{"id":"W4309157948","doi":"10.2196/39231","title":"Accuracy of COVID-19–Like Illness Diagnoses in Electronic Health Record Data: Retrospective Cohort Study","year":2022,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Centers for Disease Control and Prevention","keywords":"Medicine; Emergency department; Retrospective cohort study; Diagnosis code; Medical diagnosis; Cohort study; Cohort; Observational study; Electronic health record; Epidemiology; Emergency medicine; MEDLINE; Pediatrics; Health care; Internal medicine; Psychiatry; Population; Pathology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.01688896,0.0002001501,0.0005670899,0.0009955307,0.001264867,0.0001098677,0.004322887,0.0000441395,0.0001688529],"category_scores_gemma":[0.002451622,0.0001968278,0.00004659205,0.004032409,0.0001532601,0.001289478,0.005935343,0.00288417,0.00001973295],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004072916,"about_ca_system_score_gemma":0.004149891,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01614565,"about_ca_topic_score_gemma":0.003218445,"domain_scores_codex":[0.9873641,0.00693468,0.0008184859,0.0009899553,0.002589077,0.00130368],"domain_scores_gemma":[0.9938806,0.002739993,0.0004597976,0.002242581,0.0003625294,0.0003144839],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001049667,0.001222246,0.8894778,0.0002854562,0.00004431768,0.00004888186,0.06330793,0.0003473609,0.000001548241,0.009763063,0.01149234,0.02390406],"study_design_scores_gemma":[0.00137094,0.005683701,0.9120135,0.00003896585,0.000001743988,0.00002911802,0.0155167,0.02943631,0.000005485239,0.004425128,0.03113788,0.0003405031],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9713833,0.00102132,0.006451585,0.01172315,0.0006025614,0.007685501,0.0002122475,0.0002310062,0.0006892703],"genre_scores_gemma":[0.9972915,0.000241936,0.0002107468,0.0002959209,0.00003581648,0.001752003,0.0000740619,0.00002042254,0.00007758864],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04779123,"threshold_uncertainty_score":0.9997503,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08834359484281383,"score_gpt":0.4843026038609434,"score_spread":0.3959590090181295,"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."}}