Influence of exposure differences on city-to-city heterogeneity in PM 2.5 -mortality associations in US cities

There remains uncertainty to the cause of the observed heterogeneity in city-to-city PM_2.5-mortality associations in U.S. based multi-city studies. Often this heterogeneity has been attributed to the regional and national variation in components of PM_2.5; however, a clear difference in the air pollution mixtures has yet to be identified. A previous study examining city-to-city differences in ambient concentrations and the relationship between PM_2.5 components and gaseous pollutants with PM_2.5 mass did not provide any evidence of clear differences in ambient concentrations or sources between cities [9]. Additionally, the evidence from epidemiologic and toxicological studies has not demonstrated that any one component or source is more strongly related with specific health outcomes [8, 18, 19].

Infiltration, defined as the fraction of the outdoor concentration that penetrates indoors and remains suspended, varies between cities, between homes, and over time within homes [20]. Allen at al. developed models to predict infiltration based on behavioral factors such as air conditioning use and windows opening that can vary seasonally [21]. The wide variation observed in residential infiltration rates [21] supports the hypothesis that city-to-city differences in the personal exposure-ambient monitor relationship may also be contributing to the observed heterogeneity in PM_2.5 mortality risk estimate across cities This has led to epidemiologic studies examining factors that may influence individual exposures and ultimately associations in studies of short-term air pollution exposures and various health outcomes (e.g., mortality), such as air exchange rate [22], infiltration rates [23], and air conditioning use [2]. Each of these studies has provided initial information on the importance of accounting for factors that may influence individual exposure, but the overall air pollution exposures people encounter are dictated by a variety of factors, not just one at a time. To further expand upon some of these initial studies, Baxter et al. [3] used information on individual exposure factors representative of infiltration and commuting to create clusters of cities with similar exposure profiles to examine whether different exposure profiles help explain the city-to-city heterogeneity in PM_2.5-mortality risk estimates.

Building off the work detailed in Baxter et al. [3], the objective of this analysis was to determine whether the previously developed clusters help explain the city-to-city heterogeneity in PM_2.5-mortality associations. Overall, we reported a 0.33% increase in non-accidental mortality for a 10 ?g/m³ increase in previous 2-day moving average (lag 0–1) PM_2.5 concentrations for 77 U.S. cities for the years of 2001–2005. The examination of cluster assignment was found to be marginally significant in explaining the heterogeneity in PM_2.5-mortality associations. When comparing results between clusters, we only observed evidence of significant differences in mortality associations between Cluster 1 and 4 and Cluster 3 and 4, while the PM_2.5-mortality associations in Cluster 1 and 3 are significantly smaller in magnitude compared to those in Cluster 4. As a sensitivity analysis, we performed a meta-regression on the individual factors that comprise the clusters and found that only home size appeared to explain the heterogeneity in the PM_2.5-mortality associations, with larger associations in larger homes. However, in Baxter et al. [3] home size was not well correlated with the other exposure factors included in the infiltration exposure factors evaluations.

Upon closer examination, there are clear differences in the housing characteristics between clusters that appeared to contribute in explaining the heterogeneity in PM_2.5 mortality associations. Air exchange rates have been found to be higher in larger and older homes [24] and in homes with less central AC due to more opening of windows [25] resulting in higher exposures to outdoor PM and associations larger in magnitude. Homes in Cluster 1 were on average 426 ft² smaller, had 27% less central AC, and similar age homes compared to Cluster 4. Larger health effect were observed in Cluster 4 suggesting higher exposures in those homes. This is in agreement with the significant findings on home size in the sensitivity analyses with larger homes size associated with larger health effect estimates. Similarly, the underlying exposure profiles of Cluster 3 and 4 may help explain the difference in associations between the two clusters. Cluster 3 has a larger percentage of homes with central AC, as well as homes that are newer and smaller than those in Cluster 4. This difference between Clusters 3 and 4 would indicate that air exchange rates may be smaller in Cluster 3 resulting in lower exposures, which would subsequently result in associations smaller in magnitude for Cluster 3.

It is important to recognize that this study is subject to inherent limitations. One of the main limitations in the epidemiological analysis was the potential for exposure error from using an adjusted average of PM_2.5 concentrations from a few monitors to characterize a population exposure in each of the cities. However, PM_2.5 is relatively spatially homogenous and studies of personal exposures have shown that temporal variability in outdoor PM_2.5 concentrations are a good surrogate for temporal variability in personal PM_2.5 exposures [26, 27]. There is also potential for exposure error as the exposure factors that were used to generate the clusters are surrogates rather than direct measurements of residential infiltration. Furthermore, while reductions in residential infiltration will reduce exposures to PM_2.5 of ambient origin it will also increase exposures to PM_2.5 generated from indoor sources. This indoor PM_2.5 may be independently associated with adverse health effects. Finally, a more thorough evaluation of potential differences between the five clusters examined in this study was limited by the small number of cities that comprised Clusters 2 and 5.