Aims: Identifying individuals at the highest risk of progression to type 2 diabetes (T2D) using clinical and social determinants of health (SDoH) measures will help prioritize prevention efforts. We aimed to investigate model performance after adding SDoH to a previously validated cardiometabolic disease staging diabetes risk prediction model.
Materials and methods: We developed a Bayesian predictive model using data [clinical factors: fasting glucose, blood pressure, body mass index, high-density lipoprotein cholesterol, triglycerides; individual SDoH: income, education, health insurance status, relationship status, self-reported stress and neighbourhood SDoH: census-tract level social vulnerability index] from the REasons for Geographic And Racial Differences in Stroke (REGARDS) study to predict T2D with external validation using the Coronary Artery Risk Development in Young Adults (CARDIA) study.
Results: The analysis included 9907 REGARDS participants without T2D at baseline [mean age 63 years (SD 8.5), 54% female, 33% non-Hispanic Black] who completed a follow-up visit 10 years later. N = 1268 (12.8%) developed T2D. Adding SDoH to the clinical model modestly improved performance [Area Under the Curve: 0.802 vs. 0.804, p = 0.01]. Calibration plots indicated that the clinical model underpredicted risk in disadvantaged SDoH subgroups, whereas the clinical plus SDoH model improved prediction accuracy in subgroups. Classification tables revealed that the clinical plus SDoH model accurately reclassified individuals categorized as borderline risk in a clinical-only model.
Conclusion: Including SDoH in T2D risk prediction and stratification at the population level may aid in better classifying T2D risk among vulnerable populations, which has important implications for screening strategies.
Keywords: cardiometabolic disease; risk prediction; risk stratification; social determinants of health; type 2 diabetes.
© 2025 The Author(s). Diabetes, Obesity and Metabolism published by John Wiley & Sons Ltd.