Reanalysis of the association between reduction in long-term PM2.5 concentrations and improved life expectancy.

Abstract

Background: Much of the current evidence of associations between long-term PM_2.5 and health outcomes relies on national or regional analyses using exposures derived directly from regulatory monitoring data. These findings could be affected by limited spatial coverage of monitoring data, particularly for time periods before spatially extensive monitoring began in the late 1990s. For instance, Pope et al. (2009) showed that between 1980 and 2000 a 10 μg/m³ reduction in PM_2.5 was associated with an average 0.61 year (standard error (SE) = 0.20) longer life expectancy. That analysis used 1979-1983 averages of PM_2.5 across 51 U.S. Metropolitan Statistical Areas (MSAs) computed from about 130 monitoring sites. Our reanalysis re-examines this association using modeled PM_2.5 in order to assess population- or spatially-representative exposure. We hypothesized that modeled PM_2.5 with finer spatial resolution provides more accurate health effect estimates compared to limited monitoring data.

Methods: We used the same data for life expectancy and confounders, as well as the same analysis models, and investigated the same 211 continental U.S. counties, as Pope et al. (2009). For modeled PM_2.5, we relied on a previously-developed point prediction model based on regulatory monitoring data for 1999-2015 and back-extrapolation to 1979. Using this model, we predicted annual average concentrations at centroids of all 72,271 census tracts and 12,501 25-km national grid cells covering the contiguous U.S., to represent population and space, respectively. We averaged these predictions to the county for the two time periods (1979-1983 and 1999-2000), whereas the original analysis used MSA averages given limited monitoring data. Finally, we estimated regression coefficients for PM_2.5 reduction on life expectancy improvement over the two periods, adjusting for area-level confounders.

Results: A 10 μg/m³ decrease in modeled PM_2.5 based on census tract and national grid predictions was associated with 0.69 (standard error (SE) = 0.31) and 0.81 (0.29) -year increases in life expectancy. These estimates are higher than the estimate of Pope et al. (2009); they also have larger SEs likely because of smaller variability in exposure predictions, a standard property of regression. Two sets of effect estimates, however, had overlapping confidence intervals.

Conclusions: Our approach for estimating population- and spatially-representative PM_2.5 concentrations based on census tract and national grid predictions, respectively, provided generally consistent findings to the original findings using limited monitoring data. This finding lends additional support to the evidence that reduced fine particulate matter contributes to extended life expectancy.