{"id":"W4399269321","doi":"10.1016/j.amjsurg.2024.115787","title":"Inter-rater reliability of ACS-NSQIP colorectal procedure coding in Canada","year":2024,"lang":"en","type":"article","venue":"The American Journal of Surgery","topic":"Cardiac, Anesthesia and Surgical Outcomes","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University; Nova Scotia Health Authority","funders":"","keywords":"Inter-rater reliability; Coding (social sciences); Reliability (semiconductor); Reliability engineering; Medicine; Computer science; Statistics; Engineering; Mathematics; Rating scale","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02506726,0.0002235487,0.0003678818,0.002219007,0.001289934,0.001146128,0.001077406,0.0002865892,0.0007501974],"category_scores_gemma":[0.09091827,0.0002639261,0.0004251557,0.002394716,0.00144357,0.0003673821,0.001506728,0.000481376,0.000128976],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01369779,"about_ca_system_score_gemma":0.02116158,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.8352603,"about_ca_topic_score_gemma":0.8346653,"domain_scores_codex":[0.9659718,0.009626049,0.003222411,0.002974705,0.01673682,0.001468324],"domain_scores_gemma":[0.8696684,0.03408695,0.01255102,0.003635856,0.07765996,0.002397881],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001406505,0.00001909012,0.9769333,0.0001377917,0.0001746825,0.00004878308,0.004885702,0.000632597,0.0003264886,0.0002164494,0.001540083,0.01494435],"study_design_scores_gemma":[0.000009066608,0.000035193,0.9928238,0.00009290158,0.00004266885,0.00005443463,0.001806342,0.002605626,0.0005679865,0.00009481905,0.001845993,0.00002121089],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9850406,0.0009274444,0.003792716,0.0007009442,0.00007818006,0.0002641553,0.002427228,0.00007224057,0.006696429],"genre_scores_gemma":[0.9976727,0.0001694137,0.001171918,0.00006476667,0.00001064982,0.00005084496,0.0005695984,0.00001213485,0.0002778031],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9863022,"threshold_uncertainty_score":0.3314196,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01428065634177808,"score_gpt":0.2525767622631192,"score_spread":0.2382961059213411,"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."}}