In this study we found associations between IHD mortality and PM2.5, PM10, NOx, and ozone. The associations with PM2.5 and NOx were modestly greater among never-smokers, as were associations with cardiovascular disease mortality as a whole. Incident stroke (combining fatal and nonfatal events) was associated with PM10 and with PM2.5, particularly among women who were postmenopausal at baseline, whereas analyses focusing only on nonfatal incident stroke (i.e., hospitalizations only) identified associations with multiple pollutants, especially PM10.
This investigation represents one of the largest prospective air pollution studies undertaken to date (1
). Unlike most prior studies, we developed individualized estimates of long-term exposure to PM2.5
and other pollutants at the participants’ residences, including those who relocated during the study period. The low prevalence of active smoking in this cohort (5% at baseline), in combination with previously collected data on household second-hand smoke exposure, allowed for a potentially more clear-cut examination of the impact of air pollution exposures than in other investigations with substantial proportions of active smokers (e.g., American Cancer Society Cancer Prevention Study [ACS CPS] II, 22% active smokers; Harvard Six Cities [HSC] study, 33–40% active smokers). Moreover, unlike most other cohorts, CTS participants share a relative uniformity of occupational status, precluding the need for statistical adjustment for toxic industrial exposures based on potentially problematic job exposure matrices. Thus, the CTS design and population characteristics included an individualized exposure assessment and diminished the potential for confounding and effect modification by nonpollutant variables.
Several other long-term air pollution studies have found associations of PM2.5
with increased risk of all-cause mortality, and greater risks with either cardiopulmonary or cardiovascular disease. We found no associations with all-cause mortality in any analysis except for NOx and SO2
, and these results were based on few events. Although our results are different from those of several other U.S. cohorts, they are generally consistent with the Dutch study by Beelen and colleagues, who reported no significant increases in all-cause or cardiovascular mortality associated with measured PM2.5
). However, those investigators reported slightly increased risks for these outcomes in relation to traffic intensity on the nearest road. Because our analysis of NOx was limited to residences within either 3- or 5-km buffers, the elevated HRs that we observed for this pollutant may represent effects of local traffic emissions as well as transported products of combustion. Our results are also consistent with an analysis of the ACS cohort in the New York City region, which also found no effects for all-cause mortality, but elevated risks from PM2.5
exposure for IHD mortality that are of the same magnitude as those reported here (13
Several prior California-specific studies of air pollutant exposure and mortality have produced mixed results. Enstrom found essentially no relationship between exposure to fine PM and all-cause mortality among elderly California participants in the ACS CPS I from 1973 to 2002, although the relative risk (RR) for the 20,210 women was slightly elevated (RR, 1.027; 95% CI, 1.005–1.050) (14
). Chen and colleagues reported in 2005 that a 10-μg/m3
increase in PM2.5
was associated with an RR of 1.42 (95% CI, 1.11–1.81) for fatal CHD in a cohort of 2,090 women participating in the Adventist Health Study (15
). In an analysis of the ACS CPS II data for 22,905 Los Angeles residents, Jerrett and colleagues (6
) reported that a 10-μg/m3
increase in PM2.5
was associated with HRs of 1.11 (95% CI, 0.99–1.25) for all-cause and 1.25 (95% CI, 0.99–1.59) for IHD mortality, using a model with 44 individual-level and parsimonious contextual covariates. In a more recent analysis of these data using land use regression, these estimates were only slightly greater (13
Although we found no relationship of PM2.5
with all-cause mortality, the association between a 10-μg/m3
increase in PM2.5
and increased risk of fatal IHD (HR, 1.20; 95% CI, 1.02–1.41) was of similar magnitude to that reported by Jerrett and colleagues (6
). In an analysis of the national ACS CPS II cohort from 1983 to 1998, average PM2.5
was also associated with IHD mortality (HR, 1.18; 95% CI, 1.14–1.23) (3
). Other, more recent studies of PM and mortality from CHD in women have produced higher risk estimates. For instance, in the WHI, the risk of CHD death associated with a 10-μg/m3
increase in estimated PM2.5
was more than doubled (HR, 2.21; 95% CI, 1.17–4.16) (5
). Using modeled PM2.5
data to estimate 10-year exposures to participants in the Nurses’ Health Study, Puett and colleagues reported a similarly elevated risk of death from CHD (HR, 2.02; 95% CI, 1.07–1.54), although this estimate was markedly attenuated when the investigators used annual average data from fixed-site monitors (HR, 1.47; 95% CI, 0.73–2.99) (8
). The differences between our estimates and those of these other investigations may be related to differences in the underlying health status of the study populations; the numbers of cases (there were 773 IHD cases in our main PM2.5
analysis, far more than in these other recent studies); methods of estimating exposure; particle composition and relative toxicity; and measurement and control of potential confounders. When exposures were assigned using only the average PM2.5
concentrations estimated for the participants during the year preceding follow-up, our results remained consistent with those from the time-dependent exposure models ().
Because of the restrictions we placed on spatial interpolations for CO, NO2
, NOx, and SO2
to reduce the potential for exposure misclassification, there were far fewer participants and events in all models involving these pollutants than in those for PM2.5
, and ozone. Moreover, these gases are subject to considerable intraurban variability, depending largely on local traffic patterns; this variability may have been underestimated by IDW interpolation. NO2
levels may vary significantly over a distance of a few hundred meters (16
). Therefore, our NOx results should be interpreted with caution. Last, as is true with all air pollution epidemiology studies, differential measurement error among the pollutants may have affected both the magnitude and the precision of the effect estimates.
Only one prospective investigation of long-term exposure to PM2.5
has reported associations with incident MI and stroke (5
). The investigators followed nearly 66,000 participants in the WHI observational study without a history of cardiovascular disease for a median of 6 years, using as the exposure metric a 1-year average of PM2.5
values measured at the monitor closest to their residence at baseline. They reported HRs of 1.06 (95% CI, 0.85–1.34) for incident MI and 1.28 (95% CI, 1.02–1.61) for stroke per 10-μg/m3
increase in PM2.5
). Our results are similar in that we found associations of PM2.5
with stroke, but not MI. In our incidence analysis that included only hospital admissions, stroke was also positively associated with both PM2.5
(HR, 1.10; 95% CI, 0.93–1.31) and PM10
(HR, 1.09; 95% CI, 1.01–1.17).
Puett and colleagues modeled monthly PM2.5
levels at the residences of 66,250 women living in the Northeast and Midwest of the United States in the Nurses’ Health Study from 1992 to 2002 (8
). They found significantly elevated HRs for all-cause mortality, but the risk of incident CHD (including nonfatal MI) was not significantly elevated overall (CHD: HR, 1.11; 95% CI, 0.79–1.55; nonfatal MI: HR, 0.73; 95% CI, 0.48–1.12). Our findings are similar to those of Puett and colleagues with respect to the lack of association with incident MI and CHD. Although PM10
were highly correlated in our data set, PM2.5
showed somewhat stronger associations with IHD mortality and stroke. This may be due in part to the likelihood of greater exposure misclassification for PM10
than for PM2.5
, as the former exhibits greater spatial heterogeneity.
Using year-round and summer ozone levels, we found positive associations with IHD mortality, but not overall cardiovascular mortality. In two-pollutant models, however, there was no association of IHD mortality with ozone, whereas the HRs for PM2.5
remained elevated, suggesting that the results for ozone were most likely explicable by its positive correlation with particulate matter. In several prior cohort studies, when ozone has been included in the models of long-term exposure, no associations with cardiopulmonary mortality have been observed (1
). However, Jerrett and colleagues reported slightly elevated significant positive associations of ozone with cardiovascular mortality in their analysis of the ACS CPS II data, which diminished to null results in two-pollutant models with PM2.5
Jerrett and colleagues also reported associations of respiratory mortality with long-term ozone exposure nationwide, but when they stratified on geographic area, they found no association of ozone with respiratory mortality in Southern California (7
). Both measures of ozone in our study suggested associations with nonmalignant respiratory mortality ( and , and Table E10), which were of comparable magnitude to the ozone-associated relative risk for nonmalignant respiratory mortality among women in the Adventist Health Study (17
). Our results also suggest associations between both particulate matter metrics and nonmalignant respiratory disease mortality, especially among nonsmokers, although the latter HRs were based on relatively few events (203 deaths for PM10
and 191 for PM2.5
; Table E8). Nonetheless, the estimated HRs were of comparable magnitude to that identified for deaths due to lower respiratory infections in the ACS CPS II analysis among never-smokers (HR, 1.20; 95% CI, 1.02–1.41) (3
Our analyses of MI and stroke were limited to women who did not report a history of either of these events on the baseline questionnaire. Although some of these participants may have experienced silent events, it is unlikely that such misclassification of disease would be differentially distributed by pollutant exposure. Also, these outcomes were measured here only as hospitalizations or deaths, which could have resulted in incomplete ascertainment. Nevertheless, there is no reason to think that such unrecorded events would have biased the results in a differential manner.
Acute events such as stroke may be attributable to both short-term as well as long-term pollutant exposures (18
). However, it is unlikely that the effects reported here were due only to short-term exposures, as the magnitudes of increased risks identified in this investigation (19% for stroke among postmenopausal women) far exceed those reported in time-series investigations (e.g., 3% for stroke mortality [21
]). Without daily data for this entire time period we could not disaggregate short-term from long-term pollutant impacts. However, experimental evidence and other epidemiological studies of subclinical disease support the proposition that these long-term exposures were associated with incident disease in the CTS cohort.
In a rodent model of atherosclerotic disease, chronic exposure to low levels of PM2.5
(6-mo study average, 15 μg/m3
) was associated with enhanced progression of disease, increased vasomotor tone, and vascular inflammation (22
). In humans, progression of atherosclerotic disease can be observed subclinically as increases in carotid arterial intima medial thickness, which has been reported cross-sectionally in association with estimated residential annual mean concentrations of PM2.5
) and, more recently, in pooled data from five clinical studies conducted in the Los Angeles basin (24
). Thus, although such subclinical outcomes could not be examined in the CTS, these mechanisms underscore the biological plausibility of our finding that long-term exposure to particulate matter was associated with incident stroke (25
This study provides evidence that long-term exposures to PM2.5, PM10, and NOx were associated with increased risks for IHD mortality. The increased risk of IHD mortality associated with long-term ozone exposure was most likely explicable by its correlation with particulate matter, whereas that for NOx was based on relatively small numbers of observations. These data also suggest associations of long-term ozone and particulate matter exposure with mortality from nonmalignant respiratory disease. That both measures of PM were associated with incident stroke provides support for the notion that these pollutants may play an etiological role in the development of circulatory disease.