An exposome approach to estimating causal effects of time-varying air pollution mixtures on CVD incidence

Understanding the causal health effects of environmental exposures is a central aim of exposomics.Ambient air pollution, largely anthropogenic, contributes substantially to global disease burden, with cardiovascular diseases (CVD) accounting for a major share. Using the UK Biobank cohort, we estimated the joint causal effect of time-varying PM2.5 and NO2 exposures on cumulative CVD risk over a 10-year follow-up. Annual exposures were derived from a Europe-wide land-use regression model, providing high-resolution, time-resolved air-pollution estimates for this cohort. We estimated the health impact of a counterfactual proportional decrease in exposure and the projected co-benefits of UK net-zero and emissions-control policies using modified treatment policies. Confounders were selected using a causal structure discovery (CSD) algorithm informed by prior knowledge. Causal effects were estimated using a sequential doubly robust non-parametric estimator accounting for censoring, incorporating ensemble learning, and providing uncertainty quantification. Guided by the CSD, the main model included the assessment-center region as a potential confounder. This may control for spatial confounding but can also reduce the precision by shrinking exposure contrasts or amplify bias from unmeasured confounding. We therefore, conducted a sensitivity analysis excluding the region from the adjustment set. A 25% reduction in PM2.5 and NO2 in 2005-2019 was estimated to reduce 10-year cumulative CVD incidence by -0.62 per 100 (95%CI=(-0.70,-0.55); Relative Risk = 0.90). Based on projected regional exposure reductions under future emissions and net-zero policies, the estimated reduction in incidence reached -1.05 per 100 (95%CI=(-1.12,-1.07); RR = 0.85). Excluding the region from the adjustment set resulted in effect estimates more than twofold larger. This study is the first to integrate data-driven causal structure discovery with a non-parametric, double-robust algorithm to quantify the causal effect of longitudinal policy-relevant interventions. The results reinforce that air pollution is a significant environmental risk factor for CVD and indicate that net-zero policies could yield substantial (>15%) reductions in CVD risk.