Fine Particle Sources and Cardiorespiratory Morbidity: An Application of Chemical Mass Balance and Factor Analytical Source-Apportionment Methods

Recent interest in the health effects of particulate matter (PM) has focused on identifying sources of PM that pose the greatest health risks. Because it is likely that not all PM is equally toxic, epidemiologic models that incorporate source-resolved PM may provide a step toward targeting the most important causal agents and refining traditional mass-based PM standards. Quantifying health risks associated with sources such as biomass burning, power plants, gasoline and diesel emissions, rather than individual pollutants, may also capture complex multipollutant interactions that more accurately reflect the etiologic relationships between PM and adverse health. Few epidemiologic studies, however, have included source-apportionment data in their examinations of PM health effects (Ito et al. 2006; Laden et al. 2000; Mar et al. 2000, 2006; Ozkaynak and Thurston 1987; Schreuder et al. 2006). The limited application of source apportionment may be attributable partly to uncertainties regarding optimal methods for conducting PM source apportionment, as well as the lack of suitable air quality data for analysis. Source-apportionment methods used in previous studies have generally relied on factor analytic approaches. For example, Laden et al. (2000) grouped elemental PM concentrations from six U.S. cities into a small numbers of categories, or “factors” (Laden et al. 2000). Significant associations were found between mortality and the traffic and coal combustion factors, with the largest effect size for the traffic factor. No significant associations were observed for the oil and soil factors. A series of recent analyses examined the associations between source-resolved PM estimates and mortality in Washington, DC, and Phoenix, Arizona, using several different multivariate factor analytic methods in each city (Hopke et al. 2006; Ito et al. 2006; Mar et al. 2006; Thurston et al. 2005). In these analyses, source apportionment was conducted on samples collected twice a week, using absolute principal components analysis (PCA), UNMIX (a multivariate receptor model), and positive matrix factorization (PMF). Results showed that variability among the methods was small when compared with overall source-apportionment model uncertainty, and suggested that these apportionment methods may be useful in discerning source-specific health effects. The authors note the relatively limited sample size for these data sets and their inability to robustly identify certain source categories (e.g., specific mobile source types). Questions also remain concerning the generalizability of these findings to other locations with different aerosol compositions, the marginal benefit of using source-apportioned data over single-species tracers, and whether analyses using other source-apportionment methods, notably chemical mass balance (CMB), will show the same pattern of agreement. Here we present and compare results from epidemiologic analyses of emergency department (ED) visits and source-resolved PM2.5 (PM with aerodynamic diameter ≤ 2.5 μm; fine PM) obtained using PMF, modified CMB, and a single-species tracer approach. This analysis is the first to compare epidemiologic findings generated using both factor analysis and mass balance source-apportionment methods. The data used in this analysis were collected in Atlanta, Georgia, a unique location for conducting this type of health-effects study given the existence of an extensive time-series of daily speciated PM2.5 measurements and corresponding hospital records. These data have been previously characterized in several source-apportionment and epidemiologic analyses (Kim et al. 2004; Marmur et al. 2005; Metzger et al. 2004; Peel et al. 2005). We compare results across methods and assess the robustness of health risk estimates for cardiopulmonary ED visits. The implications of using one or several methods for understanding the sources of PM2.5-mediated health risks are also addressed.