Related Experiment Video
Updated: Aug 11, 2026
An R-Based Landscape Validation of a Competing Risk Model
Published on: September 16, 2022
Validation and Comparison of 10-Year Cardiovascular Risk Prediction Models in Two Chinese Prospective Cohorts: The
Xikang Fan1, Jiaxi Zhou2, Mengxia Li1
1Jiangsu Provincial Center for Disease Control and Prevention, Nanjing, China.
Background:
To evaluate and recalibrate commonly used 10-year cardiovascular disease (CVD) risk prediction models in two Chinese cohorts: the general population and adults with type 2 diabetes.
Methods:
We analysed data from the Wuzhong subcohort of the China Kadoorie Biobank (CKB) and the Jiangsu Biobank for the Prevention and Control of Diabetes (JBPCD). After excluding participants with baseline CVD, baseline diabetes in CKB or missing predictors, 42 937 CKB participants and 14 125 JBPCD participants were included. We evaluated CKB-CVD, Globorisk, Framingham General CVD and WHO 2019 in both cohorts, and SCORE2-Diabetes in JBPCD. Performance of original and recalibrated models was assessed by 10-year inverse probability of censoring weighting (IPCW) C-statistic, calibration measures including the expected-to-observed (E/O) event ratio and Brier score. In JBPCD, overlap in high-risk classification was examined using a 20% predicted 10-year risk threshold.
Results:
During a median follow-up of 11.96 years in the general population cohort and 10.03 years in the type 2 diabetes cohort, 3740 and 2794 ischemic heart disease or stroke events occurred, with 10-year risks of 6.04% and 19.72%, respectively. Discrimination was higher in the general population cohort than in the type 2 diabetes cohort (C-statistic 0.723-0.778 vs. 0.584-0.629). Original models were poorly calibrated, with systematic overprediction in the general population cohort (E/O 1.47-3.78) and heterogeneous miscalibration in the type 2 diabetes cohort (E/O 0.32-2.70). Recalibration markedly improved calibration and reduced Brier scores, with little change in discrimination. In the type 2 diabetes cohort, recalibrated CKB-CVD classified 48.4% as high risk, with an observed 10-year risk of 25.7%.
Conclusion:
Widely used 10-year CVD risk models showed acceptable or modest discrimination but substantial miscalibration in Chinese populations, especially among adults with type 2 diabetes. Recalibration is essential before threshold-based clinical or public-health use.