Adjusting for Long-Term Temporal Trends in Incidence Rates of All-Cause Mortality Associated with 1-year Average PM2.5 Concentrations: An Analysis of over 20 Million Medicare Beneficiaries between 2000-2012

Abstract

Introduction. Associations between chronic PM2.5 and mortality have been shown, although associations have varied. This variation may reflect confounding by long-term trends in PM2.5 and mortality. Methods. We assessed the impact of long-term time trends on the PM2.5-mortality association between 2000-2012. We did so using exposure measures based on monthly PM2.5 from 798 monitoring sites: (1) PM2.5 at each site (PMct) (2) PM2.5 centered by its mean site level, and (3) (γct), a new measure adjusting for annual PM2.5 trends. We used age-stratified log-linear Poisson models to assess the association between each PM2.5 measure and rate of mortality. We examined confounding by comparing PM2.5-associated rate ratios for data subsets based on every combination of follow-up period and baseline year. We also decomposed each PM2.5 measure into its local and global components and compared the difference in mortality rate ratios. Results We found statistically significant, positive associations between PM2.5 exposures and mortality. Centered PM2.5-associated mortality rate ratios were highest when data for the 13-year study period was examined, with a 5.5% increase in mortality rate per interquartile range (IQR), and decreased with shorter follow-up periods. When we adjusted for long-term PM2.5 time trends, we found lower rate ratios corresponding to a significant 2.7% increase in mortality rate per IQR increase in γct. Mortality rates associated with γct did not vary by follow-up period. When PMct was parsed into global and local components, their rate ratios differed, suggesting unmeasured confounding. This difference disappeared when γct was used as the exposure measure. Conclusions. We found associations between 1-year PM2.5 and mortality to be confounded by long-term time PM2.5 trends. This confounding was minimized using our year-adjusted PM2.5 measure, suggesting that future studies should adjust for long-term time trends in analyses of PM2.5 and mortality.