Prediction of coronary heart disease risk integrating polygenic risk scores and wearables

Abstract

Prediction of coronary heart disease (CHD) risk through standard equations relying on laboratory-based clinical markers has proven challenging and needs advancement. This study aims to derive and cross-validate CHD-risk prediction models based on lifestyle behaviours including wearables and polygenic risk scores (PRS), and compare their performance with the established Pooled Cohort Equations (PCE). This study included 291,151 white British individuals in the UK Biobank. Cox regression was applied to derive the Lifestyle-Based Model (LBM) for CHD-risk prediction incorporating age, sex, body mass index, dietary intake score (0-3; derived from self-reported food types), smoking status (never, previous, current), and physical activity (wearable-device-derived Euclidean Norm Minus One). Weighted PRS for CHD was calculated based on 300 genetic variants. Over a median 13.8-year follow-up, 13,063 CHD incidence cases were ascertained. The C-index (indicative of discrimination) of the LBM and PCE was 0.713 (95% Confidence Interval [CI]: 0.703-0.722) and 0.714 (95% CI: 0.705-0.724). Adding PRS to LBM and PCE increased the C-index to 0.733 (95% CI: 0.724-0.742) and 0.726 (95% CI: 0.716-0.735), respectively. The LBM with and without PRS both demonstrated good calibration, as evidenced by p-values of 0.997 and 0.999, respectively. The addition of PRS to LBM marginally improved calibration, with the slope increasing from 0.981 to 0.983. Integrating PRS resulted in a positive categorical net reclassification improvement (cut-off point: 7.5%) of 4.30% for LBM and 5.04% for PCE. Models incorporating either lifestyle behaviours alone or in combination with genetic traits demonstrated acceptable accuracy for CHD risk prediction.