Worldwide studies have shown the short-term effect of particulate pollution (PM) on hospital admissions and deaths from respiratory causes [1–6].

Epidemiologic studies indicate that hospitalizations for respiratory causes are strongly related to PM exposure. Several hypotheses have been advanced for possible underlying mechanisms [7]. For example, PM may impair ventilation in COPD patients by causing airway narrowing and increasing the work of breathing [8]. In addition some particles can cause epithelial cell damage, pulmonary edema, and eventually fibrosis [9].

Particles may be deposited in the extra thoracic airways (mouth, nose, larynx), in airways of the trachiobronchial regions and in the alveolar region where the gas exchange occurs [10].

The respiratory tract deposition patterns depend on particle size and distribution within the inspired air. Biologic effects may be a function also of particle number, composition and the total surface area of the particle.

Various factors have been shown to influence particle deposition, such as age, ventilation patterns and the presence of obstructive or inflammatory airway disease [11]. Higher ventilation increases total deposition, and obstructive airway disease, such as chronic bronchitis, emphysema and asthma results in increased deposition in the lower respiratory tract [12].

Retention of particles is a function of deposition site and clearance of particles, which again may be impaired in persons with COPD. Chronic effects may also arise from recurring cycles of pulmonary injury and repair [10]. Studies have shown that high doses of particles can trigger oxidative stress and the induction of inflammation, increased blood coagulation, impaired cellular defence, and modulation of the immune system [13, 14].

The effect of longer term PM exposure on survival in general cohorts has been examined in a number of studies, which mainly focused on total or cardiovascular mortality. The Harvard Six Cities prospective cohort study [15] showed an association between cardiovascular mortality and chronic exposure to air pollutants. Other studies [16, 17] such as the American Cancer Society Cancer Prevention study population, most recently[17] reported an association between long term average PM and specific causes of death, such as deaths from all cardiovascular disease plus diabetes, ischemic heart disease, and dysrhythmias, heart failure, and cardiac arrest mortality.

Another cohort in seven French cities [18] found that urban air pollution was associated with increased total mortality over 25 years, and they also observed a consistent pattern for lung cancer and cardiopulmonary causes.

While these previous studies have reported an association of air pollution with survival, and sometimes cause specific survival, these have not examined whether risk changed with annual changes in exposure.

Specifically, in most published cohort studies, the exposure was taken to be long term differences in air pollution concentrations across geographic location, which does not allow examination of whether changes in annual exposure result in changes in risk. However, a recent study in Dublin [19] suggests that decreasing air pollution concentrations by a sudden intervention (a coal ban) reduced cardiovascular and respiratory deaths the next year, with no further decrease in subsequent years.

This suggests a cohort study with PM₁₀ treated as a time varying covariate, with annual follow-up periods for exposure and survival, would be of considerable interest. In addition, if such studies are done within city, then by design, they eliminate the possibility of confounding by unmeasured covariates that vary across city or geography.

A recent re-analysis of the Harvard Six Cities Study, which extended the mortality follow-up period, used annual PM2.5 as a time-varying exposure, and found a significant association of PM2.5 with mortality, and moreover that most of the effect was with the previous two year's exposure [20]. However, in that study, variations in exposure both between city and between years within city contributed to the exposure gradient.

Another recent study investigated [21] whether year-to-year within city changes in annual PM exposure were associated with progression of disease or reduced survival in a study of 196,000 persons discharged alive following an acute myocardial infarction. We have applied that approach in a 34-city cohort study, with cohorts defined within each city by hospital discharge for COPD.

The focus of this study is therefore to examine the effect of exposure to particulate air pollution on persons discharged alive following an admission for chronic obstructive pulmonary disease (COPD).

Figure 1 shows a map of the USA with the state boundaries and the cities analyzed in the study.

During the study period there were 1,039,287 hospital discharges with COPD; of these, 57% died before the end of follow-up.

The average duration of the follow-up was 4.5 years with the range of survival times between 1 to 14 years.

The most common type of COPD was chronic airway obstructions (55%); 14% of cases had chronic bronchitis, and 8% had emphysema.

Figure 2 presents the box-plots of the of the individually assigned 1-year PM₁₀ mean in each city, ordered by concentration of PM₁₀. Los Angeles has the highest concentrations level, while Honolulu the smallest. The average PM₁₀ across all cities was 29.4 μg/m³.

Figure 3 shows the city specific results for PM₁₀ at lag 0. In most cities we found positive and significant associations.

When looking at the sum of the effects of PM exposure on the four years (same year and 3 preceding years) in the distributed lag model we found a hazard ratio for mortality of 1.22 (95% CI: 1.17–1.27). The PM₁₀ effect sizes were higher in the distributed lag model than in the model with a single exposure period. When looking by lag, the effect on mortality risk in each year of follow-up was lower for exposure in the same year, higher for the exposure in the two previous years, and lower again for exposure three years earlier.

There was little indication of effect modification by any of the examined variables.

We found a significant effect of long-term exposure to airborne particles on the risk of death in a large multi-city study of elderly subjects discharged alive following an admission for COPD, with a relatively large effect size compared to general population cohorts previously reported. We also found that the effect was not limited to the exposure in the each year of follow-up, with larger cumulative effects spread over the follow-up year and 3 preceding years.

This study is comparable to a similar study which analyzed admissions for myocardial infarction. That study investigated whether PM was associated with progression of disease or reduced survival in a study of 196,000 persons from 21 US cities discharged alive following an acute myocardial infarction, and found a significant effect of long term exposure to airborne particles on the risk of death, progression to heart failure and to a subsequent MI and that association persisted for several years of lag, but was falling off by lag 3 [21].

This is the first long-term study which investigated persons discharged from the hospital for COPD. One key feature of this study compared to many previously published air pollution cohort studies is that it is population based, rather than based on a convenience sample in each city. This is an important distinction. While individual exposures differ from the mean exposure in each city in each year, it seems reasonable to assume that the exposure error is Berkson, and produced no downward bias in the estimated effect [31]. That is, each person's exposure fluctuates randomly around the population average exposure, a situation that reduces power, but produces no bias in effect estimates. In contrast, for a convenience sample cohort, such as the ACS study, this is less clear, since they do not capture either the population average mortality experience in the city or the population average exposure. The ACS volunteers in city A may represent a healthier, and less exposed subset of city A then they do in city B. This, clearly, can introduce bias into the estimates.

Another key difference is the focus on shorter term, within city differences in exposure. Demonstrating that effects can be seen within a few years of exposure changes has important implications for health impact assessments. If changes in risk induced by particle exposure were permanent, we would have to wait decades to see public health improvements following changes in exposure. These results indicated that they should be seen rather within years.

The relative risks of mortality for all our four outcomes are higher than has been reported in general population studies such as the six city study (RR = 1.088, 95% CI: 1.03–1.13)[32]. This may be due to the focus on the elderly, however previous studies have not suggested such large differences in risk due to age alone, and the restriction to subjects with chronic diseases thought to increase susceptibility likely plays a role in the increased effect size. Our risks are also higher than the one observed in short term effect studies, for example our hazard ration for the same year (1.11 (95% CI: 1.06–1.15)) is higher than the previous published result of 14 US cities [6] reporting a RR of 1.02 (95% CI: 1.01–1.02) for exposure in the few days preceding a death.

Another difference between this study and previous cohort studies is the source of variation in exposure. In this study the basic analysis was conducted within city, and exposure variation comes from temporal changes in pollution concentration. In the other cohort studies, the source of exposure variation is across geographic area. For example the American Cancer Society study [17] contrasted covariate adjusted survival in each city with long term average pollution in that city. To the extent that geographic variations in unmeasured confounders such as diet, usual treatment for medical conditions, etc exist as potential confounders to those studies, that problem is obviated in our study design. Instead, we must contend with unmeasured temporal confounding. We have used separate strata within each city for 5 year intervals to capture such changes, but as with any study, this may be imperfect control. Nevertheless, the finding of an association between longer term exposure to airborne particles and survival in a study with such different susceptibilities to confounding adds considerable strength to the evidence of a much larger effect of longer term exposure to particles on mortality risk.

The use of longitudinal rather than cross-sectional exposure gradients in this study may also explain some of the differences in effect size estimates, as the variation in central station monitoring and personal exposure over time may be more correlated than similar variations over space. In addition, the American Cancer Society study used monitors within the multi-counties metropolitan areas to assign exposure, whereas our subjects are generally matched to monitors in the same city or county. A recent re-analysis of the American Cancer Society study restricted to subjects living in the same county as the monitor reported a higher relative risk [33].

This study presents some limitations; one is that Medicare does not provide the underlying cause of death, limiting our ability to study susceptibility.

The main limitation, however, is the absence of information on subject characteristics such as smoking, body mass index, or medicine use. In our model we controlled for all the available personal characteristics such as age, race, gender, severity of the index admission, and detailed data on previous and secondary diagnosis. Moreover, we conducted a city-specific analysis to remove location-specific differences from the analyses. Hence differences across cities in smoking rates, etc cannot confound the association, as only the temporal variability in pollution within city contributes to the association. We do not believe year to year changes in PM₁₀ concentrations are correlated with year to year changes in smoking rates within city, but we cannot definitively exclude the possibility. Further, the emphysema death rate in persons aged 65 and older, which is strongly associated with smoking history, did not modify the PM associated risk.

Our findings suggest that long-term exposure to particulate matter elevates the risk of mortality in susceptible population defined by COPD admissions. The finding that the effect seems increasing to the preceding two years of exposure has important implications for public health. It indicates that reductions in air pollution should be followed quickly by improvements in public health, rather than the decades some have suggested. This is in accordance with some other recent studies [20, 21, 34], and heightens the urgency for such measures.