Cardiovascular disease risk prediction in multi-ethnic Asian populations: evidence from two population-based cohorts in Singapore

TL;DR

Researchers compared and recalibrated cardiovascular disease risk models for Asian populations in Singapore to improve accuracy.

Contribution

The study recalibrated and validated CVD risk models for multi-ethnic Asian populations using local data.

Findings

The recalibrated PCE-W model outperformed the original in predicting CVD risk in Singaporean ethnic groups.

The SG-FRS-2023 model showed good calibration but overestimated risk for some ethnic groups.

Local recalibration improved model performance compared to the SCORE2 Asia–Pacific model.

Abstract

The rising burden of cardiovascular diseases (CVD) in Asia requires risk assessment tools tailored to Asian populations. Therefore, we recalibrated the ACC/AHA Pooled Cohort Equations for non-Hispanic Whites (PCE-W) and compared its performance in predicting 10-year CVD risk with two other established CVD prediction models that have been recently recalibrated for Asian populations.

We used data from the Singapore Multi-Ethnic Cohort (MEC1) and the Singapore Epidemiology of Eye Diseases (SEED) cohort comprising ethnic Chinese, Indian, and Malay participants. The PCE-W was recalibrated using data from MEC1, externally validated in the SEED cohort, and compared against the Singapore-modified Framingham Risk Score (SG-FRS-2023) and the SCORE2 Asia–Pacific model using the concordance index (C-index). Calibration was assessed using the calibration-in-the-large method, the calibration slope, and a goodness-of-fit test.

All three models demonstrated possibly helpful to clearly useful discrimination in MEC1 and SEED, with overall C-indices ranging from 0.728 to 0.811. The recalibrated PCE-W outperformed the original PCE-W in MEC1 and SEED, although some misestimations remained among Chinese men and women and Malay women (calibration-in-the-large ranged from −0.479 to 0.260). The SG-FRS-2023 displayed generally satisfactory calibration across both MEC1 and SEED but tended to overestimate risk in Chinese (calibration-in-the-large −0.671) and Indian men (calibration-in-the-large −0.214) in the SEED cohort. The SCORE2 Asia–Pacific model performed satisfactorily among Indians but overestimated risk in Chinese (calibration-in-the-large ranged from −0.570 to −1.185) and showed poor model fit in Malays.

The recalibrated PCE-W, SG-FRS-2023, and SCORE2 Asia–Pacific model demonstrated possibly helpful to clearly useful discrimination across two multi-ethnic cohorts in Singapore. In terms of calibration, the recalibrated PCE-W and SG-FRS-2023, both recalibrated using local data, performed better than the SCORE2 Asia–Pacific model. Our study supports the use of the established CVD prediction models in Asian populations following appropriate local recalibration.

This work was supported by the Singapore Ministry of Health’s National Medical Research Council and the Singapore 10.13039/501100012415Biomedical Research Council.

Figures