{"id":"W2107006322","doi":"10.1370/afm.1195","title":"Methods to Achieve High Interrater Reliability in Data Collection From Primary Care Medical Records","year":2011,"lang":"en","type":"article","venue":"The Annals of Family Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":83,"is_retracted":false,"has_abstract":true,"ca_institutions":"Élisabeth Bruyère Hospital; Bruyère; University of Ottawa","funders":"","keywords":"Medicine; Inter-rater reliability; Reliability (semiconductor); Quality assurance; Chart; Confidence interval; Audit; Statistics; Sample size determination; Data collection; Control chart; Quality (philosophy); Sample (material); Quality management; Statistic; Medical physics; Operations management; Computer science; Mathematics","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03377779,0.0001919913,0.0006673181,0.000233398,0.0000861706,0.00001622972,0.003048468,0.0001331717,0.001460184],"category_scores_gemma":[0.01376385,0.0000971556,0.0000697654,0.001058348,0.0004213185,0.0002245524,0.000875101,0.0003745536,0.00004197165],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006376385,"about_ca_system_score_gemma":0.0001830394,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009034255,"about_ca_topic_score_gemma":0.001082433,"domain_scores_codex":[0.9923752,0.00239038,0.001500649,0.0007730561,0.002670198,0.0002905233],"domain_scores_gemma":[0.993553,0.002528254,0.0002948843,0.002709132,0.0007049104,0.0002098185],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.004025218,0.0006367561,0.04423352,0.00007976044,0.0001085009,0.00001515597,0.04014425,0.0000318631,0.005974001,0.0001152385,0.3169057,0.5877301],"study_design_scores_gemma":[0.00120859,0.00158491,0.8841581,0.0006690299,0.0000490107,0.000001419501,0.01228859,0.000493831,0.00280815,0.06336588,0.03311352,0.0002590028],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9377157,0.001027701,0.01361397,0.03347982,0.002000295,0.0006954911,0.00005253811,0.00002450557,0.01138996],"genre_scores_gemma":[0.9692243,0.0004519374,0.01807741,0.01138403,0.0003401645,0.00003537812,0.00007191067,0.00001453107,0.0004003511],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8399245,"threshold_uncertainty_score":0.9994526,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7358387871238369,"score_gpt":0.5387267113294011,"score_spread":0.1971120757944358,"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."}}