Epidemiologic studies have shown positive associations of variations in air pollution between cities and cardiovascular mortality.1–3 Several studies have also found that individuals living in more polluted areas of cities are at higher risk of dying from cardiovascular disease.4–6 Road traffic is a major source of urban air pollution. Fine particulates generated from traffic and other sources have been implicated in triggering oxidative stress,7 activating systemic proinflammatory responses,8 increasing systemic arterial blood pressure,9 and promoting atherosclerosis10—all of which are possible biologic pathways to cardiovascular disease.1
The effects on cardiovascular mortality of long-term exposure to traffic-related air pollution have been investigated in only a few cohort studies, most of which relied on proximity to roads with heavy traffic,11,12 spatial interpolation of criteria air pollutants using fixed-site monitors,5 and air dispersion models.13,14 An expert panel15 has highlighted the weaknesses in these exposure methods; for example, spatial interpolation models capture regional patterns of pollution well but often fail to account for near-source impacts from local traffic. In view of the relatively few studies and the wide range of estimates of the association, this expert panel concluded that the evidence of the association between exposure to traffic pollution and cardiovascular mortality was not conclusive.15
Because traffic-related air pollution varies at small spatial scales, land-use regression (which provides more accurate estimates of the spatial variability of traffic pollutants)15,16 has become a preferred approach. Only a few cohort studies have used land-use regression models to investigate the associations between intraurban variations in traffic pollution and cardiovascular mortality.6,17,18 In addition, the induction period has not been investigated adequately, especially with respect to chronic versus acute exposures. Although longer term exposure over many years may be more hazardous than shorter term exposure,2 the underlying nature of the critical period of long-term exposure to air pollution responsible for the observed association is poorly understood1 and has been considered in only a few studies.19–21
We conducted a study of long-term exposure to traffic-related pollution in relation to cause-specific cardiovascular mortality among adults living in three cities in Ontario, Canada. We improved upon previous studies by using land-use regression models to assign exposure to nitrogen dioxide (NO2), a valid marker for traffic-related air pollution.15,16 In addition, we examined whether critical time windows of exposure affect rates of cardiovascular mortality.
The Ontario Tax Cohort
The Ontario Tax Cohort is a retrospective cohort study with approximately 660,000 participants who were followed from 1982 through 2004. Subjects were selected randomly from among Canadians who filed federal income tax returns and who were part of the Statistics Canada family income tax database. Canadian citizens or landed immigrants age 35–85 years were eligible if they had filed at least one tax return between 1982 and 1986 when they resided in any of 10 urban centers in Ontario. The income tax database is an excellent sampling frame of the Canadian adult population, containing about 98% of the population.22
Our analysis included 205,440 residents of three urban centers (Toronto, Hamilton, and Windsor) during 1982–1986. We used information on land-use regression models of NO2 concentrations23–25 to link estimates of NO2 exposure to subjects’ residences. Addresses were based on the Canadian six-character postal codes, which were also available on the income tax returns. The study was approved by the Research Ethics Boards of Health Canada and the University of Toronto.
Assessment of Exposure to Traffic-Related Air Pollution
The land-use regression models were derived from dense measurement campaigns of ground-level concentrations of NO2 conducted separately in each of the three cities.23–25 Briefly, in Hamilton and Toronto, a 2-week integrated sampling campaign of ambient NO2 was conducted using two-sided Ogawa passive diffusion samplers (Ogawa and Co., Pompano Beach, FL). In Hamilton, the Ogawa samplers were placed at 107 locations between 21 October and 6 November 2002.24 Another 30 locations in Hamilton were monitored in May 2004. The measured concentrations of NO2 for monitors at the same locations in 2002 and 2004 were reasonably well correlated (Pearson correlation coefficient [r] = 0.76).24 In Toronto, samplers were set up at 100 locations, and the sampling campaign was repeated in two seasons, one in the fall of 2002 (9 September to 24 September) and the other in the spring of 2004 (11 May to 26 May).17,23 Fifty monitors were deployed at one set of locations during both sampling periods, and the locations for 50 samplers were changed in the second round of monitoring. In Windsor, Maxxam all-season passive air samplers (Maxxam Analytics, Calgary, Edmonton) were used.25 Measurements were made in 2004 at 54 locations over four 2-week periods in February, May, August, and October.
Manual forward-selection regression procedures were used to select the best predictors of NO2. These predictors included an array of land use, traffic, physical geography, and population variables. The R2 of the land-use regression models were 70%, 76%, and 77% for Toronto, Hamilton, and Windsor, respectively. Because the measurements of NO2 were highly correlated between different seasons (Toronto: r = 0.8; Windsor: r = 0.7–0.9),17,25 we took the average of the two seasons in Toronto and the average of the four seasons in Windsor.
We created annual estimates of NO2 exposure for each subject by linking the exposure surfaces derived from the city-specific models to residential addresses recorded each year during follow-up, thereby accounting for residential mobility. The addresses were represented by six-character postal codes, which in urban areas represent a city block or a large apartment complex.
To compare with previous studies,11,12,26 we also calculated distances between subjects’ postal code addresses at the time of study entry and major traffic roads (primary urban roads, arterial roads), categorized as 0–50, 51–100, 101–200, 201–300, and >300 meters. As a sensitivity analysis, we also created a dichotomous exposure variable within 50 meters of a major road or 100 meters of a highway (see eAppendix, http://links.lww.com/EDE/A624).
Historical Estimates of Exposure in Toronto
Because the land-use regression models were developed near the end of the follow-up period, we applied methods that we developed previously to back-extrapolate the exposure surfaces.27 These methods take into account changes in the spatial structure of concentrations of NO2 that were captured by fixed-site monitors operating throughout the study period. As these methods require a large number of fixed-site monitoring sites, we could apply our methods only to subjects living in Toronto.
We used two related back-extrapolation methods to estimate concentrations of NO2 for 1982 and 1992. We made use of data from the fixed-site stations of Environment Canada’s National Air Pollution Surveillance Network.28 In the first method, we multiplied (1) the surface of NO2 concentrations produced from the land-use regression model from the fall of 2002 in each grid cell (5 meters × 5 meters) by (2) the ratio of two surfaces derived using first-order, inverse-distance weighted interpolation. The numerator represented mean concentrations of NO2 measured at six fixed-site monitors in the fall of 1982, and the denominator represented a similar surface for 2002. This process was repeated with data during the spring of 1982 and 2004. We then combined the two resulting exposure surfaces by taking an average at each grid cell, producing an estimated surface of concentrations of NO2 for 1982. We repeated the process for 1992, for which we used nine fixed-site monitors.
In the second method, we used data on land use and vehicular traffic by multiplying (1) the surface concentrations of NO2 produced by the land-use regression model for 2002 by (2) a ratio of estimates from two models. For the numerator, we developed a model by regressing the observed mean concentrations of NO2 at six fixed-site monitors in the fall of 1982 against a reduced set of spatial variables that were used to develop the 2002 model. We selected the variables using a supervised forward stepwise model selection procedure. For the denominator, we developed a model using as the dependent variable predicted concentrations at the six fixed-site monitors from the 2002 model. We then regressed these predicted values against the same covariables selected for the numerator. This process was repeated with data for spring 1982 and 2004. As in the first method, we then combined the two new exposure surfaces by taking an average at each grid cell, providing a surface of NO2 for 1982. As with the first method, we used nine fixed-site monitors for 1992.
Vital status was determined until the end of 2004 using a probabilistic record linkage to the Canadian Mortality Database.29 This database provides data on all deaths of Canadians occurring in Canada, and most of such deaths occurring in approximately 20 U.S. states. The accuracy of identifying deaths is about 98%.30 Until 1999, the underlying causes of death were coded to the ninth revision of the International Classification of Diseases (ICD-9); from 2000 onward, ICD-10 was used. Coronary heart disease was coded as ICD-9: 410–414 or ICD-10: I20–I25, cerebrovascular disease as ICD-9: 430–438 or ICD-10: I60–I69, and all cardiovascular diseases combined as ICD-9: 400–440 or ICD-10: I00–I99.
We used a stratified Cox proportional hazards model with strata defined as single-year age categories defined at date of entry. The time axis was calendar time starting from date of entry into the cohort. Those subjects who did not have a match on the mortality database were assumed to be alive at the end of follow-up.30 Subjects were censored if they no longer lived in any of the three cities.
For each city, we fitted separate Cox proportional hazards models. We we used proximity to roadways as a metric of exposure and we also modeled exposure to NO2, as estimated from the land-use regression models, as (1) annual average exposure at baseline; (2) annual average exposure during the year preceding the time of the event; (3) mean annual exposure over the 3 years before the event; (4) cumulative mean annual exposure from the time of entry until the event; and (5) temporal, orthogonal decomposition of exposure into cumulative mean annual exposure and the difference between annual average exposure during the year before the event and cumulative mean annual exposure. The motivation for this last exposure metric was to account for strong temporal correlations in exposures (eg, locations with high concentrations of NO2 in a given year likely remained high over the long-term), and thus annual mean exposure in the follow-up represents a combination of long-term and short-term components. The regression coefficients of the two components in this fifth model thus approximate the independent effects of long-term and short-term exposures.
In all of the Cox models, we adjusted for sex; marital status; individual-level annual household income (quintiles); and four area-level variables: percentage of immigrants (quintiles), percentage of adults with less than high school education (continuous), unemployment rate (continuous); and average household income (quintiles). The area-level covariables were derived from the 1981 Canadian Census using census-enumeration areas. The enumeration area is a small, relatively homogeneous geographic unit composed of one or more blocks (400–700 persons). For each model, we tested deviations from the proportional hazards assumption by assessing whether the cross-product of each variable with the natural logarithm of the time variable was statistically important.31 We also verified the assumption of linearity for continuous exposure and confounding variables by adding a quadratic term. As all associations with NO2 were linear, we report adjusted rate ratios (RRs) and their associated 95% confidence intervals (CIs) for an increase in NO2 concentrations of 5 parts per billion (ppb). We chose an increment of 5 ppb to allow pooling of city-specific effect estimates and to enable more direct comparison of our findings with other studies.17,32
To obtain estimates of effects for the entire study population, we pooled the city-specific effect estimates using a fixed-effect model with weights equal to the inverse variance of the city-specific estimates. We verified the assumption of homogeneity using a chi-square test.
In addition, we investigated whether age (<45 years, 45–54 years, 55–65 years, >65 years), sex, marital status, and individual-level household income (quintiles) modified the associations between traffic-related exposures and cardiovascular mortality.
Finally, to assess the impact of potential confounding introduced by unmeasured tobacco smoking and obesity, we performed Monte Carlo sensitivity analyses proposed by Steenland and Greenland.33 In their original derivation, dichotomous exposure variables were used; this method was further extended by Villeneuve et al34 to allow for continuously measured exposure variables (see eAppendix, http://links.lww.com/EDE/A624). To estimate area-specific prevalence of smoking and body mass index (BMI; weight [kilogram]/height [meter]2), we made use of data from the nationally representative Canadian Community Health Survey conducted in 2001.35 We used two methods: (1) probability of smoking conditional on NO2 and (2) probability of smoking conditional on NO2, age, sex, income, and other covariables, to obtain a plausible range of the prevalence of smoking. We also applied these two methods to BMI, which was modeled as a categorical variable. To estimate adjusted RRs for smoking and BMI, we made use of the Second American Cancer Society Cancer Prevention Study.36 We repeated these sensitivity analyses for each of the three causes of death and for subjects in each of the three cities.
A total of 77,890 subjects in Hamilton, 58,700 in Toronto, and 68,850 in Windsor were included in the study (Table 1). At the time of entry, the mean age of the cohort in the three cities was 52 years, 50% were men, and approximately two-thirds were married. The mean annual household incomes of the participants at entry to the cohort were $Cdn35,600, $43,500, and $34,200 in Hamilton, Toronto, and Windsor, respectively. According to the 1981 Census, the percentage of the population 15 years and older having more than high school education was 55% in Toronto and about 50% in Hamilton and Windsor. The average percentage of immigrants in the neighborhoods of subjects was 40% in Toronto, 28% in Hamilton, and 24% in Windsor.
We had over 4 million person-years of follow-up. There were 18,360, 12,410, and 17,360 nonaccidental deaths among subjects living at time of entry in Hamilton, Toronto, and Windsor, respectively (Table 2). About 40% of these deaths were attributed to cardiovascular disease.
The mean distributions of NO2 at subjects’ homes were as follows: Toronto, 21.7 ppb (interquartile range = 4.1 ppb); Hamilton, 15.5 ppb (2.9 ppb); and Windsor, 12.1 ppb (2.5 ppb). In Hamilton and Toronto, which had multiple fixed-site monitors, the spatial distributions of NO2 and the rank ordering of the fixed-site monitors did not change appreciably over the follow-up period (see eAppendix, http://links.lww.com/EDE/A624). In Toronto, there were enough monitors to use back-extrapolation methods, and we found using these methods that means and variances of NO2 were similar during the follow-up (approximately 24 ppb) (see eAppendix, http://links.lww.com/EDE/A624).
Table 3 shows adjusted RRs per an increase in NO2 of 5 ppb, using the NO2 estimates from the land-use regression models. Adjusting for all personal variables, the RRs for all cardiovascular disease were as follows: Hamilton, 1.16 (95% CI = 1.10–1.22); Toronto, 1.01 (0.98–1.05); and Windsor, 1.13 (1.06–1.22). There was considerable heterogeneity among these estimates. The addition of the contextual variables slightly reduced the strength of the association between NO2 and cardiovascular mortality in Hamilton and Windsor, but increased it in Toronto. With this adjustment, the RRs were no longer heterogeneous (pooled RR = 1.08 [95% CI = 1.05–1.11]).
For ischemic heart disease, the estimated effects were similar to those for all cardiovascular mortality (Table 3). There was no important heterogeneity in the fully adjusted estimates, and the pooled RR was 1.09 (95% CI = 1.04–1.14). In contrast, there was no association between NO2 and cerebrovascular disease-related mortality (pooled RR = 0.96 [95% CI = 0.90–1.05]).
Indirect adjustment for smoking yielded a pooled RR (Table 4) for cardiovascular mortality ranging from 1.04 (1.00–1.09) to 1.07 (1.02–1.11), similar to estimates not adjusted for smoking. Similar estimates of effect were found for mortality from ischemic heart disease. BMI (as a categorical variable) was found similar across levels of NO2, and so the indirect method would not alter the adjusted RRs (see eAppendix, http://links.lww.com/EDE/A624).
Consistent with the above observations, Figure 1 shows a clear inverse response pattern according to distance from a major roadway for all deaths attributable to cardiovascular diseases and those attributable to ischemic heart disease. Figure 2 shows risk estimates of cardiovascular mortality per 5-ppb increase in NO2 in four time windows. The strongest associations were with long-term cumulative exposure, especially for ischemic heart disease (pooled RR = 1.15 [95% CI = 1.08–1.21]). In contrast, the RRs relating ischemic heart disease to exposure in the preceding year and exposure from the 3-year moving average were 1.10 (1.04–1.16) and 1.09 (1.04–1.14), respectively. The risk estimates for annual average exposure assigned at baseline were similar in magnitude and direction to the effects of long-term exposure and the other metrics of exposure. Although we attempted to decompose exposure into short-term and long-term components, we were unable to obtain stable estimates attributable to highly inflated variances (ie, the variables were collinear) (data not shown). There was no evidence of effect modification by age, sex, marital status, and individual-level household income. (The P values from likelihood ratio tests comparing models with and without the interaction term varied from 0.11 to 0.67.)
We conducted a population-based cohort study of cardiovascular mortality in relation to NO2 exposure, with a 23-year follow-up of 205,440 adults in three cities in Ontario. We found that NO2, which is a valid marker for traffic-related air pollution, was positively associated with mortality from cardiovascular disease. RRs varied between 1.04 (95% CI = 1.00–1.09) and 1.12 (1.07–1.17) per 5-ppb increase in NO2, with long-term cumulative exposure more strongly associated with mortality than short-term exposure. We did not find any associations with cerebrovascular disease. Risk estimates for subjects’ exposure at time of entry were similar to their long-term exposure, suggesting that exposure at the baseline was a valid proxy for long-term exposure.
Long-term cumulative exposure tended to be a stronger predictor of mortality from cardiovascular disease, but we were unable to separate the effects of short-term and long-term exposure. This was attributable to much greater spatial than temporal variability (see eAppendix, http://links.lww.com/EDE/A624).
Our findings are similar to the results of a recent cohort study carried out in Vancouver, Canada.32 Gan et al32 estimated rates of mortality from ischemic heart disease in a cohort composed of all residents aged 45–85 years in Vancouver (n = 452,735) over a 4-year period. They reported that an increase of 8.4 µg/m3 (approximately 4.5 ppb) of NO2 was associated with a 4% increased risk of mortality from ischemic heart disease (95% CI = 1%–8%). In another Canadian cohort study, Jerrett et al17 showed that a 4-ppb increase in NO2 increased cardiovascular disease mortality by 40% (95% CI = 5%–86%). This last cohort included nearly half of subjects diagnosed with ischemic heart disease and 30% had chronic obstructive pulmonary disease; thus, the cohort represented a subgroup of the population susceptible to the effects of air pollution.
Similar to previous studies,11,12 we observed positive associations between living close to a major road and cardiovascular mortality. This finding is consistent with the association found using metrics derived from the land-use regression models of NO2. Because these latter estimates take into account proximity to roadways, topography, meteorologic factors that affect dispersion, and vehicle mix, we expect the land-use regression models to be more accurate estimates of traffic-related air pollution than proximity alone.15,16
Our study has several strengths. First, it was population-based and had statistical power to detect small effects. Second, we obtained individual residential histories from annual federal income tax files between 1982 and 2004. This improved locational accuracy was combined with high-resolution estimates of the concentrations of NO2 derived from land-use regression models. We further incorporated temporal variability in the spatial structure of NO2 from measurements at fixed-site monitors. This integrated approach should have reduced measurement error. Third, household income, a risk factor for cardiovascular disease, was measured accurately because it was obtained directly from the tax file. In addition, there was little impact of smoking and BMI on the risk estimates.
A limitation of the study is the use of death certificates to ascertain cardiovascular outcomes. As shown previously,37 cardiovascular diseases and coronary heart disease may be overestimated on death certificates, although it is unclear whether this applies to the present study. Nevertheless, this measurement error was likely independent of exposure, leading to an underestimation of the true effects.
A second limitation is that the land-use regression models were developed toward the end of our study and might not characterize exposure adequately during earlier periods of our study. We verified that variability in the NO2 concentrations in Toronto during the observation period was primarily spatial rather than temporal (see eAppendix, http://links.lww.com/EDE/A624). Thus, we expect the spatial contrast in NO2 using the land-use regression models provided reasonable estimates of long-term spatial exposure. Similar findings have been reported in previous studies; for example in Montreal, Canada38; in California19; in North Rhine Westphalia, Germany39; and in the Netherlands.40 We expect that the land-use regression models developed for Hamilton and Windsor would also reflect the long-term spatial pattern of NO2. Because there were few fixed-site monitors in the two cities, we were unable to extrapolate their land-use regression models back in time, and thus examine this issue adequately.
Furthermore, we did not have information on subjects’ residences before they entered the study and thus assumed that the subjects’ previous exposures were similar to those during the follow-up period. In addition, it was not possible to assess personal exposures to ambient air pollution. Other factors such as daily activity patterns may have an important influence on total personal exposure to traffic-related air pollution. A population survey reported that Canadian adults who resided in major cities during 1992 and 1997 spent on average over 80% of their time each year at home (both indoors and outdoors).41 Other studies have shown strong correlations between indoor and outdoor levels of NO2; for example, indoor-outdoor concentration ratios for NO2 varied from 0.88 to 1.42 As a result, it is reasonable to use spatially derived exposures at the residences as a surrogate for personal exposures. Nonetheless, given the inherent imprecision of these spatially derived exposures, our assessment of exposure was likely subject to nondifferential misclassification that would attenuate our estimates. Although our study showed a positive association between NO2 and cardiovascular outcomes, our findings should not be interpreted as meaning that NO2 is the causal agent, but rather a marker for traffic-related pollution.
A third limitation, shared by many cohort studies using administrative databases, was that we lacked information on individual risk factors for cardiovascular diseases such as smoking or obesity. Indirect adjustment for smoking or BMI had no appreciable impact on the RRs of cardiovascular mortality, because the prevalences of these factors were weakly correlated with ambient concentrations of NO2. We have previously shown in these data34 that participants living in areas of Ontario with higher concentrations of PM2.5 tended to smoke less and that smoking had a negative confounding effect on the association between PM2.5 and mortality. This is consistent with previous studies in which adjustment for smoking was shown to have virtually no impact on the risk estimates.20,43
The accuracy of our method of indirect adjustment depends on the true personal distributions of the prevalence of smoking and the relative risks for smoking. The Canadian Community Health Survey was used to obtain information from a large probability sample of all households in Canada; thus, it provided reliable estimates of the prevalence of smokers in our study area. To improve accuracy, we restricted data from the survey to those aged 35–85 years. In addition, we assumed that the impact of smoking on the association of air pollution and cardiovascular mortality is approximately constant across the covariates. Furthermore, our method of indirect adjustment may overadjust the effects of smoking because this adjustment was applied to the survival models that already included many correlates of smoking such as age, income, and marital status.
We were unable to examine the potential confounding effects of other risk factors such as high blood pressure and diabetes. Although we could not rule out potential residual confounding by these risk factors, it is likely that their confounding effects are small because adjustment for other variables such as age, sex, income, and smoking might in part control for the confounding effect of these unmeasured risk factors. Indeed, in the cohort study in Toronto, Jerrett et al17 showed that adjustment for smoking, diabetes, preexisting chronic obstructive pulmonary disease, and chronic ischemic heart disease resulted in a decrease of 5% in the estimate of effects of NO2 on cardiovascular mortality.17 In many other studies,44,45 these variables did not appear to be important confounders. Another limitation of the study is the lack of information on occupational exposures, although it is unclear whether such exposures would be associated with traffic-related pollution.46
We did not see a consistent pattern for the association between long-term exposure to NO2 and cerebrovascular-related mortality. The fewer number of deaths from cerebrovascular disease may have affected our ability to detect small associations. The observed null associations may also have been caused by the heterogeneity of cerebrovascular disease. Etiologically, cerebrovascular disease is far more diverse than ischemic heart disease, including both hemorrhagic and occlusive strokes, which in turn are further subdivided into distinct subtypes.47 As a result, it is plausible that different types of cerebrovascular disease may follow different biologic pathways. Whether there are important differences in the impact of traffic-related pollution on different types of cerebrovascular disease remains unresolved.
1. Brook RD, Rajagopalan S, Pope CA 3rd, et al.. American Heart Association Council on Epidemiology and Prevention, Council on the Kidney in Cardiovascular Disease, and Council on Nutrition, Physical Activity and Metabolism. Particulate matter air pollution and cardiovascular disease: an update to the scientific statement from the American Heart Association. Circulation. 2010;121:2331–2378
2. Pope CA 3rd, Dockery DW. Health effects of fine particulate air pollution: lines that connect. J Air Waste Manag Assoc. 2006;56:709–742
3. Chen H, Goldberg MS, Villeneuve PJ. A systematic review of the relation between long-term exposure to ambient air pollution and chronic diseases. Rev Environ Health. 2008;23:243–297
4. 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
5. Jerrett M, Burnett RT, Ma R, et al. Spatial analysis of air pollution and mortality in Los Angeles. Epidemiology. 2005;16:727–736
6. 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. 2008;116:196–202
7. Bräuner EV, Forchhammer L, Møller P, et al. Exposure to ultrafine particles from ambient air and oxidative stress-induced DNA damage. Environ Health Perspect. 2007;115:1177–1182
8. Riediker M. Cardiovascular effects of fine particulate matter components in highway patrol officers. Inhal Toxicol. 2007;19(suppl 1):99–105
9. Van Hee VC, Adar SD, Szpiro AA, et al. Exposure to traffic and left ventricular mass and function: the Multi-Ethnic Study of Atherosclerosis. Am J Respir Crit Care Med. 2009;179:827–834
10. Hoffmann B, Moebus S, Möhlenkamp S, et al.. Heinz Nixdorf Recall Study Investigative Group. Residential exposure to traffic is associated with coronary atherosclerosis. Circulation. 2007;116:489–496
11. Finkelstein MM, Jerrett M, Sears MR. Environmental inequality and circulatory disease mortality gradients. J Epidemiol Community Health. 2005;59:481–487
12. Gehring U, Heinrich J, Krämer U, et al. Long-term exposure to ambient air pollution and cardiopulmonary mortality in women. Epidemiology. 2006;17:545–551
13. Rosenlund M, Bellander T, Nordquist T, Alfredsson L. Traffic-generated air pollution and myocardial infarction. Epidemiology. 2009;20:265–271
14. 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
15. HEI Panel on the Health Effects of Traffic-Related Air Pollution.Traffic-Related Air Pollution: A Critical Review of the Literature on Emissions, Exposure, and Health Effects. HEI Special Report 17. 2010 Boston, MA Health Effects Institute
16. Hoek G, Beelen R, de Hoogh K, et al. A review of land-use regression models to assess spatial variation of outdoor air pollution. Atmos Environ. 2008;42:7561–7578
17. Jerrett M, Finkelstein MM, Brook JR, et al. A cohort study of traffic-related air pollution and mortality in Toronto, Ontario, Canada. Environ Health Perspect. 2009;117:772–777
18. Rosenlund M, Picciotto S, Forastiere F, Stafoggia M, Perucci CA. Traffic-related air pollution in relation to incidence and prognosis of coronary heart disease. Epidemiology. 2008;19:121–128
19. Jerrett M, Burnett RT, Pope CA 3rd, et al. Spatiotemporal analysis of air pollution and mortality in California based on the American Cancer Society Cohort: Final Report. 2011. Available at: http://www.arb.ca.gov/research/rsc/10-28-11/item1dfr06-332.pdf
. Accessed 5 March 2012.
20. Puett RC, Schwartz J, Hart JE, et al. Chronic particulate exposure, mortality, and coronary heart disease in the nurses’ health study. Am J Epidemiol. 2008;168:1161–1168
21. Villeneuve PJ, Goldberg MS, Krewski D, Burnett RT, Chen Y. Fine particulate air pollution and all-cause mortality within the Harvard Six-Cities Study: variations in risk by period of exposure. Ann Epidemiol. 2002;12:568–576
22. Spasoff RA, Gilkes DT. Up-to-date denominators: evaluation of taxation family file for public health planning. Can J Public Health. 1994;85:413–417
23. Jerrett M, Arain MA, Kanaroglou P, et al. Modeling the intraurban variability of ambient traffic pollution in Toronto, Canada. J Toxicol Environ Health Part A. 2007;70:200–212
24. Sahsuvaroglu T, Arain A, Kanaroglou P, et al. A land use regression model for predicting ambient concentrations of nitrogen dioxide in Hamilton, Ontario, Canada. J Air Waste Manag Assoc. 2006;56:1059–1069
25. Wheeler AJ, Smith-Doiron M, Xu X, Gilbert NL, Brook JR. Intra-urban variability of air pollution in Windsor, Ontario—measurement and modeling for human exposure assessment. Environ Res. 2008;106:7–16
26. Hoffmann B, Moebus S, Kröger K, et al. Residential exposure to urban air pollution, ankle-brachial index, and peripheral arterial disease. Epidemiology. 2009;20:280–288
27. Chen H, Goldberg MS, Crouse DL, et al. Back-extrapolation of estimates of exposure from current land-use regression models. Atmos Environ. 2010;44:4346–4354
28. Environment Canada. National Air Pollution Surveillance Network (NAPS). Available at: http://www.ec.gc.ca/natchem/default.asp?lang=en&n=EE0E2169-1
. Accessed 31 March 2011.
30. Goldberg MS, Carpenter M, Thériault G, Fair M. The accuracy of ascertaining vital status in a historical cohort study of synthetic textiles workers using computerized record linkage to the Canadian Mortality Data Base. Can J Public Health. 1993;84:201–204
31. Kleinbaum DG Survival Analysis: A Self-Learning Text. 1996 New York, NY Springer-Verlag
32. 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
33. Steenland K, Greenland S. Monte Carlo sensitivity analysis and Bayesian analysis of smoking as an unmeasured confounder in a study of silica and lung cancer. Am J Epidemiol. 2004;160:384–392
34. Villeneuve PJ, Goldberg MS, Burnett RT, van Donkelaar A, Chen H, Martin RV. Associations between cigarette smoking, obesity, sociodemographic characteristics and remote-sensing-derived estimates of ambient PM2.5: results from a Canadian population-based survey. Occup Environ Med. 2011;68:920–927
35. Statistics Canada. Canadian Community Health Survey (CCHS). Available at: http://www.statcan.gc.ca/cgi-bin/imdb/p2SV.pl?Function=getSurvey&SDDS=3226&lang=en&db=imdb&adm=8&dis=2
. Accessed 28 January 2011.
36. Thun MJ, Apicella LF, Henley SJ. Smoking vs other risk factors as the cause of smoking-attributable deaths: confounding in the courtroom. JAMA. 2000;284:706–712
37. Lloyd-Jones DM, Martin DO, Larson MG, Levy D. Accuracy of death certificates for coding coronary heart disease as the cause of death. Ann Intern Med. 1998;129:1020–1026
38. Crouse DL, Goldberg MS, Ross NA, Chen H, Labrèche F. Postmenopausal breast cancer is associated with exposure to traffic-related air pollution in Montreal, Canada: a case-control study. Environ Health Perspect. 2010;118:1578–1583
39. Krämer U, Herder C, Sugiri D, et al. Traffic-related air pollution and incident type 2 diabetes: results from the SALIA cohort study. Environ Health Perspect. 2010;118:1273–1279
40. Eeftens M, Beelen R, Fischer P, Brunekreef B, Meliefste K, Hoek G. Stability of measured and modelled spatial contrasts in NO(2) over time. Occup Environ Med. 2011;68:765–770
41. Leech JA, Nelson WC, Burnett RT, Aaron S, Raizenne ME. It’s about time: a comparison of Canadian and American time-activity patterns. J Expo Anal Environ Epidemiol. 2002;12:427–432
42. Blondeau P, Iordache V, Poupard O, Genin D, Allard F. Relationship between outdoor and indoor air quality in eight French schools. Indoor Air. 2005;15:2–12
43. 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
44. 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
45. Yorifuji T, Kashima S, Tsuda T, et al. Long-term exposure to traffic-related air pollution and mortality in Shizuoka, Japan. Occup Environ Med. 2010;67:111–117
46. Health Effects Institute. Reanalysis of the Harvard Six Cities Study and the American Cancer Society Study of Particulate Air Pollution and Mortality: A Special Report of the Institute’s Particle Epidemiology Reanalysis Project. 2000 Boston, MA Health Effects Institute
47. O’Donnell M, Xavier D, Diener C, et al.. INTERSTROKE Investigators. Rationale and design of INTERSTROKE: a global case-control study of risk factors for stroke. Neuroepidemiology. 2010;35:36–44