Development and Validation of a Novel Algorithm for Classifying Diabetes Type in Adolescents and Young Adults Using Electronic Health Record Data

Abstract

OBJECTIVE Accurate classification of diabetes type in early-onset populations is essential for epidemiologic and health services research but remains challenging when using electronic health record (EHR) data alone. Our objective was to develop and validate a diabetes typing algorithm distinguishing type 1 from type 2 diabetes in adolescents and young adults (AYAs). RESEARCH DESIGN AND METHODS We developed and validated an EHR-based algorithm to identify and classify diabetes type among individuals aged 10–30 years enrolled in the Kaiser Permanente Northern California network (2007–2024). Identification of AYAs with diabetes incorporated diagnosis codes, laboratory results, and medications. Diabetes type was classified using a hierarchical algorithm that leveraged pancreatic autoantibody results and longitudinal diagnostic patterns. Algorithm performance was assessed via blinded medical record review in a random subset of 100 AYAs with diabetes, representing 0.14% of the 71,385 AYAs with algorithm-identified diabetes. RESULTS Among 71,385 AYAs with diabetes, the disease was classified as type 1 diabetes in 16.6%, as type 2 diabetes in 75.2%, as other diabetes in 3.0%, and as indeterminate in 5.2%. When validated against blinded medical record review in a subset of patients, sensitivity and specificity were 100% and 98%, respectively, for type 1 diabetes and 98% and 95%, respectively, for type 2 diabetes. Inclusion of pancreatic autoantibody data prevented potential misclassification of type 1 diabetes in 609 patients, representing ∼6% of those with algorithm-classified type 1 diabetes. CONCLUSIONS This validated diabetes typing algorithm supports scalable early-onset diabetes research using EHR data.