287136,0.0001306608,0.0005489064,0.001717187,0.0012444,0.007664923],"genre_scores_gemma":[0.9739327,0.00009185811,0.02398706,0.00004900314,0.00002878793,0.0001646486,0.0008258767,0.00003181606,0.0008883369],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00666451,"threshold_uncertainty_score":0.03524572,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05832050767327143,"score_gpt":0.3448813249118499,"score_spread":0.2865608172385785,"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."}}