Cardiovascular disease risk prediction models for type 2 diabetes: verification in Chinese patients

Abstract

Objective To independently verify and compare the accuracy of 2 cardiovascular risk prediction models, Framingham model and New Zealand Diabetes Cohort Study (DCS) model in Chinese patients with type 2 diabetes mellitus (T2DM). Methods A total of 48 063 T2DM patients aged from 30 to 74 who had medical records but no cardiovascular diseases (CVD) in the Chinese electronic health records research in Yinzhou (CHERRY) were enrolled from January 1, 2010 to December 31, 2018. Kaplan Meier method was used to calculate the actual risk of CVD in following 5 years, and the recalibrated Framingham model and DCS model were adopted to estimate the predicted 5-year risks of the diseases. The performance of the models was assessed in terms of discrimination and calibration degrees, using C-index and Hosmer-Lemeshow Chi-square value respectively. Results Through a median follow-up of 5.9 years, 2 867 new CVD cases were observed among the subjects. However, both the Framingham and the DCS models underestimated the risk of subjects, by 3.6% (Framingham) and 21.1% (DCS) in males, and by 21.0% (Framingham) and 20.9% (DCS) in females, respectively. For males, the DCS model (C-index=0.716, 95%CI: 0.702~0.730) had a better discriminative ability than the Framingham one (C-index=0.662, 95%CI: 0.648~0.677), though the performance of calibration was not ideal in both models (P < 0.05). For females, C-index was 0.743 in DCS (95%CI: 0.730~0.756) and 0.686 in Framingham models (95%CI: 0.673-0.700), both showing poor performance on calibration (P < 0.05). Conclusion Both Framingham model and DCS model are confirmed inapplicable to CVD risk prediction in Chinese population with T2DM.

Keywords

• type 2 diabetes mellitus
• cardiovascular diseases
• risk prediction
• cohort study