Analysis of Personal and Home Characteristics Associated with the Elemental Composition of PM2.5 in Indoor, Outdoor, and Personal Air in the RIOPA Study.

The complex mixture of chemicals and elements that constitute particulate matter (PM*) varies by season and geographic location because source contributors differ over time and place. The composition of PM having an aerodynamic diameter < 2.5 μm (PM2.5) is hypothesized to be responsible, in part, for its toxicity. Epidemiologic studies have identified specific components and sources of PM2.5 that are associated with adverse health outcomes. The majority of these studies use measures of outdoor concentrations obtained from one or a few central monitoring sites as a surrogate for measures of personal exposure. Personal PM2.5 (and its elemental composition), however, may be different from the PM2.5 measured at stationary outdoor sites. The objectives of this study were (1) to describe the relationships between the concentrations of various elements in indoor, outdoor, and personal PM2.5 samples, (2) to identify groups of individuals with similar exposures to mixtures of elements in personal PM2.5 and to examine personal and home characteristics of these groups, and (3) to evaluate whether concentrations of elements from outdoor PM2.5 samples are appropriate surrogates for personal exposure to PM2.5 and its elements and whether indoor PM2.5 concentrations and information about home characteristics improve the prediction of personal exposure. The objectives of the study were addressed using data collected as part of the Relationships of Indoor, Outdoor, and Personal Air (RIOPA) study. The RIOPA study has previously measured the mass concentrations of PM2.5 and its elemental constituents during 48-hour concurrent indoor, outdoor (directly outside the home), and personal samplings in three urban areas (Los Angeles, California; Houston, Texas; and Elizabeth, New Jersey). The resulting data and information about personal and home characteristics (including air-conditioning use, nearby emission sources, time spent indoors, census-tract geography, air-exchange rates, and other information) for each RIOPA participant were downloaded from the RIOPA study database. We performed three sets of analyses to address the study aims. First, we conducted descriptive analyses to describe the relationships between elemental concentrations in the concurrently gathered indoor, outdoor, and personal air samples. We assessed the correlation between personal exposure and indoor concentrations as well as personal exposure and outdoor concentrations of each element and calculated ratios between them. In addition, we performed principal component analysis (PCA) and calculated principal component scores (PCSs) to examine the heterogeneity of the elemental composition and then tested whether the mixture of elements in indoor, outdoor, and personal PM2.5 was significantly different within each study site and across study sites. Secondly, we performed model-based clustering analysis to group RIOPA participants with similar exposures to mixtures of elements in personal PM2.5. We examined the association between cluster membership and the concentrations of elements in indoor and outdoor PM2.5 samples and personal and home characteristics. Finally, we developed a series of linear regression models and random forest models to examine the association between personal exposure to elements in PM2.5 and (1) outdoor measurements, (2) outdoor and indoor measurements, and (3) outdoor and indoor measurements and home characteristics. As we developed each model, the improvement in prediction of personal exposure when including additional information was assessed. Personal exposures to PM2.5 and to most elements were significantly correlated with both indoor and outdoor concentrations, although concentrations in personal samples frequently exceeded those of indoor and outdoor samples. In general, for most PM2.5 elements indoor concentrations were more highly correlated with personal exposure than were outdoor concentrations. PCA showed that the mixture of elements in indoor, outdoor, and personal PM2.5 varied significantly across sample types within each study site and also across study sites within each sample type. Using model-based clustering, we identified seven clusters of RIOPA participants whose personal PM2.5 samples had similar patterns of elemental composition. Using this approach, subsets of RIOPA participants were identified whose personal exposures to PM2.5 (and its elements) were significantly higher than their indoor and outdoor concentrations (and vice versa). The results of linear and random forest regression models were consistent with our correlation analyses and demonstrated that (1) indoor concentrations were more significantly associated with personal exposure than were outdoor concentrations and (2) participant reports of time spent at their home significantly modified many of the associations between indoor and personal concentrations. In linear regression models, the inclusion of indoor concentrations significantly improved the prediction of personal exposures to Ba, Ca, Cl, Cu, K, Sn, Sr, V, and Zn compared with the use of outdoor elemental concentrations alone. Including additional information on personal and home characteristics improved the prediction for only one element, Pb. Our results support the use of outdoor monitoring sites as surrogates of personal exposure for a limited number of individual elements associated with long-range transport and with a few local or indoor sources. Based on our PCA and clustering analyses, we concluded that the overall elemental composition of PM2.5 obtained at outdoor monitoring sites may not accurately represent the elemental composition of personal PM2.5. Although the data used in these analyses compared outdoor PM2.5 composition collected at the home with indoor and personal samples, our results imply that studies examining the complete elemental composition of PM2.5 should be cautious about using data from central outdoor monitoring sites because of the potential for exposure misclassification. The inclusion of personal and home characteristics only marginally improved the prediction of personal exposure for a small number of elements in PM2.5. We concluded that the additional cost and burden of indoor and personal sampling may be justified for studies examining elements because neither outdoor monitoring nor questionnaire data on home and personal characteristics were able to represent adequately the overall elemental composition of personal PM2.5.