Cohort studies on the long-term effects of air pollution suggest a reduction in the survival of the general population in association with ambient particles,1,2 other traffic-related pollutants,3 and indicators of traffic-related air pollution.4,5 Although most cohort studies have focused on between-city differences in exposure and survival,1,2 recent attention has been drawn to the role of differences in exposure within a given city and their relation to survival patterns. Recent studies7,8 have reported larger effect sizes for these contrasts, which are primarily the result of mobile source pollution. These prior studies assigned the same exposure to all subjects within a defined geographic area. However, studies have demonstrated that there is considerable variation in exposure to air pollution on a smaller scale, with residence on a busy street, for example, associated with greater exposure.5 There are relatively few studies examining the association between mortality and within-community variations in traffic exposure.
The mechanisms of long-term effects of particulate air pollution are not fully understood, but cohort studies in the general population have shown more pronounced associations for cardiovascular-related causes than for other causes of death.9 These findings suggest that initiation, as well as exacerbation, of atherosclerosis, pulmonary, and systemic inflammation and altered cardiac autonomic function are possible pathophysiologic pathways for the adverse effects associated with particulate air pollution.9 The proposed mechanisms have been recently supported by the results of animal studies,10,11 showing exacerbation of aortic plaque development as well as inflammation and altered vasomotor tone in mice, and by epidemiologic studies showing an association between ambient air pollution and atherosclerosis.12,13 Several studies have identified susceptible subgroups for the acute effects of particulate matter (PM) exposure, including individuals with diabetes14 and patients who had survived a recent myocardial infarction.15 Less is known about the susceptibility to long-term PM exposure, although the effects of long-term exposure appear greater than those of acute exposure.16 These adverse effects may ultimately affect the broader population,17 and (given the variety of postulated pathophysiologic mechanisms involved) patients with cardiovascular disease in particular.
We conducted a cohort study of hospital survivors of acute myocardial infarction (MI) in Worcester, MA, USA and modeled within-community traffic exposure measures to investigate the long-term effects of traffic-related air pollution on mortality in a potentially sensitive group.
The data for this investigation are from the Worcester Heart Attack Study. This is an ongoing community-wide investigation examining changes over time in the incidence and case-fatality rates of confirmed episodes of MI in residents of the greater Worcester area, who were hospitalized with MI at all area medical centers.18,19 The details of this study have been described previously in detail.18,19 In brief, during the 5 study years, 1995, 1997, 1999, 2001, and 2003, the medical records of the 11 acute care general hospitals serving residents of the Worcester metropolitan area were searched for patients with a possible discharge diagnosis of MI. The records were reviewed and independently validated according to diagnostic criteria described previously.19 These criteria included a suggestive clinical history, increased serum biomarker levels above each hospital's normal range, and serial electrocardiography findings indicative of acute MI. At least 2 of these 3 criteria were necessary for study inclusion. Information was abstracted from hospital medical records for each patient's demographic and clinical characteristics, medical history, smoking status (current vs. non), hospital survival status, MI order (initial vs. prior), and MI type (Q wave vs. non–Q wave). The present investigation was limited to adult patients 25 years and older who were hospitalized with independently confirmed acute MI. Patients' residential addresses at the time of MI were collected from information contained in hospital medical records and geocoded. Long-term survival status was ascertained through the end of calendar year 2005 through the review of records for additional hospitalizations and a statewide and national search of death certificates for greater Worcester residents. Some form of additional follow-up was obtained for nearly all (>99%) discharged hospital survivors. The study was approved by the Committee for the Protection of Human Subjects at the University of Massachusetts Medical School and the Human Subjects Committee at the Harvard School of Public Health.
A land use regression model for traffic-related air pollution20 was developed, which predicted weekly local PM2.5 filter reflectance as a latent variable. PM2.5 filter reflectance is a proxy measure of traffic-related particulate air pollution. The model was fit to monitoring data of PM2.5 filter reflectance and nitrogen dioxide (NO2) concentrations at 36 locations in the greater Worcester area chosen to represent a broad range of exposures to traffic particles. Sampling was conducted continuously from September 2003 to April 2005 for filter reflectance and from December 2003 to April 2005 for NO2 by using a rotating-site sampling design such that data were collected at 6–10 sites during any given week. Consequently, data were available for 2 winters and 2 spring seasons, and for 1 fall and 1 summer season.
The prediction model included elemental carbon measured at a central monitoring site located on the roof of Countway Library, Harvard Medical School, in downtown Boston, MA (75 km from Worcester, capturing regional elemental carbon day-to-day variation), using an aethalometer (Model 8021; McGee Scientific, Berkeley, CA), day of the year, height of the planetary boundary layer (predicted from spatial smoothing of the weekly average planetary boundary layer estimates from the North American Regional Reanalysis21 at 18 locations within 75 km of Worcester), and urbanization (US Geological Service National Land Cover Dataset), as explanatory variables. We used spline regression methods22 to allow these factors to affect exposure levels in a potentially nonlinear way. Finally, we used thin-plate splines, a 2-dimensional extension of regression splines, to model longitude and latitude and capture additional spatial variability unaccounted for after including our deterministic spatial predictors in the model. This approach is a form of universal kriging (ie, kriging extended to incorporate covariates), or a geoadditive model,23 for daily concentrations of particle levels. This spatial surface was allowed to vary by season. A Bayesian Markov chain Monte Carlo algorithm for model fitting was used.
We checked the goodness-of-fit of our models by comparing summaries of the observed data to their corresponding predicted distributions from the model fit, which in the Bayesian context is referred to as posterior predictive checking. The comparison of the observed versus fitted 5% and 95% percentiles of log-transformed filter reflectance suggested that the tails of our predictions matched the observed data well. We also compared the predictive posterior cumulative distribution function to its observed counterpart, which suggested that the model fit well the entire range of observed data.
The prediction model was used to estimate weekly exposure at the addresses of each study participant for their respective follow-up periods. For elemental carbon measured in Boston, 25% (1029) of daily values were missing, mainly because of an interruption of the measurements from spring 1997 to spring 1999. We imputed missing values of daily elemental carbon in Boston by using other air pollutants measured in Boston, meteorology, and weekday as predictors and achieved a cross-validated R 2 of 88% for these predictions. With the completed weekly elemental carbon time-series, covariate data were available to make predictions of filter reflectance at each residence and for the entire study period of 1995–2005.
Filter reflectance was then converted to elemental carbon by using the conversion factor obtained through comparisons of filter reflectance and elemental carbon measurements. We calculated 4 sets of estimated elemental carbon annual averages for each individual: (A) 1 mean for each calendar year of follow-up; (B) a fixed calendar year mean for the year of entry; (C) monthly moving annual average (mean of lag 0–11); and (D) mean for the year of 2000.
Census Block Group Data
We obtained 2000 census data from the US Bureau of the Census Summary File (III). Based on previous work,24 we used the proportion of the population with income in 1999 below the federally defined poverty level and median household income within a census block group as an area-based measure of socioeconomic position. Census block groups have a population of about 1500 individuals and are defined by the Census Bureau as small statistical subdivisions of counties with generally stable boundaries, designed to have relatively homogeneous demographic and economic characteristics. Census block group data on economic poverty have been shown to be a relatively sensitive measure of socioeconomic inequalities in health outcomes.25 The subjects of this study lived in 374 different block groups within the greater Worcester area.
We calculated weekly and monthly averages of temperature and relative humidity at Worcester Airport, provided from the National Climatic Data Center.
We used a Cox proportional hazards regression model to investigate the association between estimated annual mean elemental carbon at residence with the risk of dying for subjects discharged after MI. We selected survival time after hospital discharge as the relevant time axis for the analyses, excluding patients with a hospital stay of more than 28 days. The survival times of participants who did not die over the course of our extended follow-up were censored at the end of the study period.
We used the predicted elemental carbon value at residence for each calendar year of follow-up as exposure (exposure covariate choice A) to develop the final regression model.26
We considered age, sex, race, marital status, hospital, medical history, and clinical complications of the acute coronary event as potential confounders of the associations under study. We tested interaction terms for 3 age categories and sex. In the final regression model, we controlled for age (cubic), sex, hospital of admission, development of a Q-wave MI, occurrence of atrial fibrillation, cardiogenic shock, and heart failure during hospitalization and a medical history of stroke, heart failure, angina, diabetes, and previous MI.
To test the validity of the proportional hazards assumption, we plotted log–log survival curves and assessed the correlation of Schoenfeld residuals with the ranking of individual failure times. Both approaches suggested that all terms, except for the exposure variable of interest (annual mean elemental carbon at residence) met the proportional hazards assumption. However, using the annual mean elemental carbon at residence divided into yearly follow-up time intervals, we found that the proportional hazards assumption was met for exposure in the first 2 years, and also in the subsequent follow-up period. Therefore, in the final regression models, we simultaneously estimated a hazard ratio for the first 2 years of follow-up and 1 for the period thereafter. We repeated the final model by using the 3 other annual elemental carbon estimates as exposure covariates (B, C, and D).
Because socioeconomic position and long-term air pollution exposure are correlated, and a lower long-term survival after MI has been shown previously in more deprived neighborhoods,24 we included census block group level proportion of poverty and median household income into further adjusted models.
To assess potential effect modification by preexisting conditions, we stratified the analyses by sex, age, race, and various clinical characteristics as well as according to the presence of preexisting conditions that had also been considered as potential confounders to the associations under study.
Because information on current smoking status was missing for 20% of the study sample, we did not include this variable in the main model. In a subset analysis, however, we adjusted for smoking status (current vs. nonsmoking), which may be an important confounder.
To adjust the long-term effect of elemental carbon exposure for any acute effects, we entered weekly elemental carbon simultaneously with the mean elemental carbon of calendar year of follow-up into the model, controlling for weekly mean temperature, relative humidity, and season.
We explored the shape of the distribution of the effect of annual mean elemental carbon at residence over time, using a polynomial (cubic) distributed lag model of lag 0–11 months,27 controlling for month with a categorical variable.
Tests of Residual Spatial Autocorrelation
We examined whether there was a need to account for spatial autocorrelation in our model, which could result in underestimation of the standard errors and incorrect effect estimates. Following the work of Therneau and Grambsch,28 we used the expected rates of the full Cox regression model in a generalized additive Poisson regression model that included the same covariates as in the main model and added a smooth term for space (using penalized splines in R 2.6.0, The mgcv Package). This model did not show a change in the estimated parameters for our exposure of interest or in its standard error, and the spatial term did not improve the model fit. We also examined the spatial structure of the residuals of the Cox model calculating the empirical semivariogram of the normalized Martingale residuals (deviance residuals), which also suggested no remaining spatial correlation. Last, we repeated the analyses as pooled logistic regression,29 including all covariates as specified for the Cox model as well as a smooth term of space. Effect estimates were very similar to the Cox model and there was no evidence for spatial correlation.
We stratified by time period of elemental carbon imputation to assess exposure misclassification. We repeated the analyses, including age in 5 year classes and sex as stratification variables to verify whether the results were robust after allowing the baseline hazard functions to vary over these strata. Rather than include hospital as a categorical variable, which might have captured some of the spatial variability of exposure, we included hospital as a random variable.
Data were available on a total of 4096 residents of the greater Worcester area who survived their index hospitalization for MI. Of these, 3895 residents (95%) had complete follow-up data (demographic and clinical characteristics, as well as geocoded address location) and were discharged after less than 28 days of the acute index event. The mean age of the study sample was 69 years, the majority were men (59%), 93% were white, and 71% presented with an initial MI (Table 1). By the end of follow-up (December 2005), 44% had died; 58% of these patients had survived for less than 2 years.
Estimated annual mean elemental carbon at residence in the year 2000 averaged 0.42 μg/m3, and ranged from 0.05 to 0.92 μg/m3 (Table 1; eFig., http://links.lww.com/A935). Annual elemental carbon levels decreased by a total of 3% between 1995 and 2005 at all locations.
Survival After MI and Exposure to Elemental Carbon
Figure 1 shows that subjects living in areas with lower elemental carbon levels at their residence had higher long-term survival rates. An increase in 1 interquartile range annual elemental carbon (0.24 μg/m3) was associated with a 15% increase in death rate after the second year of follow-up, whereas there was no association with annual elemental carbon levels in the first 2 years of follow-up (Table 2). All 4 exposure averaging approaches provided essentially similar results. When including the area socioeconomic position measures in the regression model, the estimate for annual elemental carbon after the second year of postdischarge survival was attenuated (Table 2, Fig. 2). The estimate was not further attenuated when we included both socioeconomic position measures in the regression model (hazard ratio [HR] for elemental carbon in the first 2 years 0.96 (95% confidence interval [CI] = 0.88–1.06) and 1.09 (0.96–1.22) after the second year of follow-up).
Men and women did not differ substantially with regard to the associations observed (Table 2). We also could not detect effect modification according to age, race, marital or smoking status, or type of MI (eTable, http://links.lww.com/A935). Although we did not find any interactions with patient's medical history of diabetes, chronic obstructive pulmonary disease, stroke, heart failure, or hypertension, there was some indication that the effect of elemental carbon may be stronger in those with an initial MI compared with those with a prior MI and in those with previously diagnosed angina (HR for elemental carbon in the first 2 years 1.23 [1.03–1.46] and 1.22 [0.99–1.52] after the second year of follow-up) compared with those without angina (HR = 0.96 [0.87–1.06] and 1.13 [0.99–1.29], respectively). We found no effect modification by area level socioeconomic position (see eTable, http://links.lww.com/A935).
Including smoking status as a covariate did not change the effect estimate for yearly elemental carbon (Table 2).
When we included weekly residential elemental carbon exposure, controlling for relative humidity, temperature, and season, the estimate for yearly residential elemental carbon exposure remained unchanged (Table 2), and we observed no effect for weekly elemental carbon exposure (HR for weekly elemental carbon 1.00 [0.90–1.12 per 0.24]). When excluding yearly elemental carbon in these models, a weak association of weekly elemental carbon with mortality was observed (HR = 1.04 [0.97–1.10]).
The polynomial distributed lag model of elemental carbon exposure after the second year of follow-up suggested a larger effect of elemental carbon in the 6 months directly before death than for exposures in the more distant past (Fig. 3). This pattern was robust to a change in the degree of polynomial (from cubic to quartic or quintic) and allowing for an “independently” estimated lag 0 months effect.
Excluding the periods where imputed elemental carbon values entered the prediction model did not change the results (Table 2). The results appeared not to be affected by including hospital as a random effect and age and sex as stratifying variables (HR for elemental carbon in the first 2 years 1.00 [0.92–1.09] and 1.15 [1.03–1.29] after the second year of follow–up).
We conducted a cohort study to investigate the association of long-term traffic-related particulate air pollution with survival of acute MI patients after hospital discharge among residents of a large New England metropolitan area. The results of this population-based study suggest increased mortality after the second year of survival in association with chronic traffic pollution at residence. No association of long-term traffic pollution exposure was observed during the first 2 years of postdischarge follow-up. The associations were attenuated after adjustment for area-level socioeconomic position. The exposure concentrations closer to the event appeared to have a stronger effect on observed mortality patterns than longer lagged exposures.
Previous cohort studies have shown consistent effects of PM exposure on mortality when comparing communities,1,2,4,9,30 as well as more recently when comparing smaller areas within communities.7,8,31 Exposure to vehicular traffic may be of particular concern, as shown in cohort studies that specifically investigated associations of mortality with small-scale variations in proximity and density measures of long-term traffic exposure.4,5,32,33 To the best of our knowledge, the present study is the first to apply a within-community spatiotemporal model of traffic particles to examine long-term health effects of particulate air pollution.
The present study suggests a long-term effect of elemental carbon only after the second year of survival after MI. Although this finding may be because of chance or bias, there are other plausible explanations for this observation. In contrast with the majority of the earlier studies on the long-term effects of traffic-related particulate air pollution, we examined a population of survivors of MI. Typically, the general population has been investigated in studies of chronic air pollution health effects,2,30 as well as population subgroups characterized on the basis of sex4,8 or age.5 As has been shown in prior studies of MI survivors, postdischarge mortality is highest in the first several months to years after hospital discharge.34 Therefore, clinical characteristics of the acute event may be more important determinants of survival than air pollution during this high-risk period. Long-term survivors of an MI may be more affected by chronic exposure to air pollution. Another possible explanation may be patients' decreasing compliance over time to effective cardiac medications that may mask an effect of particulate air pollution.
There is 1 previous study showing associations of chronic PM exposure and adverse post-MI outcomes in persons who survived an MI.35 The investigators did not report a change in the estimated health effects of chronic exposure to PM over time. However, that study and the present one are very different in terms of setting and exposure parameters. Although that study used US Medicare data and within-city between-year contrasts in PM exposure, we had a smaller dataset with more detailed and validated case data and a spatiotemporal model for exposure to traffic air pollution.
We observed an attenuation of the effect estimate for elemental carbon after the second year of survival, after inclusion of area-level socioeconomic position. Area-level socioeconomic position may be a confounder because it reflects the socioeconomic context of the area and is likely to be correlated with individual-level socioeconomic position. A previous investigation observed that low area-level socioeconomic position is a predictor of reduced survival in this population.24 In addition to reflecting the availability and cost of healthful food, area-level socioeconomic position also reflects differences in traffic burdens and availability of public space, which may explain part of the attenuation of the effect.36,37 When we stratified our data according to 2 levels of census block group median household income, we found no effect modification by area-level socioeconomic position. We did not have adequate data to address the role of individual-level socioeconomic position, which may have led to some residual confounding of our results.
The association of long-term traffic exposure at residence with patient's postdischarge survival remained unchanged when we added weekly averages as a more acute marker of exposure to the primary model. In this model, weekly average elemental carbon itself did not show an association with survival. Although we could not adjust for day-to-day changes in air pollution, this result might be regarded as support for the notion that chronic air pollution exposure contributes to long-term pathophysiologic processes, independent of the level of pollution at the time of the acute event.38
We used estimates of yearly exposure at place of residence as a measure of long-term or chronic exposure, as suggested in previous studies examining the association between air pollution and long-term survival.39 The issue of temporal aggregation in cohort studies has recently been addressed as an area of potential concern.40 Because we had a spatiotemporal model for traffic air pollution, we were able to reduce temporal aggregation of individual-level data. Our main models also considered temporal changes in residential exposure by using the mean of each calendar year of follow-up. Of note, the models in our study in which more temporally aggregated exposure data were used gave essentially the same results, which may be because of the comparably low decrease in exposure over time (3% over 10 years).
We were not able to analyze the association of traffic particle exposure and cardiovascular death separately because cause-of-death information was not available. Although previous studies have investigated the long-term effects of air pollution on mortality in the general population and according to cause of death,9,39 the present study has assessed all-cause mortality in hospital survivors of an MI, individuals suffering from underlying coronary heart disease. Death from cardiovascular-related causes has been shown to be markedly higher in MI survivors compared with the general population.41 Therefore, the results observed for all cause-mortality are likely to be primarily related to cardiovascular disease, unlike studies of the general population. Furthermore, cause of death from death certificates are frequently improperly coded,42 which would introduce additional measurement error.
Complete residential history information was unavailable. We assumed that all cohort members lived at the same place of residence over the entire study period. This assumption might have introduced some exposure misclassification and bias toward the null hypothesis.
In the present study, we incorporated a spatiotemporal process into the assessment of the chronic health effects of traffic air pollution. As noted by previous authors,43,44 it is important to assess whether spatial autocorrelation is present. Ignoring this may lead to underestimation of standard errors and incorrect effect estimates. We investigated the empirical semivariogram of the deviance (normalized Martingale) residuals, and found no evidence of remaining spatial autocorrelation. Although these residuals from the Cox model are the closest analog to residuals from a linear regression model, the analyses of these for purposes of detecting of autocorrelation have not yet been established.43 Therefore, we also included location as a smooth function into the deterministic component44 of a pooled logistic29 and a Poisson regression28 model. None of these analytic approaches suggested remaining spatial autocorrelation, and they gave essentially identical results for the elemental carbon effect estimates.
The spatiotemporal model of elemental carbon in our study setting builds on filter reflectance and NO2 measurements at a wide variety of sites in the study area, for 2003–2005. By using the predictions from this model for earlier years and elemental carbon conversion factors, we assumed that the relations between filter reflectance, elemental carbon, and the explanatory variables remained the same over time. The main spatial predictors of yearly elemental carbon were location and urbanization. It is likely that some changes in urbanization occurred during the years under study and the impact of both factors on elemental carbon might have changed somewhat over time. In the prediction model, temporal changes in the elemental carbon level over time in the region were approximated using elemental carbon measurements in Boston. The correlation of both was 0.4. Extrapolation of the model's prediction to previous years may have led to misclassification of exposure and thereby attenuated the effect estimates in our study.
Our results suggest that long-term traffic-related particulate air pollution is associated with increased mortality in hospital survivors of acute MI. This association appeared to occur only after the second year of survival after MI.
1.Dockery DW, Pope CA III, Xu X, et al. An association between air pollution and mortality in six US cities. N Engl J Med
2.Pope CA III, Burnett RT, Thun MJ, et al. Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. JAMA
3.Filleul L, Rondeau V, Vandentorren S, et al. Twenty five year mortality and air pollution: results from the French PAARC survey. Occup Environ Med
4.Gehring U, Heinrich J, Kramer U, et al. Long-term exposure to ambient air pollution and cardiopulmonary mortality in women. Epidemiology
5.Hoek G, Brunekreef B, Goldbohm S, Fischer P, van den Brandt PA. Association between mortality and indicators of traffic-related air pollution in the Netherlands: a cohort study. Lancet
6.Citation removed in proof.
7.Jerrett M, Burnett RT, Ma R, et al. Spatial analysis of air pollution and mortality in Los Angeles. Epidemiology
8.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
9.Pope CA III, Burnett RT, Thurston GD, et al. Cardiovascular mortality and long-term exposure to particulate air pollution: epidemiological evidence of general pathophysiological pathways of disease. Circulation
10.Chen LC, Nadziejko C. Effects of subchronic exposures to concentrated ambient particles (CAPs) in mice. V. CAPs exacerbate aortic plaque development in hyperlipidemic mice. Inhal Toxicol
11.Sun Q, Wang A, Jin X, et al. Long-term air pollution exposure and acceleration of atherosclerosis and vascular inflammation in an animal model. JAMA
12.Kunzli N, Jerrett M, Mack WJ, et al. Ambient air pollution and atherosclerosis in Los Angeles. Environ Health Perspect
13.Hoffmann B, Moebus S, Mohlenkamp S, et al. Residential exposure to traffic is associated with coronary atherosclerosis. Circulation
14.Zanobetti A, Schwartz J. Are diabetics more susceptible to the health effects of airborne particles? Am J Respir Crit Care Med
15.Bateson TF, Schwartz J. Who is sensitive to the effects of particulate air pollution on mortality? A case-crossover analysis of effect modifiers. Epidemiology
16.Pope CA III. Mortality effects of longer term exposures to fine particulate air pollution: review of recent epidemiological evidence. Inhal Toxicol
. 2007;19(suppl 1):33–38.
17.Pope CA III. Epidemiology of fine particulate air pollution and human health: biologic mechanisms and Who's at risk? Environ Health Perspect
. 2000;108(suppl 4):713–723.
18.Goldberg RJ, Yarzebski J, Lessard D, Gore JM. A two-decades (1975–1995) long experience in the incidence, in-hospital and long-term case-fatality rates of acute myocardial infarction: a community wide perspective. J Am Coll Cardiol
19.Goldberg RJ, Gore JM, Alpert JS, Dalen JE. Incidence and case fatality rates of acute myocardial infarction (1975–1984): the Worcester Heart Attack Study. Am Heart J
20.Gryparis A, Coull BA, Schwartz J, Suh HH. Semiparametric latent variable regression models for spatio-temporal modeling of mobile source particles in the greater Boston area. J R Stat Soc Ser C (Appl Stat)
21.Mesinger F, DiMego G, Kalnay E, et al. North American regional reanalysis. Bull Am Meteorol Soc
22.Ruppert D, Wand MP, Carroll RJ. Semiparametric Regression
. New York: Cambridge University Press; 2003.
23.Kammann EE, Wand MP. Geoadditive models. J R Stat Soc Ser C (Appl Stat)
24.Tonne C, Schwartz J, Mittleman M, Melly S, Suh H, Goldberg R. Long-term survival after acute myocardial infarction is lower in more deprived neighborhoods. Circulation
25.Krieger N, Chen JT, Waterman PD, Soobader MJ, Subramanian SV, Carson R. Geocoding and monitoring of US socioeconomic inequalities in mortality and cancer incidence: does the choice of area-based measure and geographic level matter? The Public Health Disparities Geocoding Project. Am J Epidemiol
26.Andersen PK, Gill RD. Cox's regression model for counting processes: a large sample study. Ann Stat
27.Schwartz J. The distributed lag between air pollution and daily deaths. Epidemiology
28.Therneau TM, Grambsch PM. Modeling Survival Data: Extending the Cox Model
. Berlin, Heidelberg: Springer; 2000.
29.D'Agostino RB, Lee ML, Belanger AJ, Cupples LA, Anderson K, Kannel WB. Relation of pooled logistic regression to time dependent Cox regression analysis: the Framingham Heart Study. Stat Med
30.Pope CA III, Thun MJ, Namboodiri MM, et al. Particulate air pollution as a predictor of mortality in a prospective study of US adults. Am J Respir Crit Care Med
. 1995;151(3 pt 1):669–74.
31.Elliott P, Shaddick G, Wakefield JC, de Hoogh C, Briggs DJ. Long-term associations of outdoor air pollution with mortality in Great Britain. Thorax
32.Finkelstein MM, Jerrett M, Sears MR. Traffic air pollution and mortality rate advancement periods. Am J Epidemiol
33.Beelen R, Hoek G, van den Brandt PA, et al. Long-term effects of traffic-related air pollution on mortality in a Dutch cohort (NLCS-AIR study). Environ Health Perspect
34.Botkin NF, Spencer FA, Goldberg RJ, Lessard D, Yarzebski J, Gore JM. Changing trends in the long-term prognosis of patients with acute myocardial infarction: a population-based perspective. Am Heart J
35.Zanobetti A, Schwartz J. Particulate air pollution, progression, and survival after myocardial infarction. Environ Health Perspect
36.Smith GD, Hart C, Watt G, Hole D, Hawthorne V. Individual social class, area-based deprivation, cardiovascular disease risk factors, and mortality: the Renfrew and Paisley Study. J Epidemiol Community Health
37.Diez Roux AV, Merkin SS, Arnett D, et al. Neighborhood of residence and incidence of coronary heart disease. N Engl J Med
38.Kunzli N, Tager IB. Air pollution: from lung to heart. Swiss Med Wkly
39.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
40.Haneuse S, Wakefield J, Sheppard L. The interpretation of exposure effect estimates in chronic air pollution studies. Stat Med
41.Jaffe AS, Krumholz HM, Catellier DJ, et al. Prediction of medical morbidity and mortality after acute myocardial infarction in patients at increased psychosocial risk in the enhancing recovery in coronary heart disease patients (ENRICHD) study. Am Heart J
42.Lauer MS, Blackstone EH, Young JB, Topol EJ. Cause of death in clinical research: time for a reassessment? J Am Coll Cardiol
43.Jerrett M, Finkelstein M. Geographies of risk in studies linking chronic air pollution exposure to health outcomes. J Toxicol Environ Health A
44.Cakmak S, Burnett RT, Jerrett M, et al. Spatial regression models for large-cohort studies linking community air pollution and health. J Toxicol Environ Health A