{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4494376,0.001963734,0.002092944,0.00680213,0.003620515,0.002718403,0.00337755,0.001660893,0.00311711],"category_scores_gemma":[0.5453711,0.002009993,0.002553984,0.004614541,0.003923964,0.003040939,0.005724682,0.002076327,0.001699946],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003277385,"about_ca_system_score_gemma":0.009155004,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00165151,"about_ca_topic_score_gemma":0.00309549,"domain_scores_codex":[0.31782,0.5350838,0.06897607,0.01997205,0.05563333,0.002514652],"domain_scores_gemma":[0.2300593,0.4101933,0.1057243,0.09298165,0.1584456,0.002595827],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.006953705,0.003368411,0.2574466,0.0126812,0.002621296,0.0005600724,0.04318593,0.01061905,0.02330761,0.01655531,0.01793925,0.6047615],"study_design_scores_gemma":[0.009335961,0.02201714,0.598985,0.01045048,0.002349256,0.002809879,0.01055469,0.09721164,0.09594142,0.04278404,0.1058024,0.001758084],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.101973,0.001032896,0.8176724,0.001044604,0.000638798,0.06527662,0.0006690422,0.00163973,0.01005278],"genre_scores_gemma":[0.2864311,0.0002361743,0.5925184,0.0006057493,0.0003719126,0.1182982,0.0003944308,0.0002727814,0.0008712043],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5505624,"threshold_uncertainty_score":0.6789415,"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."}}