{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2533421,0.001648524,0.001235718,0.005998754,0.002168995,0.002442442,0.00504613,0.001812239,0.001303232],"category_scores_gemma":[0.3764794,0.001281353,0.001931202,0.003804676,0.003789785,0.002217102,0.003812567,0.001132833,0.0005716471],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004740892,"about_ca_system_score_gemma":0.008705564,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01100043,"about_ca_topic_score_gemma":0.01229051,"domain_scores_codex":[0.7068037,0.2003709,0.02855935,0.01602785,0.04557041,0.002667807],"domain_scores_gemma":[0.42368,0.2731023,0.06656221,0.09893347,0.1357356,0.001986306],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002025882,0.003001722,0.8802459,0.002506975,0.0009987602,0.000640042,0.0173688,0.001148907,0.003023274,0.001251081,0.00434998,0.08343859],"study_design_scores_gemma":[0.002176984,0.00786506,0.9269818,0.002837691,0.001653749,0.002431428,0.005129674,0.01279081,0.01250996,0.0008972691,0.02447001,0.0002556205],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9240676,0.002066727,0.04060398,0.0004504514,0.0002937868,0.02577983,0.00151862,0.0001897311,0.005029329],"genre_scores_gemma":[0.9198817,0.0005452418,0.05914389,0.0009937801,0.0002065807,0.01488611,0.003591674,0.00009600704,0.0006550307],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.746658,"threshold_uncertainty_score":0.9207622,"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."}}