{"id":"W2099954138","doi":"10.1016/j.ygyno.2005.07.040","title":"Kappa statistics to measure interrater and intrarater agreement for 1790 cervical biopsy specimens among twelve pathologists: Qualitative histopathologic analysis and methodologic issues","year":2005,"lang":"en","type":"article","venue":"Gynecologic Oncology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":90,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia","funders":"National Cancer Institute","keywords":"Medicine; Biopsy; Inter-rater reliability; Cohen's kappa; Cervical cancer; Radiology; Kappa; Family medicine; Cancer; Statistics; Internal medicine","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1495038,0.0008187036,0.001833516,0.008298258,0.005271873,0.002904084,0.002677912,0.001076376,0.001786564],"category_scores_gemma":[0.3544491,0.001212884,0.00258565,0.005299124,0.00317973,0.004034961,0.003602822,0.002238778,0.0005891897],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002996556,"about_ca_system_score_gemma":0.004290004,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006174028,"about_ca_topic_score_gemma":0.01757649,"domain_scores_codex":[0.8283029,0.1018141,0.02900114,0.006852168,0.03211521,0.001914398],"domain_scores_gemma":[0.3716541,0.5093099,0.01508072,0.02190541,0.08031054,0.001739308],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002769144,0.0004261956,0.601936,0.003629873,0.005316655,0.0004953324,0.1124669,0.006046343,0.008468976,0.02441374,0.01041317,0.2236177],"study_design_scores_gemma":[0.0005420201,0.00311548,0.7053232,0.002625778,0.003877252,0.00365687,0.1041653,0.0749438,0.02466912,0.04622641,0.02983876,0.001016136],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.725691,0.003414091,0.2400243,0.001061573,0.0009635572,0.002286069,0.001855304,0.0007466816,0.02395756],"genre_scores_gemma":[0.9000795,0.0003503212,0.09588832,0.0001164311,0.00009695974,0.001411562,0.0004019506,0.0002657988,0.001389025],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8504962,"threshold_uncertainty_score":0.7906609,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2655803757781627,"score_gpt":0.4873603395796752,"score_spread":0.2217799638015124,"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."}}