170-LB: Identifying Atypical Diabetes Using i2b2 Algorithms in Electronic Health Records

Abstract

Investigating atypical forms of diabetes (other than types 1 or 2, monogenic or secondary) is the focus of research in the Rare and Atypical Diabetes Network (RADIANT) . It is difficult to identify persons with atypical diabetes using standard electronic health record (EHR) queries because there are no specific ICD codes for these phenotypes and wide range of clinical features. We aimed to develop an algorithm to identify individuals with searchable characteristics of atypical diabetes within the Baylor Medicine EHR using Informatics for Integrating Biology at the Bedside (i2b2) . I2b2 enables integration, standardization, and analysis of heterogenous quantitative and qualitative data from healthcare and research databases. Our i2b2 algorithm intended to look for patients coded as type 2 diabetes (T2D) but were very lean and lacked clinical features suggestive of systemic insulin resistance. We included new patient visits with ICD codes for T2D in the past year but excluded those with BMI >20 kg/m2 and triglycerides >100 mg/dL, or ICD codes for hypertension and hyperlipidemia. Next, we excluded patients with secondary causes of diabetes (cystic fibrosis, acromegaly, Cushing's syndrome, chronic/acute pancreatitis or presence of islet autoantibodies (GAD65Ab, IA2Ab) . The algorithm identified 133 persons from a total EHR population of 3,685,797. Manual chart review confirmed that 18 patients (13.5%) met RADIANT criteria for atypical diabetes, while information was missing in the EHR to fully classify 8 patients (6%) and 1 patient was deceased (0.8%) . Those who fully met criteria for atypical diabetes had the following characteristics: 72% female, median BMI at diagnosis 19.8 kg/m2, median age at diagnosis 29 y, 75% with history of diabetic ketoacidosis and 19.5% currently on insulin. Algorithms built on data platforms such as i2b2 may be helpful in identifying patients who should be evaluated for atypical diabetes within standard EHR databases. They could optimize recruitment for research, identify disease trends and permit improved classification of diabetes subgroups. Disclosure N. Kikani: None. G. Montes: None. R. Gaba: None. G. Liao: None. M. Tosur: Advisory Panel; Provention Bio, Inc. M. J. Redondo: Advisory Panel; Provention Bio, Inc. A. Balasubramanyam: None. Funding The RADIANT Study is funded by U54 DK118638 and U54 DK118612 from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK)