{"id":"W3111012053","doi":"10.2196/22649","title":"Detecting Miscoded Diabetes Diagnosis Codes in Electronic Health Records for Quality Improvement: Temporal Deep Learning Approach","year":2020,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Institute of General Medical Sciences","keywords":"Medicine; Population; Diagnosis code; Gold standard (test); Medical record; Deep learning; Receiver operating characteristic; Artificial intelligence; Machine learning; Diabetes mellitus; Computer science; Artificial neural network; Health care; Test set; Data mining; Internal medicine","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":[],"consensus_categories":[],"category_scores_codex":[0.003222845,0.000251445,0.0005693971,0.0001277635,0.0002710446,0.0001486542,0.001064779,0.0002025995,0.00001753407],"category_scores_gemma":[0.002274401,0.0002327961,0.0001124256,0.0006755103,0.00005757367,0.0004970653,0.0003775108,0.001541707,0.000008583066],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002699758,"about_ca_system_score_gemma":0.0005399323,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002021691,"about_ca_topic_score_gemma":0.0001404781,"domain_scores_codex":[0.9958758,0.0003744816,0.00149273,0.0003240345,0.0008604313,0.001072481],"domain_scores_gemma":[0.9975141,0.0008171241,0.0006613389,0.0003309224,0.00008656534,0.000589942],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00002605641,0.0001417857,0.1281102,0.004547945,0.00003416219,0.000001234393,0.03931241,0.001108403,0.000005675433,0.003072293,0.0003404713,0.8232994],"study_design_scores_gemma":[0.0008973662,0.0009851718,0.001377393,0.0001216524,0.000002268575,0.000001549539,0.001745981,0.9901107,0.00007405587,0.0003240763,0.004087798,0.0002719986],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4745976,0.0003308328,0.5115311,0.01094371,0.0001634587,0.001647652,0.000004927257,0.0005294154,0.0002513404],"genre_scores_gemma":[0.9176154,0.00008182621,0.07384766,0.007487412,0.0001532922,0.0007366904,0.00004706956,0.00002274398,0.000007865353],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9890023,"threshold_uncertainty_score":0.9493151,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03224698330372816,"score_gpt":0.342191091342008,"score_spread":0.3099441080382798,"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."}}