A multi-pollutant approach to estimating causal effects of air pollution mixtures on overall mortality in a large, prospective cohort

Abstract

BACKGROUND: Several studies have confirmed associations between air pollution and overall mortality, but it is unclear to what extent these associations reflect causal relationships. Moreover, few studies to our knowledge have accounted for complex mixtures of air pollution. In this study, we evaluate the causal effects of a mixture of air pollutants on overall mortality in a large, prospective cohort of Dutch individuals. METHODS: We evaluated 86,882 individuals from the LIFEWORK study, assessing overall mortality between 2013 and 2017 through national registry linkage. We predicted outdoor concentration of five air pollutants (PM2.5, PM10, NO2, PM2.5 absorbance, and oxidative potential) with land-use regression. We used logistic regression and mixture modeling (weighted quantile sum and boosted regression tree models) to identify potential confounders, assess pollutants' relevance in the mixture-outcome association, and investigate interactions and nonlinearities. Based on these results, we built a multivariate generalized propensity score model to estimate the causal effects of pollutant mixtures. RESULTS: Regression model results were influenced by multicollinearity. Weighted quantile sum and boosted regression tree models indicated that all components contributed to a positive linear association with the outcome, with PM2.5 being the most relevant contributor. In the multivariate propensity score model, PM2.5 (OR=1.18, 95% CI: 1.08-1.29) and PM10 (OR=1.02, 95% CI: 0.91-1.14) were associated with increased odds of mortality per interquartile range increase. CONCLUSION: Using novel methods for causal inference and mixture modeling in a large prospective cohort, this study strengthened the causal interpretation of air pollution effects on overall mortality, emphasizing the primary role of PM2.5 within the pollutant mixture.