Development and evaluation of cardiovascular disease risk prediction models for patients with type 2 diabetes

Abstract

Abstract To facilitate treatment decisions in people at risk of Cardiovascular Disease (CVD), several risk equations such as the Pooled Cohort Equations and Predicting Risk of Cardiovascular Disease Events (PREVENT) equations have been developed to estimate CVD risk for primary prevention patients. However, it is unclear whether these equations achieve high predictive accuracy and fairness in patients with type 2 diabetes (T2D), and whether a T2D-specific risk equation is needed. Accordingly, we developed a Weibull Accelerated Failure Time (AFT) survival model for predicting the 3-year CVD risk in 23,795 patients with T2D from the All of Us dataset, using sociodemographic information, physical measurements, medication, and CVD history. Among patients without CVD history, our Weibull AFT (vs. PREVENT) achieved a greater C-index (0.646 vs. 0.465), greater Concordance Fractions (0.610–0.674 vs. 0.541–0.600), and comparable Concordance Imparity (0.006 vs. 0.002) across sex and race/ethnicity (0.065 vs. 0.058) subgroups. Our findings highlight the need for a T2D-specific CVD risk equation and demonstrate the value of diverse datasets for developing fair and accurate predictive models.