{"id":"W4387770265","doi":"10.21203/rs.3.rs-3370784/v1","title":"Inter-rater Reliability of ACS-NSQIP Colorectal Procedure Coding in Canada","year":2023,"lang":"en","type":"preprint","venue":"Research Square","topic":"Colorectal Cancer Screening and Detection","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University; Nova Scotia Health Authority","funders":"","keywords":"Current Procedural Terminology; Medicine; Benchmarking; Calculator; Coding (social sciences); Kappa; Inter-rater reliability; Risk assessment; Cohen's kappa; Statistics; Surgery; Computer science","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":[0.02939675,0.000226437,0.0004103524,0.002386467,0.001236201,0.001273576,0.001152805,0.0002849094,0.0006972014],"category_scores_gemma":[0.09890526,0.0003101075,0.0004404459,0.003254956,0.001348866,0.0003961594,0.001446037,0.0004587393,0.0001473939],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01531186,"about_ca_system_score_gemma":0.02018283,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.8618218,"about_ca_topic_score_gemma":0.8557467,"domain_scores_codex":[0.9605025,0.01166496,0.003684207,0.003680084,0.01883739,0.001630872],"domain_scores_gemma":[0.8728481,0.03413436,0.01342192,0.004451822,0.0727819,0.002361893],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001477499,0.00001956821,0.9713579,0.0001611249,0.000202616,0.00004485851,0.004980499,0.0006028813,0.000336624,0.0002577537,0.001951392,0.01993711],"study_design_scores_gemma":[0.000008356558,0.00003061431,0.9945154,0.00007818943,0.00003732937,0.00003673143,0.001217501,0.001938137,0.0004088996,0.00005970418,0.001649571,0.00001965342],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9853551,0.001196002,0.003562578,0.0007565537,0.00009995099,0.0002915645,0.002681467,0.00007681984,0.005979995],"genre_scores_gemma":[0.9974365,0.0002442026,0.001063218,0.00006687548,0.00001341974,0.0000605102,0.0007131919,0.00001504405,0.0003870814],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9706032,"threshold_uncertainty_score":0.2779838,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07336231149199776,"score_gpt":0.3817661653769532,"score_spread":0.3084038538849554,"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."}}