Ischemic Heart Disease Mortality and Long-Term Exposure to Source-Related Components of U.S. Fine Particle Air Pollution

1Numerous epidemiologic studies have documented associations between long-term exposure to fine particulate matter mass ≤ 2.5 μm (PM2.5) air pollution and increased mortality in urban populations (e.g.,;;;;;;;, ). This association is notably robust in the United States for cardiovascular disease mortality, and especially death from ischemic heart disease (IHD; associated with a reduction of blood supply to the heart, potentially leading to heart attack), as found in prior analyses of the American Cancer Society (ACS) Cancer Prevention Study-II (CPS-II) cohort. Based on such studies, the U.S. EPA has attributed nearly 90% of the economic valuation of human health benefits derived from the U.S. Clean Air Act to reductions in PM2.5, and nearly 1 in 5 U.S. IHD deaths are associated with PM2.5 exposure. Globally, it has been estimated by the World Health Organization (WHO) that roughly 3 million people die each year as a result of outdoor ambient particulate matter air pollution exposures, indicating that air pollution is one of the world’s largest single environmental health risks, with PM2.5 being estimated to account for some 9.4% [95% confidence interval (CI) 6.6–11.8] of all IHD globally.

2Past research has focused primarily on PM2.5 mass concentration, so the types and sources of particles most responsible for these adverse health associations are not known, limiting our ability to address this global cardiovascular disease threat. Thus, the National Academy of Sciences (NAS) and the WHO have placed a high priority on determining which constituents and components of the PM2.5 mass are most responsible for these reported health effects. As noted by the, this would “facilitate targeted abatement policies and more effective control measures to reduce the burden of disease due to air pollution.”

3The present analysis addresses this need by extending the follow-up of one of the largest of the PM2.5–mortality cohort studies that has linked individual risk factor and ambient air pollution with vital status data, while expanding the scope to consider both PM2.5 source-specific components and trace element constituent exposures. Contributions of this analysis include a) expansion of the follow-up from 16 to 22 years, increasing the number of deaths considered by more than one-third; b) substantially expanded PM2.5 air pollution exposure data, including new PM2.5 composition data; and c) consideration of source-apportioned PM2.5 mass to investigate which pollution sources are most important to PM2.5–IHD mortality associations in this cohort.

4The study population cohort was drawn from the ACS CPS-II, a prospective mortality study of approximately 1.2 million adults. Participants were enrolled by ACS volunteers in the fall of 1982, and resided in all 50 states, the District of Columbia, and Puerto Rico. Enrollment was restricted to persons at least 30 years of age, in households including someone at least 45 years of age, from whom the ACS obtained informed consent. Participants completed surveys that included questions about age, sex, weight, height, smoking history, alcohol use, occupational exposures, diet, education, marital status, and other characteristics. This secondary analysis of the ACS CPS-II data set was approved by the Ottawa Hospital Research Ethics Board, Canada, and by the Institutional Review Board at the New York University School of Medicine.

5Deaths were ascertained through personal inquiries (e.g., direct contact with participants by volunteers) through September 1988, and subsequently via the National Death Index through 31 December 2004. Mortality from IHD was studied because this was the category found most associated with PM2.5 exposure in past analyses of this cohort (e.g.,; ). More than 99% of known deaths were assigned a cause using the International Classification of Diseases, 9th and 10th Revision (ICD-9 codes 410–414; ICD-10 codes I20–I25). The analytic cohort included 445,860 participants having questionnaire, known vital status through 2004, and contextual census data, residing in 100 U.S. metropolitan areas within the contiguous United States where the required air pollution data were available, with 34,408 IHD deaths (of a total of 157,572 deaths from all causes) occurring during follow-up. Seventy-six of the metropolitan study areas were included in previously published ACS studies (e.g.,, ).

6Participants were assigned a metropolitan area of residence based on their enrollment address and three-digit ZIP code area. Complete data were available for 100 metropolitan statistical areas (MSAs) (Figure 1). The mean concentrations of PM2.5 mass and trace constituents were compiled for 2000–2005 from the Health Effects Institute (HEI) Atmospheric and Environmental Research (AER) database, derived from the U.S. EPA Air Quality System that archived Chemical Species Network (CSN) and gaseous criteria pollutant data. The present analysis focuses on further investigations of the PM2.5–IHD association, so gaseous air pollutants, evaluated in past analyses, are not considered here, except in the derivation of the source factors, as discussed below. These PM2.5 constituent data were analyzed to derive estimates of source apportioned PM2.5 mass exposure concentrations using the absolute principal component analysis (APCA) PM2.5 source apportionment method. Because this process results in orthogonal source components, the source impacts developed have the advantage that they are derived to be as independent of one another as possible. This method involved a) a factor analysis of the trace constituents; b) identification of source-related factors (based on key tracers in each component); c) adjustment of factor scores into absolute scores; and d) a regression of the PM2.5 mass data on the source-related components, yielding apportionments of PM2.5 mass to each source-related factor.

7These PM2.5 mass and speciation data have been previously characterized on a nationwide basis. Trace elements considered were arsenic (As), calcium (Ca), chlorine (Cl), iron (Fe), lead (Pb), manganese (Mn), nickel (Ni), selenium (Se), vanadium (V), silicon (Si), zinc (Zn), potassium (K), sodium (Na), elemental carbon (EC), organic carbon (OC), and sulfur (S).

8In the source apportionment analysis of these trace constituent and mass data, the major U.S. PM2.5 source categories derived (and their key source-identifier elemental tracers) were metals industry (Pb, Zn); soil particles (Ca, Si); motor vehicle traffic (EC); steel industry (Fe, Mn); coal combustion (As, Se); oil combustion (V, Ni); salt particles (Na, Cl); and biomass burning (K, OC). Although these various tracer elements are individually emitted by multiple sources, when they appear together they have been found to be useful for identifying specific source category impacts (e.g., both As and Se for coal). Sulfur (S) is considered individually here, but it is a general marker for fossil fuel sources that could have different predominant origins in different places (e.g., oil or diesel combustion in California vs. coal combustion in the Eastern United States), so it is not as uniquely associated with any one single source category, and was therefore not included in the source apportionment model. Of traffic emissions, EC derives predominantly from diesel vehicles, whereas nitrogen dioxide (NO2), also included as a traffic tracer in this source apportionment, is emitted by both diesel- and gasoline-powered vehicles. Nationwide spatial plots of the source-related PM2.5 contributions to total PM2.5 mass (hereafter referred to as source impacts) have been found to support the source-component interpretations in this analysis: Ubiquitous sources, such as traffic and soil, were found to be spread across the nation, while more unique sources were not (e.g., steel and metals processing highest in select industrialized cities, coal highest in the Ohio River Valley region, biomass burning highest in the U.S. Northwest, and residual oil combustion highest in the northeastern United States and major seaport cities). To provide directly comparable results with past nationwide analyses of this cohort (e.g.,;, ), individual-site elemental and source apportioned PM2.5 exposures were averaged over the concentration values for all monitors in each MSA for the health effects analyses (ranging from one to four PM2.5 speciation monitoring stations per MSA) over the period 2000–2005, with a majority (53%) of participants residing in an MSA with multiple sites.

9Standard and multilevel random-effects Cox proportional hazard (RE-CPH) regression models were applied to estimate mortality in relation to each of the individual indices of PM2.5 air pollution. The most extensively adjusted models estimated IHD mortality hazard ratios (HRs) for each PM2.5 mass, constituent, and source exposure using Cox proportional hazards (CPH) models that also included spatial random effects and control for contextual socioeconomic variables. We also evaluated models with random effects but not contextual variables, models with contextual variables but not random effects, and fixed-effects models adjusted for individual-level variables only. The time axis used in the models was survival time from enrollment date. Survival times were censored at time of death or end of follow-up. To facilitate comparisons with past analyses, we modeled individual PM2.5 constituents and sources as continuous variables and report HRs for interquartile range (IQR) increases (equal to the difference between the 25th and 75th percentile) in the estimated mean concentration of each pollutant during 2000–2005. We also report HRs per 1-μg/m3 increases in each source to facilitate comparisons among sources according to each source’s mortality contribution per unit mass. In addition, we conducted sensitivity analyses of simultaneous CPH models, including both a source-related PM2.5 mass and the remaining PM2.5 mass (i.e., PM2.5-source–related PM2.5), for each identified PM2.5 source category to estimate the source-specific PM2.5 IHD impacts, after adjusting for the impact of all other PM2.5. Natural spline plots, showing concentration–response curves for key source impact categories, were also prepared, using the statistical package R. In R, a model was defined using the function ns of the R “splines” package, in combination with the GAMEPHIT procedure, by including in the model a term the form ns(x, df = k), where df is the number of degrees of freedom. All survival models were stratified by 1-year age categories, sex, and race (non-Hispanic white vs. other), assigning each sex–race–age category its own baseline hazard.

10A total of 42 variables were included to control for individual characteristics that might confound the air pollution–mortality association. The variables included were individual-level risk factors from the CPS-II enrollment questionnaire, based upon previous ACS air pollution analyses model specifications (e.g., ), as listed in Table 1. These included active smoking and former smoking (yes/no for each; and cigarettes/day, duration, and initiation < 18 years among active and former smokers, respectively); ever cigar/pipe smoker (yes/no); passive smoke exposure (hours/day); possible workplace exposure to PM (yes/no); occupational dirtiness index (7 categories with an indicator of exposure to an increasingly dirty job or not, plus a missing data category); marital status (married, single, or other); education (< high school, high school, or > high school); body mass index (BMI) and BMI2; consumption of beer, wine, and other alcohol (yes, no, or missing, with separate variables for each type of alcohol); quintile of dietary fat consumption; and quintile of combined dietary vegetable, fruit, and fiber consumption. For the occupational dirtiness index, six enrollment question variables were used to record past occupational exposures to asbestos, chemicals/acids/solvents, coal/stone dust, coal tar/pitch/asphalt, diesel engine exhaust, or formaldehyde, and these were then collapsed into one variable for the statistical analyses by identifying a participant as occupationally exposed if a “1” for “yes” appeared for any one of the six variables.

11Data on ecologic risk factors representing social and economic variables were obtained from the U.S. Census and other secondary sources at the MSA level. Six ecologic covariates were obtained from the 1990 U.S. Census [median household income, percentage of persons > 16 years of age who were unemployed, percentage of adults with a postsecondary education, Gini coefficient of income inequality (ranging from 0 to 1, with 0 indicating an equal distribution of income, and 1 indicating that one person has all the income), and percent black and percent Hispanic population]. Models also incorporated an MSA-level random effects adjustment, taking into account residual mortality variation within communities.

12Key cohort characteristics are tabulated versus PM2.5 quintile in Table 1. This is a nationwide cohort, but considers only adults > 30 years of age at enrollment. It was recruited by volunteers, rather than via random sampling. The average age of the cohort was 56.6 years at enrollment, and the group has more advanced education and a higher percentage of white participants than the general population of the United States. The PM2.5 constituent and source apportionment exposure data are summarized in Table 2. These data indicate that traffic-related PM2.5 has the highest mean concentration averaged across the national cohort, whereas other, more regionalized sources (e.g., coal combustion, oil combustion, biomass burning, and soil) have somewhat lower national average PM2.5 concentration contributions.

13Figure 2 presents the PM2.5–IHD mortality HRs (and 95% CIs) from the RE-CPH model (including individual and contextual variables) for the various PM2.5 constituents (i.e., trace elements) and the source components (i.e., PM2.5 mass attributed to each source category). The results of all models considered (i.e., CPH with and without random effects, with and without contextual variables) are presented in Table 3. As shown in Figure 2 and Table 3, IHD mortality was statistically significantly associated (p < 0.05) with PM2.5 mass and a variety of PM2.5 elements across most models (e.g., Se, S, As, Cl, Pb, EC, Fe), but consistently not with others (e.g., Si, K, Mn, OC). Estimates based on reduced models (without random effects and/or adjustment for contextual variables) were generally more often statistically significant than the more extensively adjusted model estimates. However, there were some exceptions, for example, the HR central estimate for S based on the fully adjusted model (1.06; 95% CI: 1.02, 1.11) was statistically significant and also larger than the HR adjusted for individual-level variables only (1.02; 95% CI: 1.00, 1.03) (Table 3).

14The source-component results shown in Figure 2 and Table 3 are concordant with the strong associations found for coal combustion’s key tracers (Se and As), with consistently statistically significant IHD mortality–PM2.5 source category associations across models being found with coal combustion PM2.5 (Table 3). Particulate S, which is predominantly formed from sulfur dioxide emissions from various fossil fuel combustion sources, but especially from coal power plant emissions in the United States during the period of this study, is also significant in most models. Metals-related factors and mortality were also statistically significant in some models, though the average exposure is very low across the nation (mean, 0.2 μg/m3), consistent with the significant IHD association with Pb, a metal strongly associated with this source factor. Coal combustion associated PM2.5 yielded an IHD HR of 1.03 (95% CI: 1.01, 1.05, per coal combustion PM2.5 IQR = 0.64 μg/m3), whereas PM2.5 mass yielded an IHD HR of 1.03 (95% CI: 1.00, 1.06 per PM2.5 mass IQR = 3.1 μg/m3).

15When considered on a per 10-μg/m3 PM2.5 mass basis, the CI of the estimate for PM2.5 HR of 1.09 (95% CI: 1.00, 1.20) overlaps prior PM2.5–IHD estimates for another subset of this same cohort (e.g., HR = 1.18; 95% CI: 1.14, 1.21 per 10-μg/m3 PM2.5, from ). When the HRs of the source concentrations (which are much smaller than the total mass) are similarly viewed on a per 1-μg/m3 comparable basis, the coal combustion PM2.5 had an IHD HR of 1.05 (95% CI: 1.02, 1.08) per 1-μg/m3 coal PM2.5, versus a 1-μg/m3 PM2.5 HR of 1.01 (95% CI: 1.00, 1.02). This indicates that the IHD risk estimate for coal PM2.5 is roughly five times higher than that for PM2.5 mass in general, per microgram/cubic meter. Although the overall traffic-related PM2.5 component and the more gasoline engine traffic–related OC were not consistently statistically significantly associated with IHD mortality across the various models, a key tracer of diesel traffic (EC) was positively and statistically significantly associated with IHD in three of the four models considered (p < 0.05), and nearly so in the most extensively specified RE-CPH model (HR = 1.03 per 0.3 μg/m3 EC; 95% CI: 1.0, 1.06) (Table 3). However, neither wind-blown crustal/soil–related nor biomass combustion–related PM2.5 mass was statistically significantly associated with IHD mortality in any model considered.

16Results of the sensitivity analysis, with both the individual source–related components of PM2.5 and the “all other” PM2.5 mass for each included simultaneously, are presented in Figure 3. In this analysis, estimates from models of source-related components of PM2.5 (presented in Figure 2 and Table 3) were adjusted for a variable representing the remaining PM2.5 mass (i.e., after subtracting the mass of each respective source-related component from total PM2.5 mass). These results indicate that, even after adjustment for the other PM2.5 mass, the coal-related PM2.5 HR is the source component least changed, and most clearly seen to be still statistically significant (IHD HR = 1.05; 95% CI: 1.02, 1.08 per 1-μg/m3 coal PM2.5). In addition, of all the sources considered here, removal of coal-related PM2.5 from the PM2.5 mass reduces the overall PM2.5 mass HR estimate the most. Indeed, once coal-related PM2.5 is removed, the “other PM2.5” HR drops to nearly 1.0 and is clearly non-significant (IHD HR = 1.01; 95% CI: 0.93, 1.10 per 10-μg/m3 PM2.5). In these models, diesel traffic–related (IHD HR = 1.01; 95% CI: 1.00, 1.03 per 1-μg/m3 PM2.5) and metals-related (IHD HR = 1.09; 95% CI: 1.00, 1.20 per 1-μg/m3 PM2.5) PM2.5 are the only other sources that approach statistical significance. Other sources of PM2.5 are seen from these results to be less important contributors to the overall PM2.5 mass effect because they are clearly nonsignificant, and their removal from PM2.5 has a more minor reduction in the PM2.5 mass HR estimate.

17The correlations of the source-related components, as calculated from the individual trace element observations, were derived to be orthogonal (independent) from one another, but averaging the daily values over time and metropolitan areas can introduce intercorrelations among the source impacts (e.g., if two source-specific PM2.5 impacts are highest in the same city). The intercorrelations of the various source impacts across the metropolitan areas (not shown) indicate that the highest spatial correlations are found between steel impacts and metals concentrations (r = 0.67), which is likely attributable to the fact that both are especially high in Birmingham, Alabama, in this data set. The salt- and traffic-related PM2.5 components also show a relatively high spatial intercorrelation with each other (r = 0.44), which is likely caused by the common use of salt on roads in winter.

18Natural spline plots of HRs as a function of source-specific mass concentration (Figure 4) are shown for the sources that were most consistently statistically significant across the various models examined (coal and traffic), versus those sources most consistently nonsignificant across the models examined (soil and biomass burning). These are visually seen to be roughly comparable with the Figure 2 RE-CPH HR effect estimates, with coal combustion having the steepest slope (i.e., greatest effect per microgram/cubic meter) presented, and a slope consistent with a 5% increase in IHD risk per 1-μg/m3 increase in PM2.5 mass concentration from coal combustion. Also, similar to the RE-CPH results, crustal/soil and biomass burning PM2.5 are indicated by the CIs provided in these natural spline plots to have nonsignificant associations with IHD mortality.

19Although PM2.5 pollution from most industrial and fossil fuel combustion categories had estimates of association with IHD mortality > 1.0, coal combustion PM2.5 and its key emission tracer elements (i.e., Se and As) were most strongly and robustly associated with IHD mortality across all the various model specifications considered (e.g., of these source and tracer groupings, they are the only ones that were clearly all statistically significant across the four models considered). In addition, particulate S, known to result largely from fossil fuel combustion and especially coal combustion, was also found to be associated. Notably, the association between IHD mortality per 1-μg/m3 increase in coal combustion PM2.5 was larger than the effect estimate for a 1-μg/m3 increase in total PM2.5 mass. In contrast, IHD mortality was not associated with crustal soil PM2.5, biomass PM2.5, nor with the elemental tracers associated with these source categories. However, our findings do not rule out effects of sources that were not identified by the source apportionment analysis, or effects of region-specific sources and exposures.

20The associations between ambient PM2.5 air pollution and IHD mortality in our study cohort are consistent with previous reports, though few studies have estimated associations with long-term PM2.5 elemental constituents and/or source-specific PM2.5 mass components. A previous cross-sectional study of long-term PM2.5 sources and all-cause mortality rates in the United States during 1980 similarly reported mortality associations with metropolitan area-wide sulfates and coal combustion-related particle exposures. Previous analyses of the ACS CPS-II cohort and the Harvard Six Cities Study cohort also showed associations between sulfates and both all-cause and cardiopulmonary mortality. Long-term exposures to EC and S were significantly associated with total mortality in a previous analysis of the ACS CPS-II cohort. A meta-analysis of data from 19 European cohorts indicated that long-term exposure to PM2.5 sulfur was associated with natural-cause mortality, and that this association was robust to adjustment for other pollutants and PM2.5 mass. Sulfate alone, as discussed by, is an unlikely causal factor for mortality or morbidity from a toxicological perspective, so it may be a contributor to the toxicity of the PM2.5 mixture or serve as a marker for a certain type or source of particulate pollution that needs identification. A study of deaths among U.S. veterans employed data from the U.S. EPA CSN fine particle speciation network, finding mortality associations with nitrates, EC, Ni, V, and traffic density. PM2.5 constituents derived from fossil fuel combustion and constituents of crustal origin were more strongly associated than other PM2.5 constituents with mortality in a population of 45,000 California teachers. A follow-up of this cohort found statistically significant (p < 0.05) associations of IHD with PM2.5 mass, nitrate, EC, copper, and secondary organics and the sources gas- and diesel-fueled vehicles, meat cooking, and high-sulfur fuel combustion in California. However, individual trace element metals were not associated with CVD mortality in a meta-analysis of 19 European cohorts. Collectively, these past studies are largely consistent with a finding that the PM2.5 association with mortality varies with its elemental composition, but most past studies have not looked at the PM2.5 constituent issue from a collective source-specific PM2.5 component perspective, which is more readily translatable into air quality policy.

21Although effects of short-term (day-to-day) exposures to PM2.5 components and sources may differ from effects of long-term exposures, findings from studies of short-term exposures to trace elements and sources have been generally consistent with findings from the present study. For example, estimated that a 10-μg/m3 increase in PM2.5 from mobile sources was associated with a 3.4% (95% CI: 1.7, 5.2%) increase in daily mortality in six U.S. cities, and that a 10-μg/m3 increase in PM2.5 from coal combustion sources was associated with a 1.1% increase (95% CI: 0.3, 2.0%) in mortality. However, daily mortality was not associated with PM2.5 crustal soil particles. The authors concluded that combustion particles in the fine fraction from mobile and coal combustion sources, but not fine crustal particles, were associated with increased daily mortality. Using Bayesian hierarchical modeling, investigated seasonal and temporal variation in PM2.5 composition and the risk of total, cardiovascular, and respiratory hospital admissions for persons ≥ 65 years of age in 202 U.S. counties during 1999 through 2005, finding that mortality effect estimates for particulate matter mass were higher in seasons and counties with higher PM2.5 Ni content. applied a Poisson generalized linear model to estimate source-specific PM2.5 mass relative risks at lags 0–4 days for total nonaccidental, cardiovascular, and cardiorespiratory mortality adjusting for weather, seasonal/temporal trends, and day-of-week for Washington, DC, during 1988–1997. They found that the secondary sulfate component had the largest (and most consistently statistically significant across investigative teams) percent excess cardiovascular mortality risk estimates, with primary coal-related PM2.5 being similarly significantly associated, and risk estimates for traffic-related PM2.5, though significant in some cases, were more variable, and soil-related PM showed smaller effect size estimates. These various studies have largely provided results consistent with the findings of this new research, indicating that the mortality risk from PM2.5 can vary by mass composition and source.

22Overall, in this new research we found that the PM2.5 source-component associations with IHD mortality were consistent with results from individual key trace constituents used to identify pollution sources. Model results indicate that, in this data set, long-term exposures to PM2.5 from coal combustion-related PM2.5 (a source component most correlated with As and Se) and EC (a tracer element from combustion sources, especially diesel-powered motor vehicles in urban areas) were among those most robustly associated with increased risk of IHD mortality across the models considered. The fact that IHD mortality risk was more consistently statistically significantly associated with the EC tracer across the models than was overall traffic-related PM2.5 may suggest that the diesel portion of the traffic component increases risk to IHD more than PM2.5 from other vehicles. Soil and biomass burning (and their tracers, Si and Ca, and K and OC, respectively) were least often positively significantly associated with IHD mortality across models.

23This study has two important strengths. This is the first large nationwide U.S. study to comprehensively examine the characteristics of PM2.5 elemental constituents and source-specific components in relation to IHD in a prospective cohort study in which we were able to consider detailed individual-level subject risk factor information. Only recently have the availability of data from the CSN and the development of advanced source apportionment methods made studies of the constituents and sources of PM2.5 possible. Second, the large population and number of cities considered, combined with the increased long-term follow-up, provide enhanced power to assess increases in risk related to relatively small changes in PM2.5 mass and component exposures.

24Despite the strengths of this new analysis, there are several attributes of this study that may have limited its ability to detect all effects. One was the use of central-site monitors as an indicator of individual exposure, which, even though averaged across available sites in each MSA, could potentially introduce nondifferential measurement error across the various sources. Traffic markers such as EC can display significant long-term average concentration spatial variation on scales of 50–500 m within cities, whereas elevated pollution sources (e.g., power plants with tall stacks) may be more spatially homogeneous. This could potentially bias the effect estimates of localized sources, such as traffic, toward the null hypothesis of no increase in risk. Another concern might be that not all sources varied as much spatially (e.g., between cities) as others, which may limit the ability of the analysis to identify the impact of those more spatially homogeneous, and less spatially variable, sources, especially the steel industry and salt aerosols (having the smallest overall Standard Deviations in Table 2). However, coal combustion PM2.5 exposure had a smaller absolute standard deviation across the United States than two of the other sources (wood burning and traffic), but still had the most significant association, suggesting that this factor was not a major determinant of source significance in this analysis. Spatial confounding may also remain, despite adjustment for contextual variables defined at the MSA level and the use of random effects models to account for uncontrolled regional variation.

25Potential misclassification of the cause of death and changes in participant characteristics (such as smoking status) during the follow-up are also sources of potential misclassification. In addition, exposures were classified using PM2.5 data from 2000 through 2005, whereas follow-up extended from 1982 through 2004; consequently, there could have been some exposure misclassification if concentrations of PM2.5 and PM2.5 constituents changed over time, and associations with PM2.5 sources in 2000–2005 may actually reflect earlier exposures to these sources of PM2.5. Previous analyses of this cohort have shown PM2.5 levels at different time periods to be correlated, with lower levels at the end of the follow-up period, but indicating that areas with the highest exposures in the 1980s were still the most highly exposed in 2000. Estimated concentrations of many of the trace element constituents (including Si, Se, Zn, V, and Mn) and source components (soil, coal, metals, traffic, and oil) were correlated with estimates for the early 1980s, at the start of cohort follow-up, supporting the use of 2000–2005 CSN data to estimate long-term exposures. Although residential mobility is generally relatively low in older adults, exposure misclassification could also have occurred if ACS cohort participants moved to a different area after baseline.

26A lack of information on medications (e.g., statins, anti-hypertensives) could also be a potential confounding factor if they are spatially correlated with air pollution exposures, because statins have been suggested as possibly protective against PM2.5 health effects (e.g., ). Because there may have been regional differences in drug-prescribing patterns and adherence, and although adjustment has been made for random effects and for individual- and city-level socioeconomic status indicators, they may not have been adequate to address this potential source of confounding.

27These results also raise the causality question as to whether it is biologically plausible that combustion-related particles, and particularly coal combustion PM2.5, would have a greater cardiovascular toxicity, per unit mass of exposure, than other particles, especially versus those of biomass or soil origins. There is evidence that oxidant stress is involved in the cardiovascular impacts of PM2.5, and it is also known that trace metals, such as those contained in high concentration in fossil fuel combustion–related particles, can cause oxidative stress, likely contributing to atherosclerosis progression. In addition, New York University’s mouse inhalation studies at five U.S. sites showed substantial variations in aortic plaque progression by geographic region that were consistent with the regional variation in annual IHD mortality in the ACS-II cohort, with both the human and mouse responses being primarily attributable to the coal combustion source category. Moreover, fossil fuel sources, notably coal combustion, are usually higher in sulfur content than biomass particles, resulting in sulfates, a secondary aerosol component that has been associated with increased mortality. The potential interaction of trace metals and sulfur in increasing the toxicity of PM2.5 mass has been posited in the recent, which states that “Sulphates could play a role in the capacity of PM2.5 to drive pulmonary inflammation via either increasing metal mobilisation from PM, and hence increasing the ability to drive oxidative stress leading to inflammation…. Inflammation in the lung could then initiate changes in blood clotting, and/or activation of macrophages in atherosclerotic plaques increasing their instability.”

28Although evidence on acute respiratory effects of biomass PM has been well documented, evidence of cardiovascular disease impacts by particulate matter from biomass is less certain. Indeed, a recent cohort study of chronic biomass cooking pollution exposures in Bangladesh found a significant increase in respiratory mortality, but not in cardiac deaths. Although the biological mechanism(s) responsible for the apparently greater long-term cardiovascular toxicity of fossil fuel combustion particles versus others not enriched in transition metals and sulfur (e.g., soil, biomass) are not yet known, there are plausible toxicological pathways consistent with these constituents in fossil fuel PM2.5, and especially coal PM2.5, being more strongly associated with IHD mortality. Although our study is a U.S.-based analysis of older adults, these findings have potentially broad policy implications: As the developing world deals with an increasing disease burden from air pollution, reductions in fossil fuel combustion, especially coal combustion, may well be the most efficacious policy approach to reduce air pollution’s global human health burden.

29Recent risk assessments have been conducted based on exposure to concentrations of PM2.5 mass decomposed into its pollution sources (e.g., ). As discussed by, these past assessments have generally assumed that the change in risk associated with PM2.5 mass concentration is independent of source. Our new study provides evidence that there may, in fact, exist sources where the risk per microgram/cubic meter appears to be either greater than (e.g., coal) or less than (e.g., biomass) the effect estimate derived in past epidemiological studies from the overall mass concentration. Although these new findings are supported by biological plausibility (as discussed above), further research is needed to further test the generalizability of the associations found (e.g., in younger populations).

30In summary, we estimated associations between IHD mortality and long-term exposures to elemental constituents and source components of PM2.5 in a U.S. cohort of older adults with follow-up from 1982 through 2004. Exposure to fossil fuel combustion–related PM2.5 air pollution, especially from coal combustion, but also from diesel traffic–related emissions, was significantly associated with IHD mortality in our study population. This finding, combined with past results and other knowledge of PM2.5 constituent toxicity, suggests that the greatest cardiovascular mortality benefits from PM2.5 control may be achieved via reductions in fossil fuel PM2.5 exposures, especially in PM2.5 from coal combustion. This research has implications for both clean air policy and climate change mitigation policy because it provides an indication as to which of the various sources of EC, carbon dioxide, and particulate matter, if reduced, might return the greatest public health benefits.