Estimating Causal Associations Of Low PM2.5 On Daily Deaths In Boston

Abstract

Background: Many time series studies have reported associations between daily PM2.5 and daily deaths, but they have been associational studies that did not use formal causal modeling. Methods: Based on a potential outcome approach, we used two causal modeling methods with different assumptions and strengths to address the question of whether there was a causal association between daily PM2.5 in Boston and daily deaths (2004-2009) when levels never exceeded the EPA standard. We used an instrumental variable approach, using back trajectories as instruments for variations in PM2.5 concentrations uncorrelated with any other predictors of death. We also used propensity score as an alternative causal modeling analysis. The former protects against confounding by measured and unmeasured confounders, and is based on the assumption that we have a valid instrument. The latter protects against confounding by all measured covariates, provides valid estimates in the case of effect modification, and is based on the assumption of no unmeasured confounders. Results: We found a significant causal association of PM2.5 on daily deaths in Boston using instrumental variable (0.53%, 95% CI: 0.09%, 0.97%) and propensity score (0.50%, 95% CI: 0.20%, 0.80%). We also failed to reject the null association between deaths and exposure after the deaths (P=0.93). Conclusion: Given these results, prior studies, and extensive toxicological support, the association between PM2.5 and daily deaths is almost certainly causal.