Accuracy of the China-PAR and WHO risk models in predicting the ten-year risks of cardiovascular disease in the Chinese population

Abstract

Objective: To externally validate and compare the accuracy of the China-PAR (Prediction for ASCVD Risk in China) model and the 2019 World Health Organization (WHO) cardiovascular disease risk charts for East Asian in predicting a 10-year cardiovascular disease in a general Chinese population. Methods: Participants aged 40-79 years without prior cardiovascular disease at baseline in the CHinese Electronic health Records Research in Yinzhou (CHERRY) were analyzed. The Kaplan-Meier analysis estimated the observed cardiovascular events (including non-fatal myocardial infarction, fatal coronary heart disease, and non-fatal or fatal stroke) rate within ten years. The expected risks were calculated using the WHO risk charts for East Asia (including the laboratory-based and non-laboratory-based models) and the China-PAR model. The expected-observed ratios were calculated to evaluate the overestimation or underestimation of the models in the cohort. Model accuracy was assessed by discrimination C-index, calibration χ2 value, and calibration plots. Results: During a median of 7.26 years of follow-up, 13 301 cardiovascular events were identified among 225 811 participants. The C-index for the China-PAR model, WHO laboratory-based model and WHO non-laboratory-based model were 0.741 (0.735-0.747), 0.747 (0.740-0.753), and 0.739 (0.733-0.746) for men, and 0.782 (0.776-0.788), 0.789 (0.783-0.795), and 0.782 (0.776-0.787) for women, respectively. The WHO laboratory-based model and non-laboratory-based model underestimated the 10-year ASCVD risk by around 15% in women and underestimated by 0.8% and 4.4% in men, respectively. The China-PAR model underestimated the risks by 19.5% and 42.3% for men and women. Conclusions: The China-PAR and WHO models all have pretty good discriminations for 10-year cardiovascular risk assessment in this general Chinese population. However, the accuracy should be improved in the highest-risk groups, suggesting further specific models are still needed for those with the highest risk, such as patients with diabetes or older persons.