The EHR-Driven Dysglycemia Risk Score (D-RISK) Predicts Future Development of Type 2 Diabetes
Abrar F. Mamun, Molly McGuire, Vincent Merrill, Song Zhang, Noel O. Santini, Brett Moran, Luigi Meneghini, Ildiko Lingvay, Ethan A. Halm, Michael E. BowenOBJECTIVE
To evaluate whether the electronic health record (EHR)-derived Dysglycemia Risk Score (D-RISK), previously validated for detecting prevalent, undiagnosed diabetes and prediabetes, predicts future development of type 2 diabetes over 5 years.
RESEARCH DESIGN AND METHODS
We used baseline data from a cross-sectional diabetes screening study (2015–2017) to predict future development of diabetes using Cox proportional hazards models. Adults without diabetes were included and followed using EHR data through October 2024. D-RISK (age, race and ethnicity, BMI, hypertension, and random blood glucose) was calculated using baseline EHR data. Incident type 2 diabetes was identified using a validated computable phenotype requiring two or more criteria (HbA1c ≥6.5%, diagnostic codes, or glucose-lowering prescriptions). Predictive performance of D-RISK was compared with U.S. Preventive Services Task Force (USPSTF) and American Diabetes Association (ADA) guidelines and the ADA Risk Test.
RESULTS
Among 502 participants (mean age 47.5 years, 69.9% female, 68.5% Hispanic), 12% developed type 2 diabetes over a mean follow-up of 5.8 years. The area under curve for D-RISK was 0.67 (95% CI 0.53–0.80) to predict incident type 2 diabetes. A D-RISK score ≥10 predicted a 7.8% 5-year risk of developing type 2 diabetes (hazard ratio [HR] 2.56, 95% CI 1.06–6.17). Performance was similar to the ADA Risk Test (≥5 points) (HR 2.32, 95% CI 1.01–5.30) and superior to the USPSTF (HR 1.6, 95% CI 0.55–4.68) and ADA (HR 0.60, 95% CI 0.08–4.44) screening guidelines.
CONCLUSIONS
D-RISK predicts future type 2 diabetes using routinely available EHR data. Its ability to both detect prevalent dysglycemia and forecast incident type 2 diabetes may support system-level diabetes risk assessment and screening.