{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0051572,0.0007973608,0.000757643,0.002667683,0.0005002252,0.001390982,0.00175161,0.0009813323,0.0006898291],"category_scores_gemma":[0.01821775,0.0004169626,0.0007863685,0.002470293,0.0005551719,0.001582215,0.001966618,0.002183168,0.0001906573],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002847037,"about_ca_system_score_gemma":0.002985263,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03689945,"about_ca_topic_score_gemma":0.04030555,"domain_scores_codex":[0.9974189,0.0009592681,0.0002759256,0.0005872431,0.0004651784,0.0002934391],"domain_scores_gemma":[0.9889182,0.005926346,0.00227487,0.0007948446,0.001758944,0.0003267358],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0005575855,0.001209572,0.3140462,0.0003127716,0.0005670609,0.0004224847,0.0006492457,0.2892312,0.001379095,0.002684098,0.01445798,0.3744827],"study_design_scores_gemma":[0.00001897916,0.0000403734,0.008698509,0.00005161852,0.00004255582,0.00003254831,0.00007048979,0.9872914,0.0004201351,0.00279112,0.0005300253,0.00001211707],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6570574,0.002291122,0.3218529,0.00809007,0.0002387959,0.0003763956,0.005093408,0.00197364,0.003026163],"genre_scores_gemma":[0.9505594,0.0003437765,0.04377593,0.0007076846,0.0001188076,0.0001201322,0.003629241,0.0000275915,0.0007173771],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9948428,"threshold_uncertainty_score":0.07336932,"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."}}