Among this population of U.S. trucking industry employees, ambient residential air pollution exposures (PM10, SO2, NO2) 1985 through 2000, as well as PM2.5 exposures in 2000, were associated with increased all-cause mortality. These associations were attenuated for PM10, PM2.5, and SO2 in multipollutant models. Overall, results were stronger when we restricted the cohort to those individuals whose jobs allowed them to return home each evening. For cause-specific mortality, elevated HRs were observed for lung cancer, cardiovascular disease, and respiratory disease with both NO2 and SO2. Importantly, these exposures were not confounded by occupational exposures measured by job title.
Numerous chronic exposure studies have observed increases in all-cause mortality with increases in PM exposures (1
). Although direct comparisons to all studies are not appropriate due to the measurement of different particulate size-fractions, our results for PM10
are elevated compared with two studies measuring PM10
(Electric Power Research Institute [EPRI]–Washington Veteran's Cohort Study and Adventist Health Study of Smog [ASHMOG]) (9
) and quite similar to those from a recent analysis in the Nurses' Health Study (16
). Expressing our results in units of 10 μg/m3
(the most common unit in studies of PM), the association of PM10
and all-cause mortality in our study was 1.07 (95% CI, 1.02–1.13) in the full cohort and 1.17 (95% CI, 1.09–1.25) after excluding long-haul drivers. In general, long-term cohort studies in which PM2.5
, as opposed to PM10
, was examined as the size-fraction have observed higher estimates of all-cause mortality risk. Expressing our results in units of 10 μg/m3
, the association of PM2.5
in the full cohort was 1.10 (95% CI, 1.02–1.18) and 1.15 (95% CI, 1.05–1.27) after excluding long-haul drivers, which are of a similar magnitude to the other studies of long-term exposures.
Our findings for all-cause mortality and ambient NO2
exposures are also consistent with those from other studies. For all-cause mortality, we observed HRs of 1.10 (95% CI, 1.06–1.15), expressed in units of 10-ppb increase in NO2
, in the full cohort, and 1.19 (95% CI, 1.13–1.26) in the cohort excluding long-haul drivers. Expressing results from previous studies in the same units, an HR of 1.14 (95% CI, 0.87–1.49) was observed for men in the AHSMOG study (8
), 1.04 (95% CI, 0.97–1.13) in the EPRI–Washington Veteran's Study (12
), 1.03 (95% CI, 1.00–1.05) in the Dutch Netherlands Cohort Study on Diet and Cancer (NLCS) study (18
), 1.16 (95% CI, 1.06–1.26) in a cohort of Norwegian men (37
), and 1.23 (95% CI, 1.02–1.47) in a cohort of German women (38
For cause-specific analyses, our elevations in cardiovascular (HR = 1.09; 95% CI, 1.01–1.17 for the whole cohort; HR = 1.14; 95% CI, 1.03–1.25 excluding long-haul drivers) and respiratory mortality (HR = 1.07; 95% CI, 0.91–1.27 for the whole cohort; HR = 1.26; 95% CI, 1.01–1.56 excluding long-haul drivers) expressed for a 10-ppb increase in NO2
are lower than those observed in most other studies. In a study of men living in Norway, HRs were 1.32 (95% CI, 1.12–1.57) and 1.16 (95% CI, 1.06–1.26) for respiratory and ischemic heart disease mortality, respectively, for each 10-ppb increase in NOx
, which includes NO2
). An HR of 2.84 (95% CI, 1.62–4.99) for cardiopulmonary mortality was observed for each 10-ppb increase in NO2
in a cohort of women in Germany (38
). In the PAARC study, a 10-ppb increase in NO2
was associated with an HR of 1.57 (95% CI, 1.08–2.29) for cardiopulmonary mortality (15
). The HRs in the AHSMOG study for a 10-ppb increase in NO2
were 1.01 (95% CI, 0.93–1.09) for cardiopulmonary mortality in men (9
Fewer studies have examined the effects of long-term exposures to SO2
. Most long-term studies haven't reported elevated risks (9
) with all-cause mortality. Positive effects have been observed in a study in Great Britain (39
), in a recent reanalysis of the American Cancer Society cohort (40
), and in many short-term studies with cardiovascular deaths and/or hospital admissions (41
One of the unique features of this study was our ability to model the mortality effects of exposures to multiple pollutants. In our multipollutant models, adverse effects were mainly seen with NO2
exposures, but were quite attenuated for PM10
. As in previous studies (18
), these findings suggest that traffic (the primary source of NO2
) is an important source of exposure. Sources of PM10
include traffic and other combustion processes that contribute to the fine fraction (PM2.5
). However, grinding, windblown dust, and agricultural activities are also contributors, especially in the coarse fraction (PM10–2.5
, however, is primarily formed through electricity generation by power plants and fossil fuel combustion from heating oil and some mobile sources. Therefore, our multipollutant analysis suggests that traffic and other sources of fossil fuel combustion are important pollution sources that result in greater overall, lung cancer, cardiovascular, and respiratory disease mortality in this cohort.
Our study has several important limitations. First, we do not have information on other risk factors for mortality, such as cigarette smoking, body mass index (BMI), medication use, high cholesterol or blood pressure diagnoses, or existing comorbidities. In many previous studies of ambient air pollution these have not been shown to be important confounders of the air pollution–mortality association (1
); however, there is likely some residual confounding if they are also associated with pollution. The homogeneity of the cohort reduces the likelihood of potential confounding by socioeconomic status. In supplemental analyses (see
Appendix in the online supplement) we used information on smoking and BMI from a questionnaire (43
) sent to a sample of current employees and recent retirees from the same companies as the cohort members to examine the associations between these potential confounders and the 1985 through 2000 average pollution metrics used in this study. There was no association of BMI with the pollution measures. Current smoking, however, was associated with small increases in PM10
. Given that current smoking is an important predictor of many of the causes of mortality in our study, there is a possibility that positive confounding may be occurring in our analyses. In an analysis of the potential bias in our observed estimates due to unmeasured positive confounding by current cigarette smoking (Table E3), we determined that the potential bias did not fully explain our single-pollutant results for NO2
with all-cause mortality.
A second limitation is that exposure is based on exposure model predictions at the last known home address of the cohort members. Using predicted values of pollution may lead to confidence intervals that are too small, because we did not incorporate the model errors into our CIs (44
). Additionally, although in surveys of this cohort the average time living in the current residence is 17 years, we cannot be certain that we have the correct home address for all participants in all years. This would add to the nondifferential misclassification of exposures and may help to explain some of the nonsignificant results we observe. Only 81% of the cohort was successfully geocoded to the street level. For those members of the cohort geocoded to a ZIP code centroid, the predicted air pollution levels would be a poorer proxy for the levels actually experienced outside the home. However, in sensitivity analyses conducted in just those individuals geocoded to the street level, the conclusions were not different than those from the whole cohort (Table E5).
Another weakness of this study is the differences in the temporality of our exposure measures. Our exposure metrics for PM10, NO2, and SO2 cover the full period of cohort follow-up, 1985 through 2000, whereas our exposure to PM2.5 is based solely on annual exposures from 2000. Unfortunately, the U.S. Environmental Protection Agency did not start wide-scale monitoring of PM2.5 until the year 1999, providing only 1 year of data during the period of cohort follow-up. However, in this cohort we observed a high level of correlation (Spearman correlation coefficient, 0.63) between the annual values of PM10 and the overall long-term average 1985 through 2000, so this may introduce only minor exposure misclassification.
Finally, we used death certificates to determine our outcomes of interest. For many of the outcomes this may lead to some misclassification in the cause-specific analyses, because some outcomes may not always be appropriately coded. For example, death certificates have been shown to underestimate the true number of workers with severe COPD at death. In the Tucson Epidemiologic Study of Obstructive Airways Disease, 25% of deaths with clinically documented moderate to severe obstructive lung disease were identified using underlying cause of death only, whereas 81% had COPD noted as either underlying or contributing cause on the death certificate (46
). Again, this would likely lead to nondifferential misclassification. However, although we control for region in our analyses, there is still a potential for differential misclassification if physicians in more heavily polluted areas within a region code causes of death differently than in less polluted areas.
Although other studies have controlled for occupational exposures (1
), our study is the first to assess the effects of multiple air pollutants on mortality with fine control for occupation within workers from a single industry. Overall, our results are consistent with the large body of air pollution literature in general population studies that indicate that there are distinct effects of ambient air pollution exposures on health.