Air pollution is a major environmental public health issue in the United States directly affecting wellbeing and quality of life. Numerous studies attribute excess respiratory and cardiovascular morbidity and higher mortality rates to air pollution, particularly among urbanized places [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34]. Air quality regulations in the United States from the beginning of the Clean Air Act in the early 1970s have separated criteria air pollutants (CAPs), which are ubiquitous chemicals in the environment (e.g., sulfur dioxide, ozone, and particulate matter (2.5 μm), from hazardous air pollutants (HAPs). HAPs comprise 187 chemicals including arsenic, lead, cadmium, pollutant gases, solvents, and pesticides many of which are thought to have no minimum threshold of exposure that can be considered safe [35]. While many State Implementation Plans employed under the Clean Air Act have made progress in reducing CAPs and other chemicals regulated as HAPs, evidence on environmental exposure to HAPs or their associated health effects is quite limited [36,37,38].

While acute exposures to high concentrations to HAPs are clearly dangerous, the cumulative health effects of low level chronic exposure are uncertain, given the multiple organs targeted by these metals and chemicals. Point source HAP emissions typically involve small amounts of dozens of chemicals, each with potentially different health effects based on the organ targeted by the chemical. For many of these chemicals small exposures are biologically capable of significant health effects. Similar to CAPs, ambient air exposure to HAPs is typically concentrated in urban and industrial areas and among adjacent populations.

HAPs have been problematic for both risk assessment methods and risk management policy. The cumulative burden of air pollution exposure among residential communities adjacent to industrial land use and point source emissions presents significant environmental justice policy issues for air quality management, and scientific challenges for risk assessment. One of the major challenges in environmental epidemiology and risk assessment in urban settings is the confounding of health effects associated with poverty with those related to environmental exposure. Complicating differential exposure among those of lower socioeconomic position is the potential for differential susceptibility to chemical exposures. The National Academy of Sciences has challenged the adequacy of risk assessment methods due to the potential for differential susceptibility, and cites the need to improve our understanding of population vulnerability to chemical exposure, particularly among populations subject to high levels of nonchemical stressors such as socioeconomic deprivation (SED) [39].

This paper seeks to contribute to risk assessment methods by demonstrating the value of incorporating social epidemiology in analyzing social gradients in HAP exposure related to urbanization and neighborhood SED. Developing better metrics and dose-response models relating to social gradients in exposure and health outcomes, particularly among vulnerable populations, can help advance cumulative risk assessment.

The study uses national emissions inventory data as modeled in the National Air Toxics Assessment in 2005 and includes health hazard exposure indices for chemicals related to risk of cancer risk and respiratory and neurological noncancerous health endpoints. We used a social determinant of health framework to develop a multilevel model to analyze how the baseline range HAP exposure hazards shifts with level of urbanization and related economic activity and degree of SED at the local (neighborhood) level. Urbanization is conceptualized as the concentration of residential population and economic activity, and is measured by local census tract population density, and tracts’ respective county population size, number employed and aggregate wage earned the tracts’ host county. The Townsend Index of Socioeconomic Deprivation (TSI) is used to measure SED at the census tract level [40]. Assessment of differential exposure relating to SED is strengthened by including known or suspected confounders and covariates relating to HAPs.

This study conceptualizes variation in local neighborhood chemical exposure to be decisively influenced by the population size in the region (with Host County of census tract as proxy) in combination with the economic activity generated in region (host county business metrics as proxy). How SED is generated and how it is geographically distribution SED is complicated [40,41,42,43]. While there is substantial rural poverty in the USA, greater numbers of households with SED are located in urban areas, often in ‘inner cities’. The key question is whether SED is confounded with HAP because of their common urban context, or whether SED works to modify HAP exposure.

Accordingly, in this study we tested the following hypotheses:

Hypothesis 1 identifies population size and level of economic activity as primary drivers of HAP exposure hazard. The second hypothesis addresses the necessary condition for SED to be confounded with HAP exposure hazard: their common relationship with the urban concentration of population and level of economic activity. As an alternative to a spurious relationship between SED and HAP exposure hazard, hypothesis 3 presents a SED differential exposure model. The geographic concentration of HAP exposure and higher SED households, it is argued, is related to “disinvestment” and public policies that work as a form of disenfranchisement among those that live in economically and politically weak urban places. In this model, neighborhood SED is directly related to HAP exposure after adjustment for urban concentration or level of economic activity.

This study evaluated evidence of “differential exposure” of HAP exposure hazard related to neighborhood SED, independent of U.S. geographic region and level of urbanization in terms of population size, density and regional economic activities. This research addresses the question: how does level of HAP respiratory, neurological, and cancer exposure hazard vary by regional population size and level of economic activity and local population density and SED. Is there evidence of differential HAP exposure related to localized SED, independent of the regional context of population level and economic activity, or is SED confounded with urbanization and HAP exposure? Among the unique contributions of this study are the application of a multilevel exposure model that examines the relationship of HAP exposure and neighborhood SED after adjustment for the level of urbanization, population density, and the regional level of economic activity that is monitored at the county level.

The study design involves a retrospective, cross-sectional, population-based analysis of HAP exposure at the census tract level in 2005. Predictors of HAP exposure include census-tract neighborhood SED and urbanization as measured by population density at the census tract level, and two regional variables—county population size and county business activity. The study population consists of the 65,166 census tracts (99.6% of the universe of 65,443) and their respective 3121 United States (U.S.) counties (99.4% of the universe of 3141 counties) for which there is complete data available.

We assembled three databases to permit the specification of a multilevel HAP differential exposure model based on census tract, county, and regional criteria: (a) the National Air Toxics Assessment (NATA) HAP exposure data relating to health hazard exposure measures at the census tract level; (b) decennial census data (2000) on tract characteristics relating to population density and socioeconomic deprivation, and related demographics for tracts’ host counties, i.e., regional population; and (c) County Business Pattern survey data for all U.S. counties in 2005 on number employed and annual wages. Regarding the geographic dimension of the analysis, we used the Census Bureau’s categorization of States into nine regions to profile HAP exposure. Figure 1 summarizes the variables and indicators included in the multilevel exposure model used in this study.

We examined the variation in modeled HAP exposure estimates from the NATA data regarding respiratory, neurological, and cancer-related exposure hazard indices and cancer risk factors, and give particular attention to the magnitude and upper tail of the distribution of HAP exposure: first at progressive deciles of census tracts related to population size, density, economic activity, and the Townsend Index, to better understand the social gradient of HAP exposure in the United States unadjusted for covariates; and then we tested for differential HAP exposure related to tract SED after adjustment for tract and county population size, density, and economic activity focusing on three HAP hazard exposure metrics (respiratory, neurological, and cancer) for 2005 using random intercept mixed models with maximum likelihood model estimation procedures.

Table 1 shows descriptive statistics for population size, economic activity, TSI and exposure data used in this study. A breakdown of the U.S. population in 2000, by census tracts and counties according to geographic census division, is provided in Table A1 in the Appendix. Census tracts, as a geographic unit, are distributed disproportionately among urban counties; they are administrative units designed for the purpose of counting the population and conducting sample surveys of the population’s characteristics.

The contrast in the distribution in social and economic characteristics between tracts and counties illustrates that most census tracts are in urban areas while counties are predominantly rural. In spite of the skewed geographic distribution of the population in census tracts, there remains variation of economic activity both within and among urban and rural counties.

There is extensive public health literature focusing on environmental inequality and environmental justice issues associated with the disproportionate impact of environmental contaminants and air pollutants [5,8,9,37,41,43,46,47,48,49,50,51,52,53,54,55,56]. Neighborhood characteristics relating to SED have been associated with differential exposure to air pollution, as well as related to neighborhood disparities in a range of health outcomes [2,5,14,42,47,49,57,58,59,60,61,62,63,64,65,66,67,68,69]. Neighborhood disparities in SED have been linked as both confounders and effect modifiers with health outcomes related to exposure to PM2.5 and ozone, though little research has been conducted relating to exposure to HAPs [37,60,70,71]. Prior epidemiologic research on SED and air pollution has tended to focus on urban places, with little attention to framing SED and chemical exposure in relation to both rural and urban America, perhaps due to greater availability of data for major cities [62]. Less clear is whether people of lower socioeconomic position were differentially exposed to HAPs across the spectrum of urban and rural places. Prior research has shown SED to be key social epidemiology variables and where there are significant gradients that explain variation in chemical exposure and health outcomes.

The substantive findings in this research are significant in that the SED differential exposure model directly relates increased HAP exposure to a social gradient of neighborhood SED, independent of the regional context of urbanization. The models suggests that there are serious incremental exposure hazards that accrue to progressive levels of socioeconomic deprivation in the United States, independent of urbanization, when we conceptualize urbanization in terms of population size and economic activity. Generalizing from the findings, the risk doubles in magnitude between the 80th percentile and 95th percentile, and is 2.6 greater at the 99th percentile of SED.

The other contribution of this research relates to cumulative risk assessment methods. It is generally recognized that the inability to quantify risks using comparative metrics accounting for nonchemical stressors—such as socioeconomic deprivation—has limited risk assessors’ abilities to assess cumulative impacts on a consistent, reliable basis, across different locations and time periods and populations. This research illustrates a population health model that enables quantification of the interaction between non-chemical stressors and chemical exposure, including measurement of exposure is modified in a dose-response curve related to SED.

There are two major threats to the validity of these findings. First, the underlying exposure models in the NATA assessments may have misclassified HAPs exposure. Certain regions of the country may be “undercounted” in terms of point source surveillance of facilities. Additionally, the population exposure model may have flaws. Among the most important uncertainties underlying these modeling results are the use of emission estimates from multiple sources (e.g., state- or industry-submitted, computed by EPA from activity data and emission factors, or developed by EPA test programs). According to the U.S. EPA, these estimates differ across geographic regions and source categories, and may vary substantially in quality. Additionally, the exposure model which is applied by the U.S. EPA nationally employs inputs and assumptions that may not be fully representative of particular local areas and the nature of their emissions and concentrations. Among additional caveats for the use of this data is that it projected exposure estimates for the median individual within each census tract; some individuals may have substantially higher or lower exposures than the median individual. The risk estimates are also limited to consideration of inhalation exposure but do not include emissions from indoor sources of air toxics. In addition, people receive substantial additional exposures to pollutants such as mercury, and dioxins that have bioaccumulated in food. The second source of bias is from “built-in” design effects that stem from the administrative organization and long-form sampling of census among tracts which would influence the underlying standard errors of the variables used in the SED measures.

On the other hand, these types of non-random measurement error do not appear to be of the type that would result in a design effect that would confound NATA’s exposure estimates with composite measures of population characteristics of census tracts. On the other side of the argument, there are good reasons for expecting that the magnitude of exposure in inner cities is undercounted given the NATA methodology and what is acknowledged to be excluded.

Given the above limitations, the results of this study demonstrate that HAP exposure was found to be related to the urban concentration of population and economic activity as well as the level of neighborhood socioeconomic deprivation. The size of a city matters as it pertains to both HAP exposure and prevalence of SED. We found relatively strong unadjusted bivariate relationships between HAPs and measures of population density, economic activity, and SED for all three measures of HAP risks in both years examined. SED clearly has two faces in the United States—urban and rural—and there is support for the notion that there are large rural areas with low levels of economic activity, high SED and low HAPs. Multivariate statistical analysis of the geographical distribution of HAPs presents a pattern of exposure hazard that is clearly associated with county population size, the county economy, and localized population density and SED within respective counties.

Differential chemical exposure among vulnerable populations raises questions about whether we would also find differential response to this exposure. The reference concentration value approach to risk assessment of chemical exposure assumes an underlying uniformity in human physiological response to toxic exposure [35,72,73,74,75]. Two aspects of this situation are worth emphasizing: first, there is clear evidence of differential chemical exposure among higher SED populations; second, the risk metrics that provide the basis for current standards as to what is safe may understate the health risks of HAPs for vulnerable populations. The role and relationship of nonchemical stressors, such as extreme poverty or socioeconomic deprivation, to multiple chemical exposures therefore represents a key issue in advancing cumulative risk assessment (CRA) and addressing environmental justice issues [35,39]. Social epidemiology is particularly suited to study differential exposure and differential susceptibility related to the interaction of chemical and nonchemical stressors in cumulative exposure and risk assessment.