Multiple studies in the United States and worldwide have shown associations between exposure to ambient particulate matter (PM) and adverse health effects.1–6 These adverse health effects include asthma,7 cardiovascular problems,3,8–12 respiratory infections,13–17 lung cancer, and mortality.18–22
Previous studies have generally focused on either long-term (chronic) PM exposure23–26 or short-term (acute) PM exposure.15,27,28 In addition, studies of acute effects typically depend on PM2.5 monitors within their study domain. Because PM2.5 concentrations vary spatially within the study domain, this introduces exposure error and likely produces a combination of downward bias in the effect estimates and wider confidence intervals (CIs) due to a mixture of classical and Berkson error.29 The relative magnitude of the two effects is not yet clear.
The lack of spatially resolved daily PM2.5 concentration data restricts most studies to areas surrounding monitoring sites, mostly in cities, which may not be representative of the population as a whole. Such studies have used mostly city-level contrasts, although more recent studies have incorporated estimates from land-use regression. For example, cohort studies with few exceptions (eg, the Six City Study) rely upon convenience samples that are not representative of the population either in demographic or geographic characteristics, raising questions about generalizability and the possibility of selection bias. Suburban, exurban, rural, and semirural populations are particularly under-represented. Similarly, in many studies, minorities and people with lower educational attainment are under-represented. In addition, although land-use regression can produce estimates of exposure at participants addresses, the lack of monitoring in rural areas to calibrate these models has led to the exclusion of participants in such locations even when in the cohort.24
Various studies have tried to address this issue. There are some population-based chronic-exposure studies that have assessed within-city exposure and mortality. Naess et al30 looked at chronic exposure to NO2, PM10, and PM2.5 in 470 neighborhoods and all-cause and cause-specific mortality in Oslo, Norway. Exposures were consistently associated with all causes of death in both age groups for men and women. The associations were particularly strong for chronic obstructive pulmonary disease, which appeared linear whereas cardiovascular causes and lung cancer seemed to have threshold effects. Gan et al31 looked at specific traffic-related air pollutants associated with coronary heart disease (CHD) morbidity and mortality in Metropolitan Vancouver. This population-based cohort study included a 5-year exposure period and a 4-year follow-up period. Individual exposures to traffic-related air pollutants including black carbon, PM2.5, NO2, and nitric oxide were estimated at residences of the subjects using land-use regression models. An interquartile range elevation in the average concentration of black carbon was associated with a 6% increase in CHD mortality (95% CI = 3–9%) after adjusting for age, sex, preexisting comorbidity, neighborhood socioeconomic status, and copollutants (PM2.5 and NO2). Few studies on the long-term effect of exposure have looked at broader populations. Crouse et al32 looked at the risk of cardiovascular mortality in relation to long-term exposure and PM2.5 in Canada. They assigned estimates of exposure to ambient PM2.5 derived from satellite observations to a cohort of 2.1 million Canadian adults. They used both standard Cox proportional survival models and nested, spatial random-effects survival models. For the Cox models, they showed hazard ratios (HRs) of 1.15 (95% CI = 1.13–1.16) from nonaccidental causes and 1.31 (95% CI = 1.27–1.35) from ischemic heart disease for each 10-μg/m3 increase in concentrations of PM2.5. Using spatial random-effects models controlling for the same variables, they showed HRs of 1.10 (95% CI = 1.05–1.15) and 1.30 (95% CI = 1.18–1.43), respectively.
Other studies have used new methodological approaches in air pollution studies regarding short- and long-term exposure. Künzli33 presents several key approaches in studies of ambient air pollution. He discusses among other issues how cross-city longitudinal data are needed to estimate both short- and long-term effects, and how time-series studies have previously neglected the long-term component. They also show the importance of disentangling the contributions of different time domains of air pollution exposure. Künzli shows that by considering the geography of susceptibility and exposure, and by using more sophisticated approaches to acknowledge the “geographies of time,” studies on the total health effects of ambient air pollution could be considerably improved. Burnett and colleagues34 showed that results from time-series studies are equivalent to estimates obtained from a dynamic population in which each person’s mortality risk can be summarized as the daily number of deaths. The authors showed that the association between temporal variation in the environmental covariates and the survival experience of members of the dynamic population can be estimated by regressing the daily number of deaths on the daily value of the environmental covariates, as is done in time-series mortality studies.
We recently presented a novel method of assessing temporally and spatially resolved PM2.5 exposures for epidemiological studies using satellite aerosol optical depth measurements, which makes it possible to predict daily PM2.5 concentration levels at a 10 × 10 km spatial resolution.35 In this previous work, we examined the relationship between PM2.5 ground measurements and Moderate Resolution Imaging Spectroradiometer Satellite–derived aerosol optical depth measurements in New England during the period of 2000–2008. Using multistage prediction models, we initially performed day-specific calibrations of the satellite using ground PM2.5 measurements from all monitoring sites in New England and incorporating land-use regression and meteorologic variables. This use of daily calibration allows us to get considerably better predictive power, and hence lower exposure error. Later, spatial smoothing was used to predict PM2.5 concentrations for grid cell-day combinations when satellite measures are not available. Because our models produced daily PM2.5 predictions, not monthly or yearly, this allowed us to estimate the health effects of both short-term and long-term exposures. In addition, the availability of satellite measurements in every grid cell improved spatial predictions, compared with land-use regression models calibrated with monitors in a limited number of locations. Finally, we incorporated land-use regression as a final component, which estimates the difference between neighborhood level (grid cell) values and address-specific PM2.5 based on land-use terms within 50 m of the residential address, allowing us to estimate long-term exposure at an address-specific level.
In this follow-up article, we use our model predictions to study the association between both long- and short-term PM2.5 exposure and mortality in the entire population of Massachusetts during the period 2000–2008.
The study was conducted in Massachusetts (Figure). To avoid boundary effects, some PM2.5 data from neighboring states were included in the analysis.
Data for both short- and long-term PM2.5 exposures for the years 2000–2008 were assessed using recently developed prediction models.35 The Massachusetts exposure dataset encompasses daily PM2.5 predictions at a resolution of 10 × 10 km. In addition, we predicted 365-day moving averages of grid-cell PM2.5 to look at chronic effects and local deviations from the grid-cell predictions (from land-use regression) on a 50 × 50 m spatial resolution across the state (Figure) during the entire study period. Because the Boston metropolitan area within the Route 128 comprises 15 cells, this allows us to resolve exposure within the city. The local land-use component allows further resolution to the address level, but only for long-term exposure. For more detailed information on the prediction model please refer to Kloog et al.35 We then generated daily and long-term PM2.5 exposure to each decedent based on their address.
Our generated predictions have multiple advantages over other approaches commonly used in other health studies such as inverse distance weighting, kriging, etc. First, the smooth function of latitude and longitude is effectively a weighting scheme, with some important differences. The weights do not have to be isotropic, they can have a functional form be other than inverse distance squared, and they can vary every 2 months (unlike distance-based weights, which are fixed). To this we add the additional information provided by the daily satellite measurement near the residence and the local land-use regression results. There are many US cities with only one monitor close to the city. In such cases, inverse distance weighting would be of no practical use, whereas our methods would produce exposure contrasts across the various cities’ 10 × 10 km grid, as well as the local land-use regression.
Individual georeferenced mortality records were obtained from the Massachusetts Department of Public Health for all available years (2000–2008). The dataset included 468,570 deaths. Our model can include locations both proximate and far from ambient monitoring stations. We defined “near” as within 20 km of an ambient monitor. The near areas contain 80% of the population, with the rest in the “far” areas (more than 20 km from an ambient monitor). Public health records included residential location, place of death, age, sex, date of death, ethnicity, education, and primary cause of death. From these data, we constructed daily death counts for each 10 × 10 km grid cell for our time-series analysis; long-term exposure was assigned based on residential address.
As in other PM mortality studies, temperature was used as a covariate.36,37 Temperature data were obtained from the national climatic data center.38 Only continuously operating stations with daily data from 2000 to 2008 were used. Grid cells were matched to the closest weather station.
Socioeconomic variables for the tract level were obtained from the 2000 US census with data on social, economic, and housing characteristics.39 Socioeconomic variables at the census-tract level included percent minorities, age, education, and income. In addition we used individual socioeconomic variables from the Massachusetts Department of Public Health mortality records (race, education, and sex).
As a surrogate for cell-specific smoking experience, we used the long-term average lung cancer mortality rate in each cell as a control variable. Lung cancer data were obtained through the Massachusetts Department of Public Health.
Geocoded mortality data were matched with our exposure estimates. Because the mortality datasets did not include changes of residence, we had to assume when looking at long-term exposure that the decedents had lived at their current address for several years. This introduces some exposure misclassification. However, we think the misclassification is relatively minor. The average age of decedents in our study population was 75 years, and in the northeast United States, people above the age of 75 have limited mobility (only 2% of the population changed residence during 2009–2010).40
We assessed the acute effects of exposure to PM2.5 by assigning to decedents the grid exposure (on the day of death and preceding 3 days) corresponding to their residence for deaths outside of hospital. Deaths in hospital may have occurred outside the grid cell of residence, and so for those deaths, we assigned the mean predicted PM2.5 concentration in all grid cells within 30 km of the residence, again for the day of death and up to 3 preceding days. Most time-series studies have reported the strongest acute associations with mean PM2.5 for the current and previous day, rather than same day exposure41 or longer lags. We therefore took current and previous day exposures as our primary analysis. As a sensitivity analysis, we also examined PM2.5 exposure 2 days before death (lag2) and PM2.5 exposure 3 days before death (lag3). Specifically, we modeled the mortality rate λit in the ith cell on the tth day as follows:
where PM it is the daily PM2.5 concentration in cell i, λ(t) is a smooth function of time, temporal covariates are temperature and day of the week, spatial covariates are socioeconomic factors and a surrogate for long-term smoking history, and e i is the remaining unexplained difference in mortality rate between cell i and other cells (treated as a mean zero normal random effect with variance estimated from the data). This model expands the usual time-series analysis by including spatial covariates and random intercepts for small areas.
The specific covariates we used were a linear and quadratic term for temperature with the same moving average as PM2.5, age, percent minorities, median income, percent of people without high school education, and lung cancer rate as a surrogate for long-term smoking history. λ(t) was estimated with a natural cubic spline with 45 degrees of freedom (5 df per year). We used an interaction term between in-hospital death and short-term exposure, to test whether location modified the association.
Long-term exposure was analyzed through a relative incidence analysis. We defined particle-related deaths as those from cardiovascular and respiratory diseases and contrasted those with mortality unrelated to air pollution (accidental death, cancer [except lung cancer], homicide, etc.). Specifically we fit the following model:
where Mort ij is the response (particle-related or nonrelated death) for the ith subject in grid j, α and u j are the fixed and random (grid-specific) intercepts, respectively, PM it is 365-day moving average ending on day t, PML i is the local (50 m) deviation of PM2.5 from the long-term grid-cell average, X 3i, etc. denote the set of covariates of interest used in the model, and σ u 2 is the variance of the random effects.
We also looked at various interactions of interest, including interactions between near and far areas and both short-term and long-term exposure as well as interactions between low and high education groups (based on individual college education) and long-term PM2.5 exposure. We considered whether effect estimates differed between the two main cause-specific mortality rates by performing a logistic analysis of cardiovascular and respiratory mortality.
Of the 468,570 deaths included in our analyses, 46% were men, 94% were white, and 20% had higher than a high school education. The average age at death was 75 years (Table 1).
Table 2 summarizes of the exposure and temperature variables used in the analysis. Various lags were tested in the time-series analysis, with the strongest based on the mean of lag01. The results for lag01 are presented as commonly reported in many previous studies.14,24 For every10-µg/m3 increase in short-term PM2.5 exposure, there was a 2.8% increase in mortality (95% CI = 2.0–3.5). There was also a significant interaction between short-term pollution and near and far locations (P < 0.001). For every10-µg/m3 increase in short-term PM2.5 exposure in the “far” group, there was a 1.4% increase in mortality (95% CI = 0.8–2.0%). For every10-µg/m3 increase in short-term PM2.5 exposure in the “near” group, there was a 4.5% increase in mortality (95% CI = 2.6–6.5%).
Table 3 presents the odds ratios (ORs) for cardiovascular and respiratory mortality for a 10-µg/m3 increase in long-term grid-cell PM2.5 exposure for the full datasets and various interactions. The OR was 1.6 (CI = 1.5–1.8) for cardiovascular and respiratory mortality relative to deaths from other causes. Local PM2.5 had an OR of 1.4 (CI = 1.3–1.5) in the same model. Because local PM2.5 was constructed as the address-specific deviation from the grid-cell average, these effects are independent and additive. There was a significant interaction between education group and long-term pollution (P<0.001). The OR for the high education group (college education) was 1.4 (CI = 1.2–1.6), whereas for the remainder the OR was 1.9 (CI = 1.6–2.1). There was also a significant interaction between long-term pollution and near and far groups (P < 0.001). The OR for subjects living more than 20 km from a monitor was 1.3 (CI = 1.1–1.6), whereas for those living closer, the OR was 1.7 (CI = 1.5–1.9). The OR for nonwhites (OR 2.9, CI = 1.9–4.5) was higher than for whites (OR 1.6, CI = 1.4–1.7).
The logistic regression comparing cardiovascular and respiratory mortality showed no appreciable differences both for the regional PM2.5 and local PM2.5 exposure. Similarly, there were no important risk differences between people who died in the hospital and those who died at home.
The main feature of this study is the use of novel hybrid prediction models that examine short- and long-term exposure associations with mortality, and include the entire population of the state. Unlike traditional land-use regressions, this model produces daily predictions. This model also performs better in regions far from monitors because the satellite data provide exposure data. In addition, these models can 1) control for small area socioeconomic status variables in time-series analysis and look for effect modification by the same variables; 2) test whether the PM2.5 slope is different in people residing far from monitors; 3) assess chronic effects of particles using the entire population, again including people distant from monitors; 4) estimate effects of local traffic–derived particles independent of regional particles; and 5) reduce exposure error and hence downward bias in slopes, and upward bias in CIs. Key findings include differences in the slopes of acute and chronic PM2.5 between locations closer or more distant from monitoring stations, interactions of chronic exposure with race and socioeconomic status (with less advantaged groups having stronger associations), and an additional effect of traffic particles generated near the address of the decedents. The Six City Study42,43 found that a 1-year period for chronic exposure captured essentially the entire effect. Hence our use of a 1-year average should be sufficient to capture the long-term effect of PM2.5 exposure.
It is instructive to compare these results with results of previous time-series and cohort studies. Zanobetti and colleagues studied the acute effect of fine and coarse particulate air pollution on mortality in a national analysis of 112 cities.44 For cities similar to cities in our study (the northeast area, grouped as cities with warm summer-continental), a 10-µg/m3 increase in PM2.5 was associated with a 1.19% increase (95% CI = 0.73–1.64) in total mortality. A key difference is that they assigned the same daily exposure to all people in each city, whereas we were able to assign different exposures to people on a relatively small grid. Another key difference is that they analyzed only cities, whereas we could analyze an entire state. Our study finds larger and more precise effect-size estimates despite analyzing fewer deaths, possibly due to reduced exposure error.
Berkson measurement error for exposure would be expected to bias standard errors upward, but not bias effect sizes down, whereas classical exposure error would do the opposite. If exposure error was a mixture of these two types, reduced exposure error could produce results such as we observed. This does not preclude other explanations.
Miller et al45 reported a relative risk of 1.76 for cardiovascular deaths per 10 μg/m3 using within-city exposure, and Puett and colleagues24 used exposure from a land-use regression analysis to estimate a hazard ratio of 1.26 for all-cause mortality (95% CI = 1.02−1.54) with each 10-µg/m3 increase in annual PM2.5 exposure. These estimates are similar to ours and higher than in older studies that did not have geographically resolved exposure. That suggests the exposure error in cohort studies is predominantly classical. The similarly larger estimate in our study supports this conclusion, and also extends the results to suburban and rural residents. Furthermore, our use of satellite exposure data allowed a much larger sample size, with more deaths than in the American Cancer Society (ACS) study, the Nurses’ Health Study, and the Six City Study combined. We confirm the results of the ACS study that people with less education have greater susceptibility to particles, although the association in the college educated is still substantial, and with a relatively narrow CI.
Although covariate control is a limitation of this analysis, the use of fine-scale geographically resolved exposure is a clear advantage over studies such as the first reports from ACS study, which used metropolitan areas often encompassing multiple counties, and often with only one monitor available to assign exposure. A reanalysis of the data by Willis et al,46 restricted to people who live closer to the monitor, reported a doubling of the estimate slope per unit exposure, suggesting substantial downward bias by classical measurement error. These results plus the similarity between our estimates and other estimates based on geographically resolved exposure24,47 provides some assurance about the generalizability of those estimates. As noted above, we see smaller effects in locations more than 20 km from monitors, which generally have lower population density. This may reflect different composition (more traffic particles in the more densely populated regions), although further analysis is needed to confirm this. Consistent with this, we found that deviations from grid-cell-average PM2.5 predicted by land-use terms within 50 m of residence had additional predictive power for mortality. This also suggests that traffic particles are more toxic.
A key difference between our analysis and the ACS or Nurses Health study is that they oversampled the highly educated, undersampled minorities. In general, cohort studies of the long-term effects of PM2.5 have been nonrepresentative. Our finding of a substantially stronger associations in nonwhites and the less educated raises an important environmental justice concern.
Another major limitation of the present study is the spatial resolution of 10 × 10 km. We compensate for this in our analysis of chronic effects by using a land-use regression to compute a local PM variable. Better resolution of the satellite data would clearly be beneficial and should become available. As satellite remote sensing evolves, higher spatial resolution data (eg, 3 × 3 km and 1 × 1 km) will further reduce exposure error. Our mortality datasets did not include changes in residence, which will introduce some exposure misclassification into our study.
In conclusion, our novel prediction models, making use of satellite data on air pollution, perform well in assessing short-term and long-term effects of PM2.5 exposure. This enables us to examine entire populations, including exurban and rural locations, with better spatial resolution and tighter CIs for the time-series estimates. The larger effect-size estimates seen in recent cohort studies (using geographically resolved chronic exposures) are supported in this analysis for a whole population. This modeling approach presents new opportunities to study the effects of both short- and long-term particle exposures on human health.
We thank Steven J. Melly, Department of Environmental Health, Harvard School of Public Health, Harvard University.
1. Atkinson RW, Anderson HR, Sunyer J, et al. Acute effects of particulate air pollution on respiratory admissions: results from APHEA 2 project. Air Pollution and Health: a European Approach. Am J Respir Crit Care Med. 2001;164(10 Pt 1):1860–1866
2. Gold DR, Litonjua A, Schwartz J, et al. Ambient pollution and heart rate variability. Circulation. 2000;101:1267–1273
3. Le Tertre A, Medina S, Samoli E, et al. Short-term effects of particulate air pollution on cardiovascular diseases in eight European cities. JEpidemiol Community Health. 2002;56:773–779
4. Salvi S, Blomberg A, Rudell B, et al. Acute inflammatory responses in the airways and peripheral blood after short-term exposure to diesel exhaust in healthy human volunteers. Am J Respir Crit Care Med. 1999;159:702–709
5. Schwartz J. Air pollution and hospital admissions for the elderly in Detroit, Michigan. Am J Respir Crit Care Med. 1994;150:648–655
6. Schwartz J. Air pollution and blood markers of cardiovascular risk. Environ Health Perspect. 2001;109(suppl 3):405–409
7. Lin S, Munsie JP, Hwang SA, Fitzgerald E, Cayo MR. Childhood asthma hospitalization and residential exposure to state route traffic. Environ Res. 2002;88:73–81
8. Barnett AG, Williams GM, Schwartz J, et al. The effects of air pollution on hospitalizations for cardiovascular disease in elderly people in Australian and New Zealand cities. Environ Health Perspect. 2006;114:1018–1023
9. Wellenius GA, Bateson TF, Mittleman MA, Schwartz J. Particulate air pollution and the rate of hospitalization for congestive heart failure among medicare beneficiaries in Pittsburgh, Pennsylvania. Am J Epidemiol. 2005;161:1030–1036
10. Peters A, Fröhlich M, Döring A, et al. Particulate air pollution is associated with an acute phase response in men; results from the MONICA-Augsburg Study. Eur Heart J. 2001;22:1198–1204
11. Zanobetti A, Schwartz J, Dockery DW. Airborne particles are a risk factor for hospital admissions for heart and lung disease. Environ Health Perspect. 2000;108:1071–1077
12. Schwartz J, Morris R. Air pollution and hospital admissions for cardiovascular disease in Detroit, Michigan. Am J Epidemiol. 1995;142:23–35
13. Baccarelli A. Breathe deeply into your genes!: genetic variants and air pollution effects. Am J Respir Crit Care Med. 2009;179:431–432
14. Zanobetti A, Franklin M, Koutrakis P, Schwartz J. Fine particulate air pollution and its components in association with cause-specific emergency admissions. Environ Health. 2009;8:58
15. Dominici F, Peng RD, Bell ML, et al. Fine particulate air pollution and hospital admission for cardiovascular and respiratory diseases. JAMA. 2006;295:1127–1134
16. Sunyer J, Basagaña X. Particles, and not gases, are associated with the risk of death in patients with chronic obstructive pulmonary disease. Int J Epidemiol. 2001;30:1138–1140
17. Schwartz J. Air pollution and hospital admissions for respiratory disease. Epidemiology. 1996;7:20–28
18. Dockery D, Pope C, Xu X, et al. An association between air pollution and mortality in six US cities. N Engl J Med. 1993;329:1753–1759
19. Franklin M, Zeka A, Schwartz J. Association between PM2. 5 and all-cause and specific-cause mortality in 27 US communities. J Expo Sci Environ Epidemiol. 2006;17:279–287
20. Pope CA 3rd, Burnett RT, Thun MJ, et al. Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. JAMA. 2002;287:1132–1141
21. Schwartz J. Air pollution and daily mortality: a review and meta analysis. Environ Res. 1994;64:36–52
22. Laden F, Schwartz J, Speizer FE, Dockery DW. Reduction in fine particulate air pollution and mortality: extended follow-up of the Harvard Six Cities study. Am J Respir Crit Care Med. 2006;173:667–672
23. Hart J, Garshick E, Dockery D, Smith T, Ryan L, Laden F. Long-term ambient multi-pollutant exposures and mortality. Am J Respir Crit Care Med. 2011;183:73–78
24. Puett RC, Hart JE, Yanosky JD, et al. Chronic fine and coarse particulate exposure, mortality, and coronary heart disease in the Nurses’ Health Study. Environ Health Perspect. 2009;117:1697–1701
25. Schwartz J. Harvesting and long term exposure effects in the relation between air pollution and mortality. Am J Epidemiol. 2000;151:440–448
26. Jerrett M, Burnett RT, Ma R, et al. Spatial analysis of air pollution and mortality in Los Angeles. Epidemiology. 2005;16:727–736
27. Katsouyanni K, Touloumi G, Spix C, et al. Short-term effects of ambient sulphur dioxide and particulate matter on mortality in 12 European cities: results from time series data from the APHEA project. Air Pollution and Health: a European Approach. BMJ. 1997;314:1658–1663
28. Schwartz J, Dockery DW. Particulate air pollution and daily mortality in Steubenville, Ohio. Am J Epidemiol. 1992;135:12–19 discussion 20.
29. Zeger SL, Thomas D, Dominici F, et al. Exposure measurement error in time-series studies of air pollution: concepts and consequences. Environ Health Perspect. 2000;108:419–426
30. Naess Ø, Nafstad P, Aamodt G, Claussen B, Rosland P. Relation between concentration of air pollution and cause-specific mortality: four-year exposures to nitrogen dioxide and particulate matter pollutants in 470 neighborhoods in Oslo, Norway. Am J Epidemiol. 2007;165:435–443
31. Gan WQ, Koehoorn M, Davies HW, Demers PA, Tamburic L, Brauer M. Long-term exposure to traffic-related air pollution and the risk of coronary heart disease hospitalization and mortality. Environ Health Perspect. 2011;119:501–507 Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3080932/
Accessed 10 October 2012.
32. Crouse DL, Peters PA, van Donkelaar A, et al. Risk of nonaccidental and cardiovascular mortality in relation to long-term exposure to low concentrations of fine particulate matter: a Canadian National-Level Cohort Study. Environ Health Perspect. 2012;120:708–714 Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3346774/
Accessed 10 October 2012
34. Burnett RT, Dewanji A, Dominici F, Goldberg MS, Cohen A, Krewski D. On the relationship between time-series studies, dynamic population studies, and estimating loss of life due to short-term exposure to environmental risks. Environ Health Perspect. 2003;111:1170–1174 Copyright © 2003 The National Institute of Environmental Health Sciences (NIEHS). Available at: http://www.jstor.org/stable/3435505
Accessed 10 October 2012.
35. Kloog I, Koutrakis P, Coull BA, Lee HJ, Schwartz J. Assessing temporally and spatially resolved PM2. 5exposures for epidemiological studies using satellite aerosol optical depth measurements. Atmos Environ. 2011;45:6267–6275
36. Schwartz J. Is the association of airborne particles with daily deaths confounded by gaseous air pollutants? An approach to control by matching. Environ Health Perspect. 2004;112:557–561
37. Schwartz J. The effects of particulate air pollution on daily deaths: a multi-city case crossover analysis. Occup Environ Med. 2004;61:956–961
38. National Climatic Data Center. The National Climatic Data Center Data Inventories. Asheville, NC: NOAA National Climatic Data Center (NCDC); 2010
39. Census US. U.S. Census of Population and Housing. 2000 Washington, DC U.S. Dept. of Commerce
40. Census US. Current Population Survey, 2010 Annual Social and Economic Supplement. 2011 Government Printing Office
41. Schwartz J, Dockery DW, Neas LM. Is daily mortality associated specifically with fine particles? J Air Waste Manag Assoc. 1996;46:927–939
42. Lepeule J, Laden F, Dockery D, Schwartz J. Chronic exposure to fine particles and mortality: an extended follow-up of the Harvard Six Cities study from 1974 to 2009. Environ Health Perspect. 2012;120:965–970
43. Schwartz J, Coull B, Laden F, Ryan L. The effect of dose and timing of dose on the association between airborne particles and survival. Environ Health Perspect. 2008;116:64–69
44. Zanobetti A, Schwartz J. The effect of fine and coarse particulate air pollution on mortality: a national analysis. Environ Health Perspect. 2009;117:898–903
45. Miller KA, Siscovick DS, Sheppard L, et al. Long-term exposure to air pollution and incidence of cardiovascular events in women. N Engl J Med. 2007;356:447–458
46. Willis A, Jerrett M, Burnett RT, Krewski D. The association between sulfate air pollution and mortality at the county scale: an exploration of the impact of scale on a long-term exposure study. J Toxicol Environ Health A. 2003;66:1605–1624
47. Miller KA, Siscovick DS, Sheppard L, et al. Long-term exposure to air pollution and incidence of cardiovascular events in women. N Engl J Med. 2007;356:447–458