Exposure to fine particulate matter air pollution in Canada
Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please "contact us" to request a format other than those available.
by Lauren Pinault, Aaron van Donkelaar and Randall V. Martin
Fine particulate matter (PM2.5) is one of the primary components of air pollution. It refers to a mixture of particles less than 2.5 microns in diameter, including aerosols, smoke and dust. According to the Global Burden of Disease Study, PM2.5 air pollution is responsible for an estimated 2.9 million deaths worldwide each yearNote 1 and is associated with increased risk of non-accidental, circulatory and respiratory disease mortality.Note 2 In a study using the 1991 Canadian Census Health and Environment Cohort (CanCHEC),Note 3 this association was observed in Canada, a country where the level of ambient air pollution is relatively low. A subsequent study using the Canadian Community Health Survey (CCHS) and mortality cohort adjusted for behavioural covariates (for example, smoking) found an excess mortality risk at a lower concentration threshold.Note 4 The CCHS analysis used a fine-scale (1 km2) national model of PM2.5 , which provided more accurate exposure estimates than did previous models.
An important part of understanding the health effects of air pollution is describing how exposure differs across demographic and socioeconomic groups. In Canada, land use regression models developed for nitrogen dioxide (NO2) in urban centres allow researchers to assess within-city differences in exposure.Note 5Note 6Note 7Note 8 In large cities, NO2 exposure tends to be greater among lower socioeconomic status populations, although the groups affected are city-specific.Note 9Note 10Note 11Note 12 A national study in the United States reported greater NO2 exposure among non-white and low-income populations, particularly in urban centres.Note 13 The dataset developed for the CCHS studyNote 4 offers an opportunity to examine national patterns in PM2.5 exposure in Canada.
Specific groups such as recent immigrants, who tend to be healthier overall, may have greater exposure to PM2.5 (for instance, a disproportionate share may live in large cities). This creates challenges in estimating concentration-response associations for the entire Canadian population. Identifying groups with more or less exposure at a national level and the geographic regions where differences exist (urban versus rural) are useful in estimating health effects and associations between exposure and response. The present study (modelled methodologically on the U.S. NO2 study)Note 13 describes residential exposure to ambient PM2.5 by visible minority, immigrant and socioeconomic status in Canada, while stratifying the analysis across the urban-rural divide.
Data and methods
Characteristics of the Canadian population were derived from the 2006 Census long-form questionnaire, which was administered to a 20% sample of the population (except in some remote areas and Indian reserves where 100% were sampled).Note 14 The 2006 Census had a net undercoverage rate of 2.67%.Note 15 Institutional respondents were excluded from this study because they did not provide information on the characteristics of interest. Results from the long-form questionnaire were weighted to reflect the demographic and socioeconomic make-up of the entire population.Note 15
The variables considered in the study were: age; Aboriginal, visible minority and immigration status; years since immigration; household income; low-income status; labour force status (age 25 or older); and education (age 25 or older). Visible minorities are persons, other than Aboriginal people, who are non-Caucasian in race or non-white in colour.Note 14 Low-income status was derived from the low-income cut-offs (LICO), which are based on after-tax household income, household size and area of residence. LICO identifies people spending 20% more of their income on food, shelter and clothing than does the average household in their region.Note 14 The geography variables were urban core/urban fringe/rural area and Census Metropolitan Area (CMA)/Census Agglomeration (CA). The urban core is the urban area within a CMA or CA with a population of at least 50,000 (CMA) or 10,000 (CA). Urban fringe refers to the urban areas within a CMA/CA that are not contiguous with the main core. Rural is all other regions with a population density less than 400 persons per km.Note 2Note 14
The place of residence of non-institutional census respondents was mapped in a Geographic Information Systems (ArcGIS v. 10; ESRI) using the postal code reported on the census and Statistics Canada’s Postal Code Conversion File plus (PCCF+, v6B). The PCCF+ uses a population-weighted random allocation algorithm to provide a geographical representative point for postal codes.Note 16 Census respondents were spatially linked in GIS to gridded estimates of PM2.5 at 1 km2 resolution from a published dataset for 2006. The PM2.5 dataset was derived from column aerosol optical depth retrievals from the Moderate Resolution Imaging Spectroradiometer (MODIS) and related to near-surface PM2.5 using information from the GEOS-Chem chemical transport model.Note 17 Geographically weighted regression (incorporating the gridded estimates, land use data and ground monitoring data) was applied to produce a surface layer of yearly average PM2.5 at an approximately 1 km2 resolution.Note 17 Outlier PM2.5 estimates (values > 20 μg/m3; fewer than 1% of respondents) were excluded from the analysis, because they are likely due to data quality issues associated with satellite retrievals.
Estimates of mean (± standard deviation) PM2.5 exposure were derived for selected characteristics. Means were calculated for the total population and for urban core, urban fringe and rural populations. Graphs were used to examine changes in mean exposure by age over income, by community size, and for immigrants by years since immigration.
To compare differences in PM2.5 exposure among selected groups, Student’s t-test was used for significance testing. Because of the very large sample (and therefore, highly significant results), Cohen’s d effect size was used to interpret the magnitude of differences between groups.Note 18Note 19 Cohen’s d is independent of sample size; d = 1 signifies that group means differ by one standard deviation. Cohen’s approximate definitions of effect size are: small (d = 0.2), medium (d = 0.5) and large (d = 0.8).Note 18Note 19 To determine if differences in PM2.5 exposure exist in urban centres, a standard measure of inequality (difference in exposure between high-income whites (HIW) and low-income non-whites (LIN)Note 13 was calculated for all CMAs. A sensitivity analysis examined the urban population by residence in the Toronto CMA versus other urban cores.
Estimates of PM2.5 exposure were assigned to 6,306,700 non-institutional respondents to the 2006 Census. National mean (standard deviation) exposure was 7.05 (5.57) μg/m3, with a range of < 0.1 to 19.1 μg/m3.
The urban core estimate was 8.03 (4.90) μg/m3. Mean urban fringe and rural estimates were, respectively, 5.62 (4.24) and 4.32 (2.77) μg/m3―2.41 and 3.71 μg/m3 lower than the urban core (p < 0.001 for both; urban fringe: d = 0.53 and rural: d = 0.93) (Table 1). PM2.5 exposure estimates were very high in parts of southern Ontario (particularly around Toronto and Windsor) and in major cities (Map 1).
Unlike geographic variations, differences in PM2.5 exposure among demographic and socioeconomic groups were small. No obvious associations were apparent by age, except in urban cores where people aged 66 or older had slightly greater exposure than did other age groups (p < 0.001; d = 0.05). To examine this in more detail, mean (± 95% CI) PM2.5 exposure at each year of age was determined for urban respondents (Figure 1). Exposure was lowest during childhood, peaked at age 25 (8.25 μg/m3), declined throughout adulthood, and peaked again at age 80 (8.39 μg/m3). The difference in exposure between any two years of age was < 0.50 μg/m3.
Visible minorities and immigrants
Nationally, exposure of visible minorities to PM2.5 was 1.61 μg/m3 higher than that of the white population (p < 0.001; d = 0.32)(Table 1). However, almost all (96%) the visible minority population lived in urban cores, compared with 66% of the white population. When only urban cores were considered, the difference was reduced to 0.69 μg/m3 (p < 0.001; d = 0.15). The highest exposures in urban cores were among Latin American (8.95 μg/m3), Black (8.89 μg/m3) and Arab (8.87 μg/m3) residents.
PM2.5 exposure among Aboriginal people was 1.55 μg/m3 lower than that for white persons (p < 0.001; d = 0.33).
Immigrants’ PM2.5 exposure was 1.55 μg/m3 greater than that of non-immigrants (p < 0.001; d = 0.30) (Table 1). As was the case for visible minorities, a much larger percentage of immigrants (90%) than non-immigrants (65%) lived in urban cores. Among urban core residents, the difference in exposure between immigrants and non-immigrants was 0.82 μg/m3 (p < 0.001; d = 0.18) (Table 1). Urban core immigrants’ PM2.5 exposure peaked 1 year after they immigrated to Canada (8.94 μg/m3) and remained high (8.41 to 8.80 μg/m3), never approaching the mean for the non-immigrant urban core population (7.81 μg/m3) (Figure 2).
No strong associations emerged between household income and PM2.5 exposure overall. However, in urban cores, exposure was greater (0.56 μg/m3; p < 0.001; d = 0.11) for people in lower- versus higher-income households; in rural areas, their exposure was less (0.16 μg/m3 ; p < 0.001; d = 0.05) (Table 1). Exposure differed very little by labour force status or education.
Nationally, visible minority individuals in low-income households had the highest PM2.5 exposure: 8.86 μg/m3, which was 2.08 μg/m3 greater than that of white people who did not live in low-income households (p < 0.001; d = 0.42) (Table 1).
PM2.5 exposure of visible minority and white populations with similar household incomes was compared (Figure 3). Exposure did not differ greatly in urban fringe and rural areas, but in urban cores, at all household income levels, members of visible minorities had consistently higher exposure than did white individuals.
Census Metropolitan Areas
The difference in PM2.5 exposure between low-income visible minorities and high-income white people was calculated for each CMA. In five CMAs, greater exposure of low-income visible minorities was observed, but in the vast majority of CMAs, exposure was greater among high-income white people (Table 2).
Mean Toronto urban core PM2.5 exposure (9.33 μg/m3) surpassed the mean for urban cores outside Toronto (7.68 μg/m3) (Table 3). Among urban core residents, 43% who were members of visible minorities lived in Toronto, compared with 15% of those who were white. Similarly, 40% of urban core immigrants lived in Toronto versus 14% of urban core non-immigrants.
To disentangle the effect of Toronto, PM2.5 exposure for residents of Toronto’s urban core was compared with that of residents of other urban cores combined. In Toronto, exposure was generally similar between white and visible minority populations and immigrants and non-immigrants. However, the difference in PM2.5 exposure between lower- and higher-income residents was greater in Toronto than in other urban cores.
The national mean estimate of PM2.5 exposure in 2006―7.05 μg/m3―was somewhat below the estimate for the 2001-to-2006 period (8.7 μg/m3), which was based on an earlier PM2.5 model.Note 3
In the present study, PM2.5 exposure was 1.61 μg/m3 higher for visible minorities than for the white population, and 1.51 μg/m3 higher for immigrants than for non-immigrants. This disparity appears to be due, at least in part, to the relatively large percentages of visible minorities and immigrants living in urban cores. Immigrants’ exposure did not decline with years in Canada.
Associations between household income and exposure depended on location―in urban cores, PM2.5 was weakly positively associated with decreasing income. When visible minority status was also considered, exposure among low-income visible minority populations was 2.08 μg/m3 greater than among high-income white people. However, at the CMA level, exposure among low-income visible minorities was higher than among high-income white people in only a handful of cities.
A unique strength of this study is the largest population dataset in Canada (census long-form questionnaire). Respondents were point-matched to their location of residence and geocoded using postal code to a very fine-scale spatial model for PM2.5 (~1 km2). Population characteristics and PM2.5 estimates were derived for the same year (2006), and therefore, provide a relatively accurate cross-sectional match between the two datasets.
The findings highlight differences between the United States and Canada in patterns of exposure to air pollution. In the United States, although Clark et al. examined NO2 rather than PM2.5, they also reported greater exposure among visible minority (versus white) and lower-income groups.Note 13 As well, they observed a weakly positive association between household income and air pollution in rural areas, and a negative association in urban areas. However, in the United States at the regional, state, county and urban area levels, the difference in NO2 exposure between low-income non-white and high-income white people was usually positive and often relatively high.Note 13 By contrast, in the present study, differences between these groups were rare at the city level; differences at the national level appear to be largely due to concentration of visible minorities in urban centres with high levels of PM2.5.
Earlier studies of air pollution exposure using aggregate data and spatial regression techniques have been conducted for specific Canadian cities. Low income was associated with NO2 exposure in Toronto, Vancouver and Montreal, although other social and material deprivation indicators were also significant.Note 9Note 10Note 11 For example, greater NO2 exposure was observed in Montreal neighbourhoods with higher percentages of unemployed adults and persons living alone.Note 9 Not all these variables were considered in the present study of PM2.5 exposure, although associations with education and labour force status were not apparent at the national level. However, the geographic distributions of PM2.5 and NO2 are not the same, so it may not be appropriate to expect associations between socioeconomic characteristics and exposure to be the same. An analysis of national patterns in NO2 exposure would be useful to bridge the results of this study with previous Canadian work on NO2 and allow a better comparison with patterns in the United States. It is possible that considering NO2 rather than PM2.5 at the CMA level would reveal exposure patterns similar to those in the United States.Note 13
Small changes in PM2.5 concentrations can have substantial population health impacts. Pinault et al.Note 4 documented an association between non-accidental, circulatory and respiratory mortality and PM2.5 in a Canadian cohort adjusted for socioeconomic, ecological and behavioural covariates with a relatively low exposure distribution (mean = 6.3 μg/m3).
Concentration-response curves (and survival model relationships) from studies of the overall population may not be applicable to specific subpopulations. Differences in exposure are an important component in the framework of “triple jeopardy,” whereby people experiencing greater deprivation may have disproportionate air pollution-related health effects owing to a combination of: 1) lower socioeconomic status (and associated stress); 2) greater exposure to air pollution; and 3) effect modification of the health effects air pollution due to lower socioeconomic status, and possibly, greater stress.Note 9Note 20Note 21 It would be useful to examine concentration-response associations among groups identified in this study as having greater PM2.5 exposure to determine if they do, indeed, have a stronger response. In one study, a stronger concentration-response association was observed among Canadians in the lowest income quintile,Note 22 which suggests that effect modification may occur among other groups.
The approach to estimating PM2.5 exposure in this analysis has several limitations. The PCCF+ program is accurate to within a block face (a few households) in most urban areas, but less so in rural areas.Note 16 To some degree, this inaccuracy might be mitigated because PM2.5 is more uniformly low in rural regions than in urban centres. Additional research is needed to determine the extent of exposure misclassification in rural regions when using the PCCF+ program.
The use of residential point estimates for air pollutionNote 3Note 4Note 22 yields only a rough estimate of true exposure. Exposure occurs outside the home―for example, while commuting and in the workplace. Inclusion of workplace exposures might improve estimates and reduce exposure misclassification. Owing to data limitations, this study was also unable to account for occupational or behavioural differences that would increase air pollution exposure.
The study depended on PM2.5 estimates from a LUR model. While the model was validated using ground-based measurements, the model itself likely contributes to exposure misclassification, compared with direct measures.
In 2006, the national mean PM2.5 exposure was 7.05 μg/m3. The mean estimate for the urban core was 8.03 μg/m3, compared with 5.62 μg/m3 for the urban fringe and 4.32 μg/m3 for rural areas. Exposure estimates were very high in parts of southern Ontario (particularly around Toronto and Windsor) and in major cities. PM2.5 exposure was higher for visible minority (versus white) populations and for immigrants (versus non-immigrants). These exposure differences were smaller when residential location (for example, urban core) was considered. Exposure among urban immigrants did not decrease substantially with time since immigration. In urban cores, residents of low-income households had marginally higher exposure than did people who were not in low-income households.