Estimating the causal effect of annual PM2.5 exposure on mortality rates in the Northeastern and mid-Atlantic states

Abstract

Background: Dozens of cohort studies have associated particulate matter smaller than 2.5 μm in diameter (PM2.5) exposure with early deaths, and the Global Burden of Disease identified PM2.5 as the fifth-ranking mortality risk factor in 2015. However, few studies have used causal modeling techniques. We assessed the effect of annual PM2.5 exposure on all-cause mortality rates among the Medicare population in the Northeastern and mid-Atlantic states, using the difference-in-differences approach for causal modeling. Methods: We obtained records of Medicare beneficiaries 65 years of age or more who reside in the Northeastern or mid-Atlantic states from 2000 to 2013 and followed each participant from the year of enrollment to the last year of follow-up. We estimated the causal effect of annual PM2.5 exposure on mortality rates using the difference-in-differences approach in the Poisson survival analysis. We controlled for individual confounders, for spatial differences using dummy variables for each ZIP code and for time trends using a penalized spline of year. Results: We included 112,376,805 person-years from 15,401,064 people, of whom 37.4% died during the study period. The interquartile range (IQR) of the annual PM2.5 concentration was 3 μg/m3, and the mean annual PM2.5 concentration ranged between 6.5 and 14.5 μg/m3 during the study period. An IQR incremental increase in PM2.5 was associated with a 4.04% increase (95% CI = 3.49%, 4.59%) in mortality rates. Conclusions: Assuming no omitted predictors changing differently across ZIP codes over time in correlation with PM2.5, we found a causal effect of PM2.5 on mortality incidence rate.