Associations of Mortality with Long-Term Exposures to Fine and Ultrafine Particles, Species and Sources: Results from the California Teachers Study Cohort

PDF Version (1.4 MB)

Introduction

Several cohort studies have reported associations of long-term exposure to fine particles (PM_2.5; particulate matter ≤ 2.5 μm in diameter) with cardiovascular mortality (Hoek et al. 2013; Laden et al. 2006; Lipsett et al. 2011; Pope et al. 2002). Because PM_2.5 is a heterogeneous mix of particle sizes and chemistry and is generated from multiple sources, the specific constituents and sources of concern have not been fully elucidated. Until recently, among the constituents of PM_2.5, long-term exposures (i.e., ≥ 1 year) for cohort studies have been generated only for sulfates and black carbon (Dockery et al. 1993; Pope et al. 1995; Smith et al. 2009). In addition, because of the difficulty in measuring exposure, there has been little focus to date on the health effects of long-term exposure to ultrafine (UF) particles (particles ≤ 0.1 μm in diameter).

Epidemiologic analysis of the effects of particulate matter constituents is hindered by their spatial heterogeneity and the reliance on a few fixed-site monitors to represent exposures in large metropolitan areas. For example, for PM₁₀ (particles ≤ 10 μm in diameter), many metropolitan areas have only a small proportion of their total population within 15 km of a monitor, such as New York, New York (3.5%), Detroit, Michigan (23%), Boston, Massachusetts (39%), Seattle, Washington (31%), and Philadelphia, Pennsylvania (35%) [U.S. Environmental Protection Agency (EPA) 2009]. The proportion for those > 65 years of age, a well-documented susceptible subgroup, who are within 15 km are only slightly higher: New York (4%), Detroit (27%), Boston (41%), Seattle (32%), and Philadelphia (38%). Although coverage for PM_2.5 is much higher given its spatial homogeneity, its constituents are known to be spatially variable and often very localized (Kim et al. 2005). The exposure misclassification will be even greater for measurement of mass and constituents of UF particles given their spatial heterogeneity (Sakurai et al. 2003; Sioutas et al. 2005). Some cohort studies have made use of land use regression (LUR) models to estimate PM_2.5 or nitrogen dioxide (Beelen et al. 2014) at finer spatial scales, but LUR models for particle sources or species are not widely available.

In a previous study, we examined the relation between mortality and long-term exposure to constituents of PM_2.5 using data from the California Teachers Study (CTS) cohort (Ostro et al. 2010). Started in 1995, the CTS is a prospective study of > 130,000 current and former female teachers and administrators identified through the State Teachers Retirement System. Because of limited data on particle species, this earlier report relied on PM_2.5 data collected and further analyzed by the U.S. EPA at eight fixed-site monitors as part of the Speciation Trends Network (U.S. EPA 2008). The 24-hr averaged measurements were usually obtained on an every third- or sixth-day basis. To minimize exposure misclassification, catchment buffer areas of 8 and 30 km were drawn around each monitor. The 30-km buffer is likely too large to capture exposure contrasts of many of the species, while the 8-km buffer significantly reduced the sample size, resulting in more unstable estimates and reduced statistical power. Although these buffers were an improvement over studies using a single or multiple monitors to represent exposure over large metropolitan areas, they may not sufficiently measure concentrations of many of the PM constituents, such as elemental carbon (EC) and transition metals that are known to exhibit high spatial variance. Specifically, Hu et al. (2014a, 2014b) reported significant bias for several species of fine and UF particles when comparing the central-site monitor readings applied to the entire metropolitan area population versus our estimated population-weighted concentrations. The latter are derived as the product (both at the 4-km grid scale) of the population and our model-based estimates of the pollutants.

Model estimates are highly correlated (r > 0.8) with observations at the monitoring locations. But for the seven major California Metropolitan Statistical Areas that had available data, the estimated population-weighted concentrations for EC were generally lower than central-site monitor predictions, with a maximum bias of –50% in Los Angeles, California, and an average bias of –33%. Although measurement and model predictions were in good agreement at the monitor locations, the bias was introduced by spatial variability around the monitor.

For the present study, we combined data from the CTS with newly developed exposure data generated from the UCD/CIT (University of California Davis/California Institute of Technology) Source Oriented Chemical Transport model. The UCD/CIT chemical transport model uses calculated meteorological fields and emissions estimates for different sources to predict airborne particulate matter concentrations. Particulate matter emissions are assigned a size and composition distribution based on measurements in source-testing experiments. The source-identity of all particulate matter emissions is retained through the simulated atmosphere. In the present study, ground-level mass concentrations for 50 PM constituents were estimated over the major population regions in California at a 4-km resolution for the period of 2000–2007. For many species of fine and UF particles, model predictions are highly correlated with measured values, particularly for longer averaging times (> 2 weeks). For example, correlations were > 0.8 for comparisons between annual modeled and measured concentrations of 10 different PM_2.5 components for five of the seven metropolitan regions with available monitoring data (Hu et al. 2014b).

Below, we report our findings of an analysis of the associations of long-term exposure to 19 constituents and sources of both PM_2.5 and UF particles on mortality from all natural causes, cardiovascular disease, ischemic heart disease (IHD), and pulmonary disease.

Methods

Data. The CTS is a prospective study of 133,479 current and former female teachers and administrators who completed baseline questionnaires mailed to them in 1995 to investigate the incidence of breast cancer in public school teachers and administrators, as described in detail by Bernstein et al. (2002). Subsequent questionnaires were mailed to CTS participants in 1997 and 2000. The design and ongoing follow-up of the CTS cohort is a multi-institutional collaboration involving researchers with diverse and complementary areas of expertise. Record linkage is conducted annually to mortality files administered by the California Department of Public Health. In addition, residential addresses of each CTS participant were updated annually for the mailing of newsletters. The mean age of CTS participants at enrollment was 54 years, with 90% between 30 and 80 years. The cohort is multi-ethnic but primarily non-Hispanic white (86.7%) and born in the United States (93.6%). For this study, we used cohort follow-up data from January 2001 through July 2007. Women < 30 years of age at the start of the study were excluded to focus on mid-life and older women. Use of data on human subjects in the main CTS cohort study was reviewed and initially approved by the California Committee for the Protection of Human Subjects, Health and Human Services Agency, and by the institutional review boards (IRB) for each participating institution in June 1995 and annually thereafter. Informed consent was obtained upon entry into the cohort. Analysis for this manuscript was approved in August 2013 by the IRB of the Cancer Prevention Institute of California, the center of one of the principal investigators.

Health outcomes. In this analysis, we focused on associations between long-term exposures and mortality. Deaths were assigned codes based on the International Classification of Diseases, 10th Revision (ICD-10) for the following outcomes: all-cause deaths excluding those with an external cause (A00–R99), cardiovascular deaths (I00–I99), IHD deaths (I20–I25), and pulmonary deaths (C34, J00–J98). Person-days at risk were calculated as the number of days starting from 1 January 2001 until the earliest of three dates: a) the date of death; b) a move out of California for at least 4 months; or c) 31 July 2007, the end of follow-up for this analysis. If a woman moved out of state for < 4 months, exposures during that time were not included in the calculations of the long-term average. Women who died from a cause other than the outcome of interest during the follow-up period were censored at the time of their deaths.

Air pollution exposure estimates. The UCD/CIT model was used to estimate ground-level concentrations of 50 PM constituents over the major population regions in California using a 4-km grid resolution for the period from 2000 through 2007 (Hu et al. 2014a, 2014b). A sensitivity analysis conducted at 250-m resolution over Oakland, California (Joe et al. 2014), indicated that 4-km resolution captures 55–70% of concentration variability within the urban area.

Using the extensive emissions inventory in California, the model calculations track the mass and number concentrations of the PM constituents in particle diameters ranging from 0.01 to 10 μm through calculations that describe emissions, transport, diffusion, deposition, coagulation, gas- and particle-phase chemistry, and gas-to-particle conversion (Hu et al. 2014b). The model solves the coupled set of differential equations that describe how atmospheric processes change pollutant concentrations in regularly spaced atmospheric grid cells. Thus, the predicted exposure concentrations primarily reflect the balance between emissions inventories of fresh particles and meteorological fields that drive dispersion and chemical reaction.

Model predictions were saved at hourly time resolution and averaged to longer times as needed. Predicted concentrations were evaluated against ambient measurements at all available locations and times. PM_2.5 mass predictions had a mean fractional bias within ± 0.3 (meeting accepted performance criteria) at 52 sites of the total 66 sites across California after correcting for bias in the dust emissions because many studies have shown that dust emissions in the current emission inventory are overestimated (Hu et al. 2015). Good correlations between predictions and measurements (r > 0.8) were demonstrated for many of the PM_2.5 and UF species at most of the monitoring stations, particularly for the monthly, seasonal, and annual averages. For example, monthly PM_2.5 nitrates were correlated with measurements with r = 0.76 (15 sites), monthly PM_2.5 EC was correlated with measurements with r = 0.94 (8 sites), and monthly PM_2.5 concentrations of potassium, chromium, zinc, iron, titanium, arsenic, calcium, manganese, and strontium were correlated with measurements with r ≥ 0.8 (5 sites of a total of 8). For EC in the UF range, the correlation was above r = 0.9 for 117 available measurements made at 13 locations during nine intensive field campaigns that each lasted several weeks (Hu et al. 2014b). The quality of the model predictions summarized above reflects the accuracy of the emissions inventories that have been refined over three decades in California, the development of reactive chemical transport models that include important aerosol transformation mechanisms, and the development of prognostic meteorological models that allow for long simulations of historical meteorology.

Coarse particle predictions (2.5 μm < Dp < 10 μm) have undergone only preliminary comparisons to measurements and were not used for exposure estimates in the present study. Likewise, UF number concentrations were not used because our modeling did not include nucleation, the process by which particles are formed directly from gas molecules, which would greatly impact this parameter. UF mass concentrations are highly correlated with particle surface area (Kuwayama et al. 2013) and serve as a good metric for the potential exposure to UF particles. The measured correlation between UF mass and particle surface was 0.97 in Sacramento, a typical city in central California. For many of the fine and UF species, Hu et al. (2014a) observed strong spatial variability within metropolitan areas (Figure 1).

On the basis of previous studies (Ostro et al. 2007, 2010; Peng et al. 2009; Zanobetti et al. 2009), we chose to examine a subset of the available constituents. Additionally, some constituents were eliminated given their high intercorrelation or low concentrations.

Thus, for each particle size, we analyzed the following constituents: copper (Cu), iron (Fe), manganese (Mn), nitrate, EC, organic carbon (OC), “other” species (i.e., mineral dusts and constituents not measured), “other” metals (those besides Cu, Fe, and Mn that were explicitly resolved), and secondary organic aerosol (SOA). SOA formation was simulated with the mechanism in the U.S. EPA’s Community Multi-scale Air Quality (CMAQ) model version 4.7 (Carlton et al. 2010). SOAs were divided into anthropogenic (SOA_ant: derived from long-chain alkanes, xylenes, toluenes, and benzene and their oligomers) and biogenic (SOA_bio: derived from isoprenes, monoterpenes, and sesiquiterpenes and their oligomers). Nitrate was not estimated for the UF size fraction. Once the constituent concentrations were obtained, Positive Matrix Factorization was used to develop source profiles. Estimates were provided for sources of primary aerosols including on-road gasoline, off-road gasoline, on-road diesel, off-road diesel, wood smoke, meat cooking, high-sulfur fuel combustion (including distillate oil, marine vessel fuel, aircraft jet fuel, liquid and solid waste fuel), and “other anthropogenic.”

Ultimately, the exposure metrics were combined with the updated addresses. Monthly individual exposure estimates were developed through spatial linkage of the geocoded residential addresses. All residences within a given grid in a given month were assigned the modeled pollutant value for that grid for that period. The average long-term pollution exposure for a participant was obtained by calculating the mean of her monthly averages. At the time of each death, the long-term average for each individual remaining in the cohort was recalculated, allowing comparison between the decedent’s long-term average exposure and those of the members remaining in the risk set.

Covariates. The individual-level covariates included as explanatory variables in the regression models were based on previous results from air pollution studies for this cohort (Lipsett et al. 2011). Specifically, the covariates included twenty individual-level covariates (a total of 58 terms): age (divided into 2-year categories between 30 and 79 years of age, 3-year categories between 80 and 88 years, and one category for women ≥ 89 years); race [non-Hispanic white, other (African American, Hispanic, Asian, Pacific Islander, and Native American), or unknown]; marital status (married/living with partner, not married, and unknown); smoking status (never, former, and current smokers) and pack-years of smoking (continuous variable for former and current smokers); secondhand smoke exposure (none, household exposure, unknown); body mass index (BMI) (16–19, 20–24, 25–29, 30–39, 40–55 kg/m²); lifetime physical activity (tertiles, unknown); alcohol consumption [beer (no/yes/unknown), wine (no/yes/unknown), liquor (no/yes/unknown)]; average daily dietary intake of fat (tertiles, unknown), fiber (tertiles, unknown), and calories (tertiles, unknown); menopausal status and hormone replacement therapy use combined (premenopausal, peri/postmenopausal and no HT use, peri/postmenopausal and past HT use, peri/postmenopausal and current use of estrogen, peri/postmenopausal and current use of estrogen plus progestin, and unknown menopausal status or HT use); family history of myocardial infarction (yes/no) or stroke (yes/no); and use of blood pressure medication (low, medium, high, unknown) or aspirin (low, medium, high, unknown). Data on all individual-level variables except marital status (which was assessed in the 2000 questionnaire) were obtained from the baseline questionnaire.

Statistical methods. We fitted Cox proportional hazards models to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for associations between each pollutant and the outcomes of interest. We examined each pollutant with a separate regression model adjusted for the covariates described above. The Cox model was stratified by age and race/ethnicity. To ensure that we would be examining associations with chronic rather than acute exposures, study exposures began in January 2000, the cohort follow-up began in January 2001, and both continued until July 2007. Two additional sensitivity analyses were conducted. First, we reran the models after including six census-derived contextual (neighborhood) variables including income (median household income), income inequality (percent below poverty level), education (percent with college degree), population size, racial composition (percent white, percent black, percent Hispanic), and unemployment (percent unemployed). These variables were derived from the 2000 census at the block-group level based on the subject’s residence at the time of the baseline questionnaire. These variables represent social, economic, and environmental settings at a group level that may be associated with disease outcomes at the individual level. As such, they may provide additional control for residual confounding. The second sensitivity analysis involved two-pollutant models for a selected set of constituents for the outcome IHD mortality. Specifically, we took the constituent of PM_2.5 and UF particles with the highest HR and ran additional regressions that included each of the other constituents in the same particle size. The HR and CIs are presented for a change in their respective interquartile ranges (IQRs) unless otherwise noted. Statistical significance was based on a p-value < 0.05, and model goodness-of-fit was based on the Akaike information criterion (AIC). The analysis was conducted using PHREG in SAS software (SAS Institute Inc., Cary, NC).

Results

Of the 133,479 women who completed a baseline questionnaire, we excluded 21,302 with no pollution data (of whom 14,670 had a lack of information on residential addresses and 6,632 lived in areas for which exposure estimates were not available); 1,363 women who had died or moved before the start of follow-up; 406 who were < 30 years of age in January 2001; 4,684 who had unknown or outlier BMI; 3,609 who were missing smoking data; 14 who were excluded because they consented to be included only in breast cancer studies; and 217 who had < 6 months of pollution values during January 2000 through December 2000. The final total was 101,884 women eligible for the study. The average length of follow-up was 6.3 years, with total person-years of 642,269. A total of 6,285 deaths occurred during the follow-up from January 2001 through July 2007; of these, 2,400 were due to cardiovascular diseases, 1,085 were due to IHD, and 929 were due to pulmonary diseases. As indicated in Table 1, the average age of eligible cohort members at the start of follow-up was 57 years, 86% of these women were non-Hispanic white, and 5% were current smokers. Table 2 summarizes the mean and distributions of the concentrations of PM_2.5 and UF constituents used in the analysis. For example, the mean of PM_2.5 was 17.9 μg/m³ with OC (3.9 μg/m³) and nitrate (3.7 μg/m³) the largest constituents. For UFs, the mean was 1.3 μg/m³ with OC the largest contributor at 0.9 μg/m³.

A majority of the species were moderately to highly correlated (r = 0.5–0.8) (see Supplemental Material, Tables S1 and S2). PM_2.5 nitrates had correlations of 0.55, 0.43, 0.65, and 0.84 with EC, OC, Cu, and SOA_ant, respectively. For UFs, EC had correlations of 0.67, 0.19, and 0.40 with OC, Cu, and SO_ant, respectively. The average inter-constituent correlation for PM_2.5 was 0.59 and for UFs was 0.64. The implications of these correlations are described in “Discussion.”

In the Cox proportional hazards regression analysis for PM_2.5, the only statistically significant association (p < 0.05) observed between sources and constituents and all-cause mortality was for the source of high-sulfur fuel combustion (HR = 1.03; 95% CI: 1.01, 1.05 for a change in its IQR), and there were no statistically significant associations with pulmonary disease mortality (see Supplemental Material, Table S3). For cardiovascular disease mortality, statistically significant associations were demonstrated only for nitrate (HR = 1.10; 95% CI: 1.02, 1.18) and high-sulfur fuel combustion (HR = 1.05; 95% CI: 1.02, 1.09). Associations with p-values < 0.10 were observed for PM_2.5 mass (HR = 1.05; 95% CI: 0.99, 1.12) and SOA_ant (HR = 1.06; 95% CI: 0.99, 1.13) (see Supplemental Material, Table S3). As summarized in Table 3 and Figure 2, however, there were many statistically significant associations with IHD mortality. Among the constituents, nitrate (HR = 1.28; 95% CI: 1.16, 1.42) and SOA_ant (HR = 1.23; 95% CI: 1.11, 1.36) were the most statistically significant, and both had higher HRs and fit the data slightly better, based on the (lower) AIC, than that of PM_2.5 mass (HR = 1.18; 95% CI: 1.08, 1.30). Among the emission sources, we found statistically significant associations between IHD and all four of the vehicle sources, meat cooking, and high-sulfur content fuel combustion.

For UFs, no statistically significant associations were observed for either all-cause or pulmonary mortality (see Supplemental Material, Table S4). For cardiovascular mortality, significant associations were noted for Cu (HR = 1.03; 95% CI: 1.00, 1.05), and the sources of high-sulfur fuel combustion (HR = 1.04; 95% CI: 1.01, 1.07). However, many statistically significant associations were again demonstrated for IHD mortality (Table 3). Among the species, this includes Cu, Fe, EC, OC, other compounds and metals, SOA_ant, and SOA_bio. The largest estimated risks were for SOA_ant (HR = 1.25; 95% CI: 1.13, 1.39), EC (HR = 1.15; 95% CI: 1.06, 1.26), and other metals (HR = 1.13; 95% CI: 1.05, 1.21), each of which had lower p-values and slightly better fitting models based on AIC than did UF mass (HR = 1.10; 95% CI: 1.02, 1.18). Many of the other constituents also had better model fits than PM_2.5. Among the sources, associations were seen for both on and off-road diesel and gasoline, meat cooking, high-sulfur fuel combustion, and other anthropogenic sources.

The analysis of IHD mortality showed that although PM_2.5 mass had a lower p-value than UF mass, UF mass and each of the UF constituents provided a better fit and had a lower p-value than their corresponding PM_2.5 constituent (except for Mn, for which there was no statistical significance for either particle size).

In our sensitivity analysis, we found that adding the six contextual variables to the model did not quantitatively alter any of the results (HR or p-value) except in one case where PM_2.5 SOA_bio became nonsignificant (data not shown). We also examined two-pollutant models with the PM_2.5 constituent with the largest effect estimate for IHD (PM_2.5 nitrate) in a regression with each of the other PM_2.5 constituents. Likewise, we examined two-pollutant models for UF (SOA_ant) with each of the other UF constituents (Tables 4 and 5). For the two-pollutant models with PM_2.5 nitrate, the HRs for nitrate were basically unchanged and none of the other PM_2.5 constituents, including mass, were statistically significant. For UFs SOA_ant, the HR was again basically unchanged and only one other constituent, Cu, was also statistically significantly related to IHD mortality.

Discussion

Our analysis of long-term exposure to the mass and constituents of PM_2.5 and UF particles revealed several statistically significant associations with all-cause, cardiovascular, and IHD mortality. For PM_2.5, high-sulfur content fuel combustion was associated with all three end points, and nitrates were associated with cardiovascular and IHD mortality. Several other constituents reached statistical significance with IHD mortality including PM_2.5 mass, Cu, EC, and the SOAs, as well as the sources including gas- and diesel-fueled vehicles, meat cooking, and high-sulfur fuel combustion. Among the PM_2.5 constituents, based on their associated IQRs, nitrate had the highest HR and provided the best fit of the data. For UFs, constituents such as SOA_ant, EC, and “other” metals exhibited statistically significant associations with IHD mortality, as did all of the mobile sources and high-sulfur fuel combustion. For both PM_2.5 and UF particles, several constituents generated higher HRs based on their relevant IQRs than their associated mass measurements and in some cases (e.g., UF mass vs. SOA_ant) the differences were statistically significant based on methodology suggested by Schenker and Gentleman (2001). In addition, for all of the constituents, there were better model fits, based on AIC, for UFs than for PM_2.5.

In a previous analysis of the CTS (based on 73,489 women), exposures to PM_2.5 were estimated using data from 77 existing monitors located throughout the state (Lipsett et al. 2011). Smoothed surfaces were produced through inverse distance weighting, and grids of 250 m were created. Monthly concentrations were assigned to residents within each grid with the added constraint that participants were required to be within 30 km of a monitor. That study produced an HR for the association of PM_2.5 and IHD mortality of 1.20 (95% CI: 1.02, 1.41) for a 10-μg/m³ increase in PM_2.5. This result comports with the HR estimate (converted to 10 μg/m³ change) in the present study of 1.19 (95% CI: 1.08, 1.31). Our estimate is also similar to those for IHD mortality based on analyses of the American Cancer Society (ACS) cohort in which the HRs for a 10-μg/m³ increase in PM_2.5 were 1.18 (95% CI: 1.14, 1.23) for the United States and 1.11 (95% CI: 1.05, 1.18) for California (Jerrett et al. 2013; Pope et al. 2002). They were also comparable to those of the Harvard Six Cities Study of 1.26 (95% CI: 1.08, 1.47) (Laden et al. 2006) for a 10-μg/m³ increase in PM_2.5.

We can also compare the estimates of a few constituents of PM_2.5 with those obtained in a prior analysis of a smaller subset (n = 43,220) of the CTS (Ostro et al. 2010). In this prior analysis, we used a 30-km buffer catchment area around each of eight U.S. EPA Speciation Trends Network monitors. The HR for cardiovascular mortality associated with a 1-μg/m³ increase in nitrate in the previous study was 1.03 (95% CI: 1.01, 1.06) versus the present study estimate of 1.02 (95% CI: 1.01, 1.04). For EC, the previous study generated an HR of 1.11 (95% CI: 0.91, 1.36) for a 1-μg/m³ change compared with 1.05 (95% CI: 0.98, 1.11) in the present study. Several cohort studies have estimated the effects of EC or its correlates on cardiovascular mortality. For example, Smith et al. (2009) estimated its effects among 352,000 participants in the ACS cohort and reported a relative risk (RR) of 11% (95% CI: 3, 19) per 1 μg/m³. The estimated RR of coronary heart disease mortality associated with EC was 1.08 (95% CI: 1.04, 1.12) per 1 μg/m³ in a cohort study in Vancouver, Canada (Gan et al. 2011). In addition, the RR of cardiovascular mortality from long-term exposure to black smoke, another EC correlate which measures the light reflectance of particles was reported in cohort studies in the Netherlands and Scotland (Beelen et al. 2008; Beverland et al. 2012). Based on a conversion factor calculated by Janssen et al. (2011), the HRs were 1.04 (95% CI: 0.95, 1.12) and 1.06 (95% CI: 1.0, 1.11) per μg/m³ of EC, respectively. Finally, a recent study (Lippman et al. 2013) examined the effect of PM_2.5 components and sources using a subset of the national ACS cohort. The results of the Cox regression model for IHD were generally supportive of our findings. Among the components measured, the authors observed statistically significant associations with IHD for EC and several of the metals (e.g., iron, lead, nickel, and zinc). Nitrates were not included in the ACS study, but statistical associations were observed for sulfur, likely from the combustion of coal and residual oil, which was not included in our study. In addition, among the sources, traffic was dominant in both studies.

We did not observe any positive associations with long-term exposure to wood smoke, although associations of short-term exposure with respiratory outcomes have been reported (Naeher et al. 2007). This may be attributable to the episodic nature of the wood smoke or to possible confounding by socioeconomic status. In California, most of the population-weighted exposure occurred in relatively high-income counties, such as San Francisco, San Mateo, and Santa Clara, where greater longevity prevails.

Given its large spatial variability, assessing exposure to UF particles among participants in cohort studies has been challenging. Therefore, very few studies have measured or estimated long-term exposures to UFs at a fine enough spatial gradient to examine its impact on health. As an alternative, several studies have attempted to estimate the effects of exposure to traffic, often a major source of UFs, using metrics such as nitrogen dioxide, distance to major roadways, and/or local traffic density (Health Effects Institute 2010). In general, within the first 250 m or so of a major roadway, UFs may be highly correlated with other pollutants such as black carbon, nitrogen dioxide, and carbon monoxide. However, the relation between UFs and these other pollutants, especially away from major roadways, is not precise and the correlations may be fairly low (Sioutas et al. 2005; Zhu et al. 2008).

In contrast, several studies have estimated the effects of daily changes in UFs where only the time-varying component is needed (Atkinson et al. 2010; Forastiere et al. 2005; Peters et al. 2009). The previous studies were based on counts of UFs rather than mass, so their estimates are not directly comparable to ours. In support of these and our findings on UFs, Delfino et al. (2009) followed a panel of 60 elderly subjects with coronary artery disease and reported associations between biomarkers of inflammation and several components of UF particles, including EC and primary OC. Other animal and human studies have implicated transition metals in generating inflammation and oxidative stress (Chen and Lippmann 2009; Costa and Dreher 1997; Gurgueira et al. 2002).

Our study has both strengths and limitations. Among the strengths are the relatively large size of the cohort, the low prevalence of active smoking, and the relative similarity of occupational status and activity patterns. These factors all help to reduce residual confounding in our estimates. Second, the study population included a large number of women at risk of developing cardiovascular disease by virtue of their age and postmenopausal status. Third, because of the level of spatial detail in the pollution estimates and the information on residential history, the temporal and spatial resolution of the pollution exposure is enhanced relative to many previous cohort studies.

One limitation is that the study was restricted to women, and these women were not necessarily representative of all women. Second, only about 1,000 women were diagnosed with IHD or pulmonary mortality, which may introduce some instability in the risk estimates. Third, our estimates could be impacted by possibly correlated and unmeasured co-pollutants. Fourth, there was high intercorrelation (most between 0.5 and 0.8) among the particle constituents, different levels of uncertainty and bias in their modeled estimates, and potentially different exposure patterns. These factors could affect the estimates of their relative toxicity. The high correlations reflect a) a consistent chemical signature of multiple pollutants associated with PM emitted from major sources, b) that some elements are dominated by a small number of sources, and/or c) the similarity of certain pollutants from different sources such as gasoline and diesel vehicles. A similar range of intercorrelation among the constituents was reported by Ostro et al. (2010), which used monitored values for the same cohort as the present study, but only included eight metropolitan areas. Thus the high correlations are not simply a result of the modeling methodology. However, this feature does make it difficult to identify unique components and sources that are associated with adverse health effects. Fifth, stationary sources contribute < 15% of the PM_2.5 in California (Air Resources Board 2013), so sources such as coal burning and industrial processes and their specific constituents are not included in this study. Finally, although our exposure method had some significant enhancements over previous assessments, some misclassification will continue to exist.

Nevertheless, this study represents an innovative effort to estimate the effects of long-term exposure to the constituents of two pollutants, fine and ultrafine particles, that are ubiquitous in our environment. As such, it provides evidence of the public health impact of a subset of these constituents and helps contribute to our understanding of air pollution–related cardiovascular disease.

References

Air Resources Board. 2013. 2012 Estimated Annual Average Emissions. Sacramento, CA:California Environmental Protection Agency. Available: http://www.arb.ca.gov/ei/emsmain/emsmain​.htm [accessed 9 March 2014].

Atkinson RW, Fuller GW, Anderson HR, Harrison RM, Armstrong B. 2010. Urban ambient particle metrics and health: a time-series analysis. Epidemiology 21:501–511.

Beelen R, Hoek G, van den Brandt PA, Goldbohm RA, Fischer P, Schouten LJ, et al. 2008. Long-term effects of traffic-related air pollution on mortality in a Dutch cohort (NLCS-AIR study). Environ Health Perspect 116:196–202; doi: 10.1289/ehp.10767.

Beelen R, Raaschou-Nielsen O, Stafoggia M, Andersen ZJ, Weinmayr G, Hoffmann B, et al. 2014. Effects of long-term exposure to air pollution on natural-cause mortality: an analysis of 22 European cohorts within the multicentre ESCAPE project. Lancet 383:785–795.

Bernstein L, Allen M, Anton-Culver H, Deapen D, Horn-Ross PL, Peel D, et al. 2002. High breast cancer incidence rates among California teachers: results from the California Teachers Study (United States). Cancer Causes Control 13:625–635.

Beverland IJ, Cohen GR, Heal MR, Carder M, Yap C, Robertson C, et al. 2012. A comparison of short-term and long-term air pollution exposure associations with mortality in two cohorts in Scotland. Environ Health Perspect 120:1280–1285; doi: 10.1289/ehp.1104509.

Carlton AG, Bhave PV, Napelenok SL, Edney ED, Sarwar G, Pinder RW, et al. 2010. Model representation of secondary organic aerosol in CMAQv4.7. Environ Sci Technol 44:8553–8560.

Chen LC, Lippmann M. 2009. Effects of metals within ambient air particulate matter (PM) on human health. Inhal Toxicol 21:1–31.

Costa DL, Dreher KL. 1997. Bioavailable transition metals in particulate matter mediate cardiopulmonary injury in healthy and compromised animal models. Environ Health Perspect 105(suppl 5):1053–1060.

Delfino RJ, Staimer N, Tjoa T, Gillen DL, Polidori A, Arhami M, et al. 2009. Air pollution exposures and circulating biomarkers of effect in a susceptible population: clues to potential causal component mixtures and mechanisms. Environ Health Perspect 117:1232–1238; doi: 10.1289/ehp.0800194.

Dockery DW, Pope CA III, Xu X, Spengler JD, Ware JH, Fay ME, et al. 1993. An association between air pollution and mortality in six U.S. cities. N Engl J Med 329:1753–1759.

Forastiere F, Stafoggia M, Picciotto S, Bellander T, D’Ippoliti D, Lanki T, et al. 2005. A case-crossover analysis of out-of-hospital coronary deaths and air pollution in Rome, Italy. Am J Respir Crit Care Med 172:1549–1555.

Gan WQ, Koehoorn M, Davies HW, Demers PA, Tamburic L, Brauer M. 2011. Long-term exposure to traffic-related air pollution and the risk of coronary heart disease hospitalization and mortality. Environ Health Perspect 119:501–507; doi: 10.1289/ehp.1002511.

Gurgueira SA, Lawrence J, Coull B, Murthy GG, González-Flecha B. 2002. Rapid increases in the steady-state concentration of reactive oxygen species in the lungs and heart after particulate air pollution inhalation. Environ Health Perspect 110:749–755.

Health Effects Institute. 2010. Traffic-Related Air Pollution: A Critical Review of the Literature on Emissions, Exposure, and Health Effects. HEI Panel on the Health Effects of Traffic-Related Air Pollution. Special Report 17. Boston, MA:Health Effects Institute. Available: http://pubs.healtheffects.org/getfile.ph​p?u=553 [accessed 29 Alpril 2015].

Hoek G, Krishnan RM, Beelen R, Peters A, Ostro B, Brunekreef B, et al. 2013. Long-term air pollution exposure and cardio- respiratory mortality: a review. Environ Health 12:43; doi: 10.1186/1476-069X-12-43.

Hu J, Zhang H, Chen S, Ying Q, Wiedinmyer C, Vandenberghe F, et al. 2014a. Identifying PM_2.5 and PM_0.1 sources for epidemiological studies in California. Environ Sci Technol 48:4980–4990.

Hu J, Zhang H, Chen SH, Ying Q, Wiedinmyer F, Vandenberghe F, et al. 2014b. Predicting primary PM_2.5 and PM_0.1 trace composition for epidemiological studies in California. Environ Sci Technol 48:4971–4979.

Hu J, Zhang H, Ying Q, Chen SH, Vandenberghe F, Kleeman MJ. 2015. Long-term particulate matter modeling for health effect studies in California—part 1: model performance on temporal and spatial variations. Atmos Chem Phys 15:3445–3461; doi: 10.5194/acp-15-3445-2015.

Janssen NA, Hoek G, Simic-Lawson M, Fischer P, van Bree L, ten Brink H, et al. 2011. Black carbon as an additional indicator of the adverse health effects of airborne particles compared with PM₁₀ and PM_2.5. Environ Health Perspect 119:1691–1699; doi: 10.1289/ehp.1003369.

Jerrett M, Burnett RT, Beckerman BS, Turner MC, Krewski D, Thurston G, et al. 2013. Spatial analysis of air pollution and mortality in California. Am J Resp Crit Care Med 188:593–599.

Joe DK, Zhang H, DeNero SP, Lee HH, Chen SH, McDonald BC, et al. 2014. Implementation of a high-resolution Source-Oriented WRF/Chem model at the Port of Oakland. Atmos Environ 82:351–363.

Kim E, Hopke PK, Pinto JP, Wilson WE. 2005. Spatial variability of fine particle mass, components, and source contributions during the regional air pollution study in St. Louis. Environ Sci Technol 39:4172–4179.

Kuwayama T, Ruehl CR, Kleeman MJ. 2013. Daily trends and source apportionment of ultrafine particulate mass (PM_0.1) over an annual cycle in a typical California city. Environ Sci Technol 47:13957–13966.

Laden F, Schwartz J, Speizer FE, Dockery DW. 2006. Reduction in fine particulate air pollution and mortality: extended follow-up of the Harvard Six Cities study. Am J Respir Crit Care Med 173:667–672.

Lippmann M, Chen LC, Gordon T, Ito K, Thurston GD. 2013. National Particle Component Toxicity (NPACT) Initiative: Integrated Epidemiologic and Toxicologic Studies of the Health Effects of Particulate Matter Components. Research Report 177. Boston, MA:Health Effects Institute.

Lipsett MJ, Ostro BD, Reynolds P, Goldberg D, Hertz A, Jerrett M, et al. 2011. Long-term exposure to air pollution and cardiorespiratory disease in the California Teachers Study cohort. Am J Respir Crit Care Med 184:828–835.

Naeher LP, Brauer M, Lipsett M, Zelikoff JT, Simpson CD, Koenig JQ, et al. 2007. Woodsmoke health effects: a review. Inhal Toxicol 19:67–106.

Ostro B, Feng WY, Broadwin R, Green S, Lipsett M. 2007. The effects of components of fine particulate air pollution on mortality in California: results from CALFINE. Environ Health Perspect 115:13–19; doi: 10.1289/ehp.9281.

Ostro B, Lipsett M, Reynolds P, Goldberg D, Hertz A, Garcia C, et al. 2010. Long-term exposure to constituents of fine particulate air pollution and mortality: results from the California Teachers Study. Environ Health Perspect 118:363–369; doi: 10.1289/ehp.0901181.

Peng RD, Bell ML, Geyh AS, McDermott A, Zeger SL, Samet JM, et al. 2009. Emergency admissions for cardiovascular and respiratory diseases and the chemical composition of fine particle air pollution. Environ Health Perspect 117:957–963; doi: 10.1289/ehp.0800185.

Peters A, Breitner S, Cyrys J, Stölzel M, Pitz M, Wölke G, et al. 2009. The influence of improved air quality on mortality risks in Erfurt, Germany. Res Rep Health Eff Inst 137:5–77.

Pope CA III, Burnett RT, Thun MJ, Calle EE, Krewski D, Ito K, et al. 2002. Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. JAMA 287:1132–1141.

Pope CA III, Thun MJ, Namboodiri MM, Dockery DW, Evans JS, Speizer FE, et al. 1995. Particulate air pollution as a predictor of mortality in a prospective study of U.S. adults. Am J Respir Crit Care Med 151(3 pt 1):669–674.

Sakurai H, Tobias HJ, Park K, Zarling D, Docherty KS, Kittelson DB, et al. 2003. On-line measurements of diesel nanoparticle composition and volatility. Atmos Environ 37:1199–1210.

Schenker N, Gentleman JF. 2001. On judging the significance of differences by examining the overlap between confidence intervals. Am Stat 55:182–186.

Sioutas C, Delfino RJ, Singh M. 2005. Exposure assessment for atmospheric ultrafine particles (UFPs) and implications in epidemiologic research. Environ Health Perspect 113:947–955; doi: 10.1289/ehp.7939.

Smith KR, Jerrett M, Anderson HR, Burnett RT, Stone V, Derwent R, et al. 2009. Public health benefits of strategies to reduce greenhouse-gas emissions: health implications of short-lived greenhouse pollutants. Lancet 374:2091–2103.

U.S. EPA (U.S. Environmental Protection Agency). 2008. Chemical Speciation. Available: http://www.epa.gov/ttn/amtic/speciepg.ht​ml [accessed 19 June 2013].

U.S. EPA (U.S. Environmental Protection Agency). 2009. Integrated Science Assessment for Particulate Matter (Final Report) EPA/600/R-08/139F. Available: http://cfpub.epa.gov/ncea/cfm/recordispl​ay.cfm?deid=216546 [accessed 4 March 2013].

Zanobetti A, Franklin M, Koutrakis P, Schwartz J. 2009. Fine particulate air pollution and its components in association with cause-specific emergency admissions. Environ Health 8:58; doi: 10.1186/1476-069X-8-58.

Zhu Y, Fung DC, Kennedy N, Hinds WC, Eiguren-Fernandez A. 2008. Measurements of ultrafine particles and other vehicular pollutants inside a mobile exposure system on Los Angeles freeways. J Air Waste Manag Assoc 58:424–434.