{"id":"W4411941372","doi":"10.1097/wnp.0000000000001185","title":"Inter-Rater Reliability of EEG-Based Encephalopathy Grading","year":2025,"lang":"en","type":"article","venue":"Journal of Clinical Neurophysiology","topic":"Epilepsy research and treatment","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Manitoba","funders":"","keywords":"Confusion; Electroencephalography; Kappa; Inter-rater reliability; Correlation; Gold standard (test); Grading (engineering); Psychology; Reliability (semiconductor); Cohen's kappa; Medicine; Physical therapy; Audiology; Physical medicine and rehabilitation; Statistics; Internal medicine; Rating scale; Psychiatry; Mathematics; Developmental psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007020148,0.0001034429,0.0008598688,0.0001517295,0.00002299236,0.000003865343,0.0001617822,0.0001025628,0.00008136859],"category_scores_gemma":[0.004156298,0.00006437474,0.000507566,0.0001776837,0.000338597,0.00003993603,0.00005518064,0.0006440029,0.00001110689],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000227692,"about_ca_system_score_gemma":0.0003994718,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000005445419,"about_ca_topic_score_gemma":5.288107e-7,"domain_scores_codex":[0.9975256,0.0005494569,0.00136215,0.0001896301,0.0001845827,0.0001886349],"domain_scores_gemma":[0.9969047,0.001737597,0.0004154859,0.0002989956,0.0004664076,0.0001767927],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.006161664,0.002092546,0.9436361,0.0001710676,0.0001601544,0.0005253637,0.00001469165,0.00004434347,0.03129146,0.0001491853,0.000967472,0.01478594],"study_design_scores_gemma":[0.003408413,0.008678056,0.9800533,0.0002022877,0.00011156,0.00001679224,0.00001069872,0.0002911777,0.003534872,0.001652718,0.001997171,0.00004297108],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9949253,0.00006722709,0.0003985752,0.003281174,0.0006762255,0.0001215975,0.000002496175,0.000005984932,0.0005214201],"genre_scores_gemma":[0.9973633,0.0001702,0.0008366691,0.001336422,0.0001558415,0.000001745357,0.000001495368,0.000006797015,0.0001275276],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03641718,"threshold_uncertainty_score":0.4975777,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04887345618865522,"score_gpt":0.4262930050511849,"score_spread":0.3774195488625297,"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."}}