211-LB: Developing "Gold Standard" Diagnoses for Type 1 and Type 2 Diabetes in Adults from Electronic Health Record Data

Abstract

Type 1 (T1D) and type 2 (T2D) diabetes are distinct diseases, but most federal surveys used to monitor diabetes prevalence have not included questions on diabetes type. To design and validate survey questions that will distinguish between T1D and T2D in adults, we developed a survey sample in whom we first determined a "gold standard" for diabetes type from review of electronic health record (EHR) data. From >50,000 adult patients in the UNC Health Care System with diabetes, we assembled a sample of 2479 patients using a "strawman" definition based on diabetes-related diagnostic codes, laboratory results, and drugs. We stratified the sample using age, sex, race/ethnicity and crude diabetes type, oversampling T1D. For visits between 10/2014 and 9/2017, we collected clinical and demographic data including body mass index, diagnosis codes, medications, family history, and laboratory tests. We also abstracted chart notes, using trained personnel, to find additional information including diabetes duration and historical treatments. We created two models to determine diabetes type. The first was a decision tree, in which higher nodes sorted straightforward cases; lower nodes included more specific factors for those not categorized in higher nodes. The second was a weighted model, in which factors drawn from clinical or epidemiologic literature weighed weakly, moderately, or strongly in favor of or against each type of diabetes. The two models were 89% congruent for T1D and 96% congruent for T2D. An endocrinologist reviewed and determined type for the 10% of patients for which results of the two models were not congruent or for which neither was able to determine type. Overall, 70% of our sample has T2D, 27% has T1D, and 3% has indeterminate or other diabetes. We will use this sample to validate survey questions to determine type of diabetes. Such "gold standard" determinations may support validation of new and improved methods for measuring diabetes type among the adult population. Disclosure M. Kirkman: Research Support; Self; Bayer AG, Novo Nordisk A/S, Theracos, Inc. J.G. Nooney: None. S.R. Benoit: None. K.A. Bergamo: None. K.M. Bullard: None. J.R. Campione: None. R. Mardon: None. E. Pfaff: None. D.B. Rolka: None. S. Saydah: None. Funding Centers for Disease Control and Prevention