Particulate Air Pollution and Socioeconomic Position in Rural and Urban Areas of the Northeastern United States

Abstract

Objectives. Although differential exposure by socioeconomic position (SEP) to hazardous waste and lead is well demonstrated, there is less evidence for particulate air pollution (PM), which is associated with risk of death and illness. This study determined the relationship of ambient PM and SEP across several spatial scales.

Methods. Geographic information system-based, spatio-temporal models were used to predict PM in the Northeastern United States. Predicted concentrations were related to census tract SEP and racial composition using generalized additive models.

Results. Lower SEP was associated with small, significant increases in PM. Annual PM₁₀ decreased between 0.09 and 0.93 micrograms per cubic meter and PM_2.5 between 0.02 and 0.94 micrograms per cubic meter for interquartile range increases in income. Decrements in PM with SEP increased with spatial scale, indicating that between-city spatial gradients were greater than within-city differences. The PM–SEP relation in urban tracts was not substantially modified by racial composition.

Conclusions. Lower compared with higher SEP populations were exposed to higher ambient PM in the Northeastern United States. Given the small percentage change in annual PM_2.5 and PM₁₀, SEP was not likely a major source of confounding in epidemiological studies of PM, especially those conducted within a single urban/metropolitan area.

Although exposure to some pollutants, such as lead, is clearly associated with race/ethnicity and socioeconomic position (SEP), there is less evidence for particulate air pollution (PM). Further, a key issue in epidemiological studies of air pollution was whether findings were confounded or modified by SEP. ^1–6 Previous studies generally found lower SEP areas to have higher air pollutant exposures, specifically to total suspended particles in Hamilton, Ontario^7,8 and to traffic pollution.⁹ These studies relied on existing PM monitoring networks, which generally limited their focus to single urban areas. As a result, the spatial scale for which confounding or effect modification by SEP is most important is not well understood.

In this study, recently developed and validated air pollution models were used to predict annual PM₁₀ and PM_2.5 levels across the Northeastern United States. The study examined whether areas with lower SEP were associated with higher annual PM pollution levels and whether the associations of SEP and PM exposure were consistent across multiple spatial scales.

METHODS

This study examined the relation between ambient PM and SEP for census tracts located in Pennsylvania, New York, New Jersey, Connecticut, Rhode Island, and Massachusetts.

Data Sources

Socioeconomic and population density. SEP for the census tracts in the study area were derived from US Census 2000 Summary File 3 information¹⁰ by the Harvard School of Public Health Public Health Disparities Geocoding Project. SEP measures included measures of poverty (percentage of census tract population below poverty line), education (percentage of persons with less than high school education), and income (median household income and median household income adjusted for local cost of living). These variables were selected based on their associations with other environmental exposures and on their previously observed influence on the PM-mortality association.^5,11–13 Census tracts were used as the areal unit, because they were previously shown to capture socioeconomic gradients and to allow maximal geocoding of health records.^14–17 Census tract population density was obtained from Environmental Systems Research Institute Census 2000 Tract Information.¹⁸ Census tracts with a population density of 500 people per square mile or greater were classified as “urban.” All other tracts were classified as “rural.”

Racial minority composition.

Racial minority composition for each census tract was determined using the percentage of the tract population who identified themselves as Black, Asian Pacific Islander (API) (“Asian,” “Native Hawaiian and Other Pacific Islander”), or “some other race.” Data were obtained from Environmental Systems Research Institute Census 2000 Tract Information.¹⁸ Ninety-seven percent of those Census 2000 respondents who reported themselves some other race alone also reported their ethnicity as being of Hispanic or Latino origin.¹⁹ To examine effect modification by racial composition, racial composition within census tracts was further classified into “high” and “low” using the 75th and 25th percentile of the racial category values.

Cost of living adjusted median household income.

Cost of living information for each census tract was estimated using the American Chamber of Commerce Research Association (ACCRA) Composite Cost of Living Index (COLI) for the third quarter of 1999. The ACCRA COLIs are calculated based on their relation to the average of all US values, which is set at 100.²⁰ ACCRA COLI was available for 6633 of the 12 381 tracts. For tracts with missing data, COLI values were estimated using that for the nearest census tract centroid determined using a geographic information system (GIS). COLI data were used to calculate the cost of living adjusted median household income:

Predicted PM₁₀ and PM_2.5.

Annual PM₁₀ and PM_2.5 for 1999 were estimated for each study census tract using previously published GIS-based spatio-temporal models for PM₁₀ and PM_2.5.^21,22 Both models predict monthly outdoor PM concentrations using monthly smooth spatial terms and smooth regression terms of GIS-based and meteorological predictors. The model for PM₁₀ covered 1988 to 2002, whereas the GIS-based spatio-temporal model PM_2.5 model covered 1999 to 2002. Cross validation R² values for monthly PM₁₀ and PM_2.5 model predictions across these periods were 0.62 and 0.77, respectively, and 0.73 and 0.81, respectively, for long-term average predicted levels. Further details are found in Yanosky et al.^21,22 and Paciorek et al.²³

The models were used to estimate PM₁₀ and PM_2.5 concentrations across a 200 meter by 200 meter grid over the study area. This grid size allowed average PM concentrations to be calculated for each census tract in the study area, ensuring many points in all but the smallest tracts. Average PM₁₀ and PM_2.5 values for each census tract were calculated by first averaging across months to obtain an annual average for 1999 at each prediction point, and then averaging the predicted concentrations at all points within the tract. Exceptions to this method were 5 census tracts within New York City that were too small to include any grid points. For these tracts, the annual PM₁₀ and PM_2.5 were generated by taking the arithmetic averages of the predicted PM₁₀ and PM_2.5 for the 6 prediction points closest to the boundaries of each tract.

Data Analysis

Predicted particulate air pollution and socioeconomic position.

The relation of predicted PM and SEP was first examined without regard to race across the study region:

where y_i was the predicted annual average PM for 1999 for i = 1…i = 12 381 census tracts; UR_i was a binary variable categorizing each tract as urban (1) or rural (0) for 12 381 census tracts; SEP_i was 1 of the 4 SEP measures; s(e_i,n_i;edf) was a smooth spatial function of spatial location based on projected easting and northing coordinates of the census tract centroids for a specified number of estimated degrees of freedom (edf); and ϵ_ι was an error term. Models were fitted using the gam() function in the mgcv package in R.²⁴

The PM–SEP relationship was examined at different spatial scales by varying the amount of spatial smoothing (i.e., edf) performed by the smooth spatial function of location.²⁵ With increasing edf (decreasing spatial smoothing), spatial variability was filtered out at increasingly finer scales, thereby identifying SEP effects on PM from variability at scales finer than the spatial smoothing used. Three edf values were used to filter out (1) regional, (2) subregional, and (3) all variability, except within-metropolitan area variability. Each model was also run without a spatial smoothing term to examine the effects of SEP on predicted PM independent of the effects of spatial smoothing, allowing all scales of variability to influence the regression estimates. The amount of smoothing was specified in the gam() function in the mgcv package²⁶ in R using a fixed smoothing parameter.

Predicted particulate air pollution and race.

The relation of predicted PM and racial composition was examined, without regard to SEP in urban census tracts. Rural census tracts were not examined given the low percentages of Black, API, and some other race (75th percentile values = 1.63%, 0.94%, 0.61%, respectively). The analyses were based on the same structure and similar spatial smoothing levels as in the SEP models.

Effect modification by racial minority percentage.

Effect modification of the PM–SEP relationship by census tract racial composition was examined for each SEP measure:

where %Race_i was the percentage of the tract population who identified themselves as race i (e.g., Black, API, or some other race). Predicted annual average PM₁₀ and PM_2.5 for each combination of “higher” (75th percentile) and “lower” (25th percentile) racial composition and higher (75th percentile) and lower (25th percentile) SEP measures were calculated to facilitate interpretation of results.

RESULTS

Within the Northeastern United States, PM concentrations varied substantially by census tract, with annual mean PM₁₀ ranging between 9.82 and 33.45 micrograms per cubic meter and PM_2.5 between 6.02 and 18.74 micrograms per cubic meter. Annual PM₁₀ and PM_2.5 were higher in urban locations compared with rural locations, especially near the Pittsburgh, Philadelphia, and New York City metropolitan areas. PM concentrations were not substantially lower in coastal areas, as would be expected given their generally higher on-shore winds, likely the result of the monthly averaging and the fact that coastal areas were located on the edge of model surfaces. Measures of census tract-based SEP were significantly correlated with percentage of Blacks and other race categories within the census tracts, corresponding to lower SEP with higher percentages of these racial minorities. For example, correlation coefficients for comparisons of percentage of Blacks and other race categories with percentage of persons below the poverty level equaled 0.49 and 0.54, respectively, whereas those for comparisons with household adjusted income equaled −0.42 and −0.47, respectively. Conversely, SEP was weakly correlated with percentage of APIs. (Summary statistics for SEP, annual PM₁₀ and PM_2.5, and racial composition are presented in supplementary Table S1 at www.ajph.org.)

Figure 1 shows spatial variability in annual PM_2.5 for 3 levels of spatial smoothing. The greatest level of smoothing (332 edf; Figure 1A) estimated the PM–SEP relationship without the influence of their regional scale correlation and produced a reasonably smooth surface over the study region. The relationship, excluding regional and subregional variability in PM, was examined using an intermediate amount of smoothing, or 949 edf (Figure 1B). Figure 1C, the model with the largest edf, gave the most spatially variable surface, having filtered out most between-metropolitan area variability in PM. This model allowed the impacts of only within-metropolitan area variation of SEP on annual PM concentrations to be examined. Results for PM₁₀ for 3 similar levels of smoothing were comparable.

FIGURE 1.

Predicted census tract mean fine particulate a pollution (PM_2.5) with 3 levels of spatial smoothing: (a) excluding regional variability: 332 estimated degrees of freedom (edf); (b) excluding regional and subregional variability: 949 edf; and (c) allowing within-metropolitan-area variability only: 1769 edf.

Predicted Change in Particulate Air Pollution With Socioeconomic Position

Table 1 shows the change in annual PM₁₀ and PM_2.5 for an interquartile range (IQR) increase in census tract-based SEP measures for 3 spatial smoothing levels. In urban census tracts, lower SEP was consistently associated with small but significant increases in PM₁₀ and PM_2.5, largely independent of the specific SEP measure. Annual PM₁₀ and PM_2.5, for example, decreased between 0.09 and 0.93 micrograms per cubic meter (depending on spatial scale) and between 0.02 and 0.94 micrograms per cubic meter, respectively, for an IQR increase in income. Low educational attainment and increased poverty were also significantly associated with higher annual concentrations of PM for each spatial scale. An IQR increase in the percent of adult residents with less than a high school education, for example, was associated with PM₁₀ increases of 0.10–0.48 micrograms per cubic meter and PM_2.5 increases of 0.09–0.60 micrograms per cubic meter for urban tracts. Similar results were found for rural census tracts, although results were less consistent.

TABLE 1.

Predicted Change in 1999 Mean Census Tract Particulate Air Pollution (PM₁₀) and PM_2.5 Concentrations (micrograms per cubic meter) With an Increase of 1 Interquartile Range (IQR) in Socioeconomic Position Indicator

		PM10			PM2.5
	IQR	edf	Urban Tracts	Rural Tracts	edf	Urban Tracts	Rural Tracts
% < High school education	15.99
		1964a	0.11**	0.12*	1982a	0.09**	0.04*
		731b	0.10**	0.16**	922b	0.13**	0.06**
		236c	0.11**	0.12*	319c	0.20**	0.09**
		No smooth termd	0.48**	0.65**	No smooth termd	0.60**	0.52**
Median HH income, $	27 358
		1 966a	−0.25**	−0.09*	1982a	−0.16**	0.00
		756b	−0.25**	−0.07	950b	−0.20**	−0.02*
		246c	−0.25**	−0.01	333c	−0.27**	−0.05**
		No smooth termd	−0.51**	−0.15	No smooth termd	−0.42**	−0.07
Adjusted median HH  income, $	23 505
		1966a	−0.35**	−0.4**	1982a	−0.22**	−0.12**
		756b	−0.38**	−0.4**	950 b	−0.27**	−0.15**
		246c	−0.43**	−0.33**	333 c	−0.35**	−0.18**
		No smooth termd	−0.93**	−0.12	No smooth termd	−0.94**	0.00
% Below poverty	12.58
		1968a	0.11**	0.33**	1984a	0.08**	0.09**
		767b	0.12**	0.36**	962b	0.11**	0.12**
		250c	0.15**	0.35**	338c	0.16**	0.15**
		No smooth termd	0.47**	0.39**	No smooth termd	0.48**	0.29**

Note. edf = estimated degrees of freedom; HH = household.

^aAllows within-metropolitan variability only.

^bExcludes regional and subregional variability.

^cExcludes regional variability.

^dAllows all geographic variability.

*P < .05; **P < .001.

For most measures of SEP, the magnitude of change in predicted PM concentrations decreased when filtering out increasingly smaller scales of variation in PM, especially for PM_2.5. For these measures, the largest changes in predicted PM₁₀ and PM_2.5 concentrations with SEP were seen for models with no spatial smooth term. These values ranged in magnitude from 0.39–0.93 micrograms per cubic meter for PM₁₀ and 0.29–0.94 micrograms per cubic meter for PM_2.5. Because models without a spatial term did not filter out any of the spatial variability in PM among the census tracts, their higher effect estimates suggested that much of the effect of SEP on PM was driven by its association over large spatial scales, perhaps from proximity to regional PM sources, such as the coal-fired power plants of the Ohio River Valley. For low education and unadjusted median household income, observed changes in PM₁₀ were generally consistent across levels of spatial smoothing.

Predicted Change in Particulate Air Pollution With Racial Minority Percentage

As shown in Table 2, the relationship between annual PM and census tract racial composition varied by racial group and spatial scale. The percentage of Black residents, for example, was associated with PM₁₀ increases of 0.02–0.09 micrograms per cubic meter in models with no smooth spatial term and at the largest spatial scale. The association was attenuated and nonsignificant in models at smaller spatial scales (i.e., larger edf). Correspondingly, the associations were largest in models with no spatial smoothing term. For PM_2.5, increases of 0.01–0.25 micrograms per cubic meter were significant across spatial scales. In contrast, PM₁₀ decreased with increased percentage of API residents at the smallest spatial scale, but increased at larger spatial scales. Similarly, effects of the percentage of persons of other race on PM₁₀ varied across spatial scales: at smaller spatial scales, the percentage of persons of other race was associated with PM₁₀ decreases of 0.02–0.03 micrograms per cubic meter, whereas it was associated with increases of 0.16 micrograms per cubic meter in models with no smooth spatial term. For PM_2.5, increases of 0.01–0.19 micrograms per cubic meter were associated with the percentage of persons of other race across all spatial scales, with the slopes again decreasing when filtering out variability at increasingly smaller scales.

TABLE 2.

Predicted Change in 1999 Urban Census Tract Mean Particulate Air Pollution (PM₁₀) and PM_2.5 Concentrations With an Increase of 1 Interquartile Range (IQR) in Racial Minority Percentage

		PM10		PM2.5
	IQR	edf	(μg/m3)	edf	(μg/m3)
Percent Black	11.97	1967a	−0.00	1984a	0.01*
		751b	0.01	945b	0.02**
		244c	0.02*	329c	0.03**
		No smooth termd	0.09**	No smooth termd	0.25**
Percent Asian	3.59	1967a	−0.02*	1984a	0.00
		751b	0.01	945b	0.00
		244c	0.04**	329c	0.01*
		No smooth termd	0.12**	No smooth termd	0.18**
Percent “some other race”	3.93	1967a	−0.03**	1984a	0.01*
		751b	−0.03**	945b	0.02**
		244c	−0.02**	329c	0.04**
		No smooth termd	0.16**	No smooth termd	0.19**

Note. edf = estimated degrees of freedom.

^aAllows within-metro politan variability only.

^bExcludes regional and subregional variability.

^cExcludes regional variability.

^dAllows all geographic variability.

*P < .05; **P < .001.

Table 3 shows the predicted annual PM within urban census tracts with higher and lower percentages of Black residents, adjusted income, and percentages of residents below the poverty line. The relations between annual PM concentrations and SEP were not consistently or substantially modified by the census tract percentage of the examined racial groups, when holding the SEP measure fixed (API and other race results not shown). In some instances, the impact of effect modification was opposite to that hypothesized. For example, decreased predicted PM was associated with higher percentages of Black residents, when controlling for SEP. Although the interaction terms in the model were generally statistically significant, the modest changes suggested that racial composition had only a modest impact on the PM–SEP relationship.

TABLE 3.

Predicted Urban Tract Mean PM (micrograms per cubic meter) for Lower (25th Percentile) and Higher (75th Percentile) Values of Socieconomic Position (SEP) and Percentage of Black Variables

	PM10, μg/m3 (SE)		PM2.5, μg/m3 (SE)
	Lower % Black	Higher % Black	Lower % Black	Higher % Black
Lower adjusted income	19.18 (0.28)	18.81 (0.27)	11.35 (0.11)	11.22 (0.11)
Higher adjusted income	18.51 (0.28)	18.15 (0.27)	10.96 (0.11)	10.84 (0.11)
Lower % below poverty	18.57 (0.25)	18.38 (0.25)	11.15 (0.11)	11.11 (0.10)
Higher % below poverty	19.24 (0.25)	19.05 (0.25)	11.50 (0.11)	11.47 (0.10)

DISCUSSION

This study found that annual PM was consistently and significantly higher in census tracts of lower SEP compared with those of higher SEP, although the relative impact was smaller than for some other environmental contaminants. Controlling for spatial variability at increasingly smaller spatial scales across both urban and rural census tracts, this study observed increased levels of PM₁₀ of 0.09–0.43 micrograms per cubic meter and of PM_2.5 of 0.02–0.35 micrograms per cubic meter with an IQR decrease in SEP. Among urban tracts, larger increases in PM were generally seen when controlling for only regional spatial variability (i.e., the same increment in income within-city was associated with a smaller increment in PM exposure than if the increment was between-city or between-region). The observed changes in predicted PM_2.5 associated with changes in SEP were less than those of PM₁₀, which was expected given the lower concentrations in PM_2.5. The study's findings provided evidence that in the Northeastern United States, populations of lower SEP might be exposed to higher annual concentrations of ambient PM pollution than populations of higher SEP, although observed differences in PM concentrations were small.

In addition, this study showed that the magnitude of PM–SEP relation varied by spatial scale, with large-scale differences in SEP in the Northeastern United States more strongly associated with annual PM than smaller scale differences in SEP, such as those between neighborhoods within a city. This finding suggested that in the Northeastern United States “between” compared with “within” city differences in census tract level SEP were more important predictors of annual PM concentrations. Further, results suggested that local PM sources, such as traffic, truck terminals, or industrial sources, had a smaller influence on the PM–SEP relationship than regionally transported pollutants and larger PM sources, such as coal-fired power plants. Because other studies showed bus/truck depots and large roadways influenced PM concentrations significantly^27,28 these findings further suggested that the influence of local PM sources on the PM–SEP relationship might only be observed at smaller spatial scales, such as the block group, or over shorter time periods, such as a day or week. In addition, the results suggested that confounding of long-term PM health effects by SEP was more likely in studies of multiple metropolitan areas compared with within just one area.

The study also found that census tract concentrations of PM_2.5 in particular were associated with the racial composition of census tracts, with PM concentrations higher in census tracts with higher percentages of racial minorities. As was the case with SEP, the magnitude of association of PM and racial composition was lower when the models controlled for regional spatial variability, again suggesting that much of the effect of race on PM was driven by large-scale spatial variation. The similar relations of SEP and race on PM were consistent with their observed significant correlation with each another. However, the change in PM with an IQR change in percentage of racial composition was smaller than for SEP.

In these analyses, racial composition did not modify the PM–SEP association, which suggested that on the census tract level, the effects of socioeconomic disadvantage on ambient PM were independent of racial composition. However, it was also possible that the study's reliance on the census tract as the unit of analysis and the observed large influence of regional PM sources on local PM might have obscured the ability to examine the impact of race and local sources on the PM–SEP relation. Further, given moderate correlations between area SEP and minority composition, it was also possible that race and SEP were acting as measures of similar social factors. If so, race would be unlikely to modify the PM–SEP association. These issues warrant further study, not only for PM but also for other pollutants and other types of environmental media.

These findings generally agreed with studies conducted within single urban areas. In Hamilton, Ontario, Jerrett et al.⁵ found significant negative associations between SEP (assessed using housing value, unemployment, and income) and total suspended particulate (TSP) levels, whereas Finkelstein et al.⁷ found that as SEP increased, exposures to TSP and sulfur dioxide decreased significantly. In Worcester, Massachusetts, Yanosky et al.²⁹ found higher levels of nitrogen dioxide, a marker of traffic pollution, associated with lower socioeconomic status areas, suggesting that lower socioeconomic status individuals were disproportionately exposed to traffic-related pollutants, which included PM. For California's South Coast Air Basin, Marshall³⁰ showed that higher than average exposures to benzene, butadiene, chromium particles, and diesel particles were found for people who were non-White, living in lower SEP households, and living in areas with high population density. Interestingly, opposite relations were shown for ozone.

A limitation of this study was that the use of census tracts as the areal unit did not allow determination of whether finer-scale variability in SEP was associated with differences in PM. Although Figure 1C depicts the finest scale of spatial smoothing, the study was only able to examine the PM–SEP relationship at a spatial scale greater than that of the census tract level. A similar assessment using block group or individual level data is needed. In addition, because estimates of uncertainty in the regression estimates and their P values did not account for residual spatial variability apart from that modeled in the smooth spatial terms, uncertainty might be underestimated, particularly for models with smoother spatial surfaces; however, the extremely high levels of statistical significance for the primary effects of SEP made it unlikely that the findings were a result of chance.

For rural tracts, several of the observed SEP-mediated changes in PM were not statistically significant. This finding might be the result of a true lack of association. Alternatively, it might be the result of the relatively larger size of rural tracts in the study area (mean area = 42 square miles) compared with urban tracts in the study area (mean area = 1.5 square miles) or to the smaller number of rural tracts (19.5% of all tracts) to observe differences in predicted PM. The limited spatial resolution of the analysis in rural areas suggested that future studies of pollution in rural areas should be based on smaller geographic units, such as individual-specific or block group-specific data.

Household income is an often used indicator of SEP in the United States. This study used 2 measures of household income: median household income and COLI-adjusted median household income. The latter accounts for differences in the cost of living and thus may more accurately reflect the relationship of real household buying power and predicted PM concentrations.^31,32 Although COLI-adjusted income may improve the utility of income as a SEP measure, it nevertheless remains imperfect. In this study, COLI reported by local chambers of commerce was only available for slightly more than half of the census tracts, with those tracts located almost entirely in urban areas. Although the method of assigning COLI information to the remaining census tracts might be reasonable, it almost certainly failed to capture some of the variation in cost of living, especially for rural census tracts. The impact of missing COLI data on these results was not known, but might explain some of the observed difference in the association between PM and the COLI adjusted and unadjusted income. Interestingly, the COLI adjusted income measure seemed to be more strongly related to PM in rural areas than the unadjusted income measure, especially for PM₁₀. This observation suggested that the imputation method for missing values did not result in more misclassification in rural compared with urban areas, although few rural areas collected COLI data.

Finally, it should be noted that the increased levels of PM concentrations associated with lower SEP observed in this study were relatively small. Nevertheless, almost all associations were statistically significant, and given evidence of a linear concentration–response curve and lack of threshold for PM,^33–35 the increased public health burden borne by socioeconomically disadvantaged and minority populations because of their exposure to PM was not insubstantial. Further, these findings suggested that the potential for confounding by SEP in epidemiology studies of PM was small, because SEP explained a small part of the variation in PM exposure. PM concentrations increased by less than 1 microgram per cubic meter per IQR decrease in SEP, with increases of less than 0.5 microgram per cubic meter when a spatial smoothing term was included in the model. These increases were small relative to their mean levels, which for PM₁₀ and PM_2.5 equaled 12.73 and 12.60 micrograms per cubic meter, respectively. For PM₁₀, the increases were more meaningful compared with its IQR, whereas for PM_2.5, they remained small. The observed increases in PM with SEP were also small relative to the 10 micrograms per cubic meter in PM₁₀ or PM_2.5 that were generally used in epidemiological studies to relate long-term PM exposures to health. For example, the Harvard Six Cities study found a relative risk of cardiopulmonary mortality of 1.16 (95% CI = 1.07, 1.26) per 10 micrograms per cubic meter of PM_2.5,³⁶ with similar relative risks reported in analyses of the American Cancer Society cohort.⁵

The results provided, among the first evidence, that annual exposure to particles was negatively associated with SEP, and to a lesser extent, racial composition in the Northeastern United States, suggesting that persons residing in socioeconomically disadvantaged census tracts could be exposed to higher annual levels of PM. If, as previously found, individuals living in these lower SEP census tracts were also more vulnerable to the impacts of air pollution exposures,¹ these study findings might have important implications for understanding the long-term health impacts of PM.