{"id":"W2772720671","doi":"10.9778/cmajo.20170077","title":"Manual review of electronic medical records as a reference standard for case definition development: a validation study","year":2017,"lang":"en","type":"article","venue":"CMAJ Open","topic":"Chronic Disease Management Strategies","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary; University of Alberta","funders":"","keywords":"Medicine; Medical record; Confidence interval; Gold standard (test); Medical diagnosis; Depression (economics); Chart; Internal medicine; Pathology; Statistics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001412143,0.0001217831,0.0003725022,0.00004549581,0.0002255104,0.0001003412,0.0003930909,0.00004305322,0.001567246],"category_scores_gemma":[0.00104056,0.0001059953,0.00004346914,0.00005063072,0.00005385537,0.000257582,0.0003152465,0.0001057834,0.00003661609],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001882548,"about_ca_system_score_gemma":0.001615412,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004670063,"about_ca_topic_score_gemma":0.000818459,"domain_scores_codex":[0.9986091,0.00005582297,0.0003938656,0.0002794554,0.0004465347,0.000215183],"domain_scores_gemma":[0.9987996,0.00005026693,0.0002769076,0.000577856,0.0001883003,0.0001071072],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.007274103,0.004445974,0.02239361,0.05544667,0.002518355,0.003732517,0.002512538,6.487089e-7,0.00003481822,0.09364228,0.06639859,0.7415999],"study_design_scores_gemma":[0.0708757,0.01962771,0.08017936,0.08997894,0.006716679,0.002489132,0.0104218,0.0001268639,0.002519825,0.02945637,0.6851929,0.002414756],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8190913,0.002352208,0.0007895277,0.003218565,0.0001920306,0.01130771,0.00005997151,0.00006355057,0.1629251],"genre_scores_gemma":[0.995285,0.001635915,0.0007987033,0.0003551427,0.00004942553,0.0006739832,0.00024273,0.00001850507,0.0009406301],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7391852,"threshold_uncertainty_score":0.9993455,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1488989919725432,"score_gpt":0.4507671598376044,"score_spread":0.3018681678650612,"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."}}