2229

Background: Multicollinearity is a common challenge in environmental epidemiology research. However, existing methods are mostly difficult to implement or produce results that are not readily interpretable. Objectives: Our aim is to use sparse principal component analysis (SPCA) to address the multicollinearity issue and improve interpretability of the prospective association between environmental exposures and risk of incident cardiovascular disease (CVD) using data from a large population-based cohort. Methods: We derived principal components using SPCA techniques on 24 different environmental exposures from the UK Biobank (405,583 participants) and built Cox regression models to estimate the hazard ratio (HR) for incident cardiovascular disease (CVD) associated with each component. We then back-calculated the hazard ratio (HR) to each individual environmental exposure from the coefficients derived from the SPCA Cox model and loading scores using a commonly adopted method for instrumental variable analysis. Results: During an average follow-up of 9.7 years, 25,655 participants developed CVD. Components characterised by higher exposure to traffic intensity on the nearest major road or traffic-related air pollutants (NO₂, NO_X and PM_2.5) were significantly associated with incident CVD risk, as identified by the SPCA Cox model. For each 1-SD (1.05 ug/m³) higher PM_2.5, the HR was 1.09 (95% confidence interval 1.06-1.12); risk estimates were similar in magnitude for NO₂ or NOx. Results for other exposures were variable and inconsistent. Conclusion: Using SPCA, our study consistently identified traffic-related air pollution as an important risk factor of CVD. SPCA offers a potential solution to address multicollinearity and the advantage for to identify relatively more important environmental exposures in relation to incident CVD outcome.