Health Reports
The association between ambient air pollution concentrations and psychological distress

by Lauren Pinault, Errol M. Thomson, Tanya Christidis, Ian Colman, Michael Tjepkema, Aaron van Donkelaar, Randall V. Martin, Perry Hystad, Hwashin Shin, Daniel L. Crouse, and Richard T. Burnett

Release date: July 29, 2020


In addition to having well-established associations with respiratory and cardiovascular morbidity and mortality,Note 1Note 2 air pollution has been linked to a range of neurological and psychiatric disorders, including dementia,Note 3 cognitive decline or impairment,Note 4Note 5 anxiety and depression,Note 6Note 7Note 8Note 9 and suicide.Note 10Note 11 A growing body of controlled experimental research supports these epidemiological associations. This research includes evidence that air pollution can impair spatial learning and memory, and provoke depressive-like behaviour in mice.Note 12 A number of biological mechanisms have been proposed to explain these associations.Note 13 For example, exposure to inhaled pollutants has been shown to activate the hypothalamic-pituitary-adrenal (HPA) stress axis in rats, resulting in increased blood levels of the glucocorticoid corticosterone, and systemic regulation of stress, metabolic, and inflammatory pathways across multiple tissues.Note 14Note 15Note 16 Acute activation of the HPA axis is a critical adaptive response to stressors. However, chronic stress and glucocorticoid dysregulation are associated with many disease processes, including anxiety, depression, impaired cognition, chronic pain and fatigue syndrome, obesity, diabetes, and cardiovascular disease.Note 17 Chronic exposure to air pollution—in conjunction with exposure to other stressors and interaction with host susceptibility traits—could lead to HPA axis dysfunction and observable stress-related or distress-related outcomes, including neurological, metabolic, cardiovascular and reproductive disorders.Note 18 Accordingly, stress-related outcomes in the causal pathway between air pollutant exposure and disease should be investigated to substantiate the link between pollutant-dependent neuroendocrine stress responses and impacts on the brain.Note 19

Psychological distress, formally assessed using the Kessler Psychological Distress Scale (K10), is a measure of common symptoms of depression and anxiety, and has been suggested as an objective measure of the effect of stress on mental health.Note 20Note 21Note 22 In rats, acute administration of corticosterone causes anxiety,Note 23 while repeated exposure produces a model of depression,Note 24 which suggests a causal relationship between glucocorticoids and neurobehavioural outcomes. Moreover, reductions in cortisol levels in individuals with generalized anxiety disorder are associated with improvements in anxiety.Note 25 Thus, it is plausible that exposure to air pollution could increase distress, possibly through effects on the HPA axis. At the same time, associations between psychological distress and air pollution may be relevant to the identification of vulnerable populations, independent of a potential causal relationship.

Several exploratory studies have been undertaken to determine whether distress or related mental health conditions are correlated with exposure to air pollution, with mixed results. A study of 6,000 adults in the United States found a positive association between distress and exposure to ambient fine particulate air pollution (PM2.5), after adjusting for personal characteristics, health behaviours and neighbourhood poverty.Note 26 However, other studies of European cohorts have reported mixed associations (null, positive and negative) between depressed mood and both nitrogen dioxide (NO2) and PM2.5 air pollution.Note 27Note 28 These associations have not yet been assessed in Canada.

The purpose of this study was to examine the association between psychological distress and outdoor air pollution using a cross-sectional spatial analysis. The examination of a spatial association between air pollutants and distress has relevance both as a possible link in the pathway between exposure and disease, and as a method to identify potential vulnerable populations.

Materials and methods

Data and linkage

The Canadian Community Health Survey (CCHS) is a national survey of the health of Canadians aged 12 years or older, conducted every two years from 2000 to 2007 and annually thereafter. Institutional respondents, persons residing on Indian reserves and in Aboriginal settlements, and full-time members of the Canadian Forces are excluded from the CCHS (exclusions represent less than 3% of the target population).Note 29 In addition to questions about demographic and socioeconomic characteristics, the survey includes questions about selected health behaviours (e.g., alcohol consumption) in all cycles.

An optional module for distress was selected by specific provinces in different cycles, and it included a formal assessment using the K10. The K10 is a 10-item self-reported measure of common depression and anxiety symptoms that have occurred over the previous month.Note 20 Respondents were asked, “During the past month, about how often did you feel: (1) tired out for no good reason, (2) nervous, (3) so nervous that nothing could calm you down, (4) hopeless, (5) restless or fidgety, (6) so restless you could not sit still, (7) sad or depressed, (8) so depressed that nothing could cheer you up, (9) that everything was an effort, or (10) worthless.” Respondents answered each of these questions using a five-point scale that ranged from “none of the time” to “all of the time.”

An ordinary least squares (OLS) regression model assumes error terms to be independent. This assumption is often violated when data are geographically ordered. Often, air pollution and population characteristics are spatially autocorrelated since similar characteristics can be more closely clustered in space. When spatial autocorrelation is detected in the residuals, a spatial analysis approach is warranted. A spatially weighted regression is most appropriate when datasets are both spatially contiguous and cross-sectional, and when the respondent population is large enough for locally based analyses. Therefore, study populations that had the largest sample of distress module respondents, by province and cycle, were selected. All cycles from 2005/2006 (cycle 3.1) to 2011/2012 were considered for inclusion since there were also fine-scale air pollution data available for these years. The largest samples were derived from Quebec (n=29,900), Alberta (n=11,800) and British Columbia (n=15,400) in 2005/2006 (cycle 3.1), and Ontario (n=42,900) in 2011/2012. Respondents from Alberta and British Columbia were combined into a single dataset since they were contiguous in space and time. Once children (aged 17 years or younger) and those who provided an unclear response to the stress and distress questions (e.g., “don’t know,” “not stated”) were excluded, the final sample sizes were 25,800 in Quebec, 23,000 in Alberta and British Columbia, and 36,000 in Ontario. Estimates were rounded to the nearest hundred for institutional confidentiality.

Air pollution data were derived from published national surface models for PM2.5,Note 30 NO2Note 31 and ozone (O3).Note 32 PM2.5 concentrations were derived from total column optical depth retrievals from the Moderate Resolution Imaging Spectroradiometer satellite instrument, related to near-surface observations from the GEOS-Chem chemical transport model, and further calibrated with local ground-based monitors using a geographically weighted regression. Final results demonstrated an R2 of 0.82 with long-term mean ground-based monitors over North America.Note 30 Outliers greater than 20 μg/m3 were excluded from the analysis.

NO2 concentrations were estimated using the 2006 annual mean from a national land-use regression model that used National Air Pollution Surveillance (NAPS) fixed-site monitoring data combined with satellite NO2 estimates, road length within 10 km, industrial land-use areas within various buffers, and mean summer rainfall.Note 31 Ground-level NO2 estimates were derived using the GEOS-Chem chemical transport model from satellite tropospheric NO2 columns.Note 33 During validation, the model explained 73% of the variance in 2006 NAPS estimates. Local variation in near-roadway NO2 was captured by applying kernel density measures of highways and major roads within 300 m and 100 m as a multiplier to the model.Note 31

O3 concentrations were derived from average eight-hour daily maximum concentrations from May to September from 2002 to 2009 at a 10 to 21 km2 grid scale.Note 32 O3—estimated by the air quality forecast model based on the Canadian Hemispheric and Regional O3 and NOx System—was interpolated to generate the surface.Note 34

NO2 and O3 data were year-adjusted using time series measurements from 24 census divisions (CDs) between 1981 and 2012.Note 35 Missing air pollution data were imputed using an interpolation algorithm that combined classical prediction techniques and phase and frequency fitting tools using the multitaper method.Note 36 For each CD time series, a cubic spline was fit to model the association between year and air pollutant. Then, the ratio between the year of the original modelled data and the year or years of the CCHS dataset was determined. Respondent residence locations were matched to the closest CD using a Geographic Information System (ArcGIS version 10, Environmental Systems Research Institute, 2010), and the corresponding time adjustment ratio was used to adjust data for annual differences in concentration.

Postal codes were used to spatially link respondents in each of the three regional datasets to the corresponding years of data for PM2.5, NO2 and O3. Respondent postal codes were used to produce approximate residential locations using Statistics Canada’s Postal Code Conversion File Plus (PCCF+), version 6D. PCCF+ uses a population-weighted random allocation algorithm to assign postal codes to representative points based on census geography, and it is most accurate in urban centres.Note 37

Statistical models and covariates

For Ontario, Quebec, Alberta and British Columbia, OLS regression models were used to estimate the overall association between air pollutants and K10 distress scores, after adjusting for age, sex, individual socioeconomic position (SEP) (i.e., marital status: married or common law vs. not married or common law; income quintile: lowest vs. all other quintiles; employment status: employed vs. unemployed), health behaviours (i.e., alcohol consumption: drinks five or more drinks on one occasion, more than once a month vs. all others; and current smoking: current daily smoker vs. not current daily smoker) and neighbourhood marginalization. K10 distress scores were not normally distributed and were skewed towards lower scores, so in order to meet the assumptions of normality, they were transformed prior to analysis using a normalized z-score. All individual variables except age (i.e., sex, SEP and behaviour variables) were dichotomized for the full regression analysis, as described above, since the spatial regression technique does not consider categorical variables.

The Canadian Marginalization Index (CAN-Marg) was used to characterize neighbourhood-level deprivation at the dissemination-area scale. The CAN-Marg was used because it includes many SEP covariates that cover four dimensions (i.e., residential instability, material deprivation, dependency and ethnic concentration),Note 38 and because some of its factors were originally developed to capture psychological depression at a neighbourhood scale.Note 39 The CAN-Marg was developed using a principal component analysis to derive four factors from a combination of 18 census variables.Note 38

Residuals were assessed for spatial autocorrelation using Moran’s I statistic and, when significant, a simultaneous autoregressive (SAR) model was fit.Note 40 Spatial weights for the SAR were developed from a Queen’s contiguity matrix, and Lagrange multiplier statistics were used to determine whether a spatial lag or spatial error model would provide a better model fit.Note 40 As a secondary analysis, the mean K10 distress scores were determined by quintile of the pollutant.


Characteristics of the three samples are provided in Table 1. In general, characteristics of the three samples were similar, although the Ontario sample included a greater proportion of older respondents and fewer current daily smokers. The Alberta and British Columbia sample included fewer respondents who were unemployed. The percentage of Quebec respondents with a K10 distress score of 11 or more was notably greater (15.1%) than in the other two samples (10.1% for Alberta and British Columbia, and 10.5% for Ontario). Because they were not intended for creating population estimates, these samples are unweighted. Therefore, the results are not representative of regional estimates. Characteristics of the CAN-Marg and the three air pollutants are provided for the three datasets in Table 2. Estimates of PM2.5 were greater in Quebec than in Ontario.

Table 3 provides the results of a multiple regression analysis between K10 distress scores (z-score) and all individual and neighbourhood covariates in the final model. In all samples, age was negatively associated with distress, while other variables considered (female sex, not married or common law, lowest income quintile, unemployed, high alcohol use and current daily smoking) were positively associated with distress. Associations between distress and neighbourhood marginalization were less consistent across datasets. For example, distress was positively associated with neighbourhood-level residential instability in the three datasets, and with neighbourhood-level material deprivation in Ontario and Alberta and British Columbia, but not in Quebec. Distress was negatively associated with increasing neighbourhood-level dependency in Ontario and increasing neighbourhood-level ethnic concentration in Ontario and Alberta and British Columbia, but not in Quebec.

Table 4 summarizes associations between air pollution and distress, in unadjusted and fully adjusted models. Residuals were spatially autocorrelated in all cases, so an SAR lag model was applied to adjust for spatial autocorrelation. In all cases, the results of the SAR model were attenuated slightly compared with those from the OLS models. In all three regional samples, higher PM2.5 exposures were associated with increasing distress scores. After adjusting for all individual and neighbourhood covariates, the positive association between PM2.5 and distress remained only for Quebec. NO2 was positively associated with distress scores in Ontario (in unadjusted models) and Quebec (in both unadjusted and adjusted models), but not in Alberta and British Columbia. In Alberta and British Columbia, NO2 was negatively associated with distress scores, but only in the fully adjusted model. Associations between O3 and distress were less consistent. Distress was positively associated with O3 in both unadjusted and fully adjusted models in Quebec, but negatively associated with O3 in Ontario (unadjusted model).

The mean K10 score per quintile of pollutant is provided in Figure 1. An increasing distress score across pollutant quintiles was observed for PM2.5 in all three regional samples, and for NO2 in Ontario and Quebec, consistent with the results in Table 4.


This study identified a small, but positive spatial association between psychological distress and some measures of air pollution (PM2.5 and NO2) in large regional datasets in Canada. Associations between psychological distress and O3 were less consistent, with a positive association only in Quebec. In Quebec, these associations were observed after adjusting for individual SEP, health behaviours, neighbourhood marginalization and spatial autocorrelation. Because PM2.5 levels have declined over the period examined, the differences observed across regions may be at least partly due to older survey and PM2.5 data used for Quebec compared with Ontario.Note 30 A larger range of PM2.5 exposure in the Quebec data may have contributed to stronger associations between air pollution and distress. Further, K10 scores were higher overall in Quebec than in the other two areas considered, which may have contributed to the stronger associations observed.

The association between distress and air pollution was consistent with a similar analysis conducted in the United States, which also identified a positive association between distress and PM2.5 air pollution.Note 26 However, studies in Europe have found mixed results with the association between different air pollutants and mental health disorders and symptoms of distress.Note 27 While stress axis activation has been proposed as a plausible biological pathway that links air pollution exposure to deleterious effects in the brain,Note 18Note 19 it is clear that the development of mental illness and distress in humans is influenced by a wide array of genetic and environmental factors.

In all regional datasets, indicators of social and material deprivation were positively associated with higher distress scores, and the associations between air pollution and distress were relatively weak. Individual SEP and behaviour are likely to have a larger influence on distress relative to air pollution, as more proximal contributors to distress (e.g., financial crisis, job insecurity) may also be more prevalent in these groups. In large Canadian cities, particulate matter and NO2 are known to be spatially correlated with lower SEP, including both material deprivation (e.g., low income, low dwelling value) and social deprivation (e.g., unemployment, lower educational attainment), both at the individual and neighbourhood scale.Note 41Note 42Note 43 Additionally, individuals with lower SEP may be more likely to have poorer health and may be more susceptible to the effects of air pollution.Note 41Note 42Note 43 Both the larger influence of individual and neighbourhood SEP on distress and the correlation of air pollution with individual and neighbourhood SEP (i.e., the double burden of deprivation and exposure) may explain the attenuated associations between pollutants and distress in Ontario (PM2.5 and NO2) and in Alberta and British Columbia (PM2.5), after including several SEP and behavioural covariates in the models.

Previous epidemiological studies have shown that psychological stress may contribute to vulnerability to air pollutants.Note 44Note 45Note 46Note 47 Further to the possibility that exposure to air pollution contributes to psychological distress, spatial associations between air pollutants and distress may help to identify potentially vulnerable populations. The physiological burden of responding to chronic exposure to multiple stressors, conceptualized as allostatic load, can increase the risk of disease.Note 48 A recent study showed that measures of SEP, a surrogate for chronic exposure to stressors, were negatively associated with allostatic load scores in Canada.Note 49 Whether distress predisposes individuals to the adverse health effects of air pollutants remains to be determined.


Owing to data and methodological considerations, the study was not national in scope and therefore is not nationally representative. The study samples included respondents in the four most populous provinces, and respondents in Canada’s seven largest cities, which may affect the generalizability of these results to more rural populations. A similar analysis of a province or territory with more rural dwellings may not yield the same results because of lower spatial variability in stress, distress and air pollution. Additionally, individuals in large urban centres may have better access to health care and social capital than those in more remote areas. However, despite not being national in scope, the overall sample size (n=84,800 persons) was still larger than that of other comparable studies.Note 26Note 27

Since this study was cross-sectional, its design could not account for changes in exposure over time, including from residential mobility. The study was also unable to assess long-term exposures, including those that occurred during critical years of development. A longitudinal study in London, England, found that PM2.5 and NO2 exposure among children at age 12 was associated with increased odds of depressive disorder at age 18, which suggests that there may be a time lag between exposure to air pollution and development of symptoms, or that exposure during critical periods of childhood may be more important in the development of later mental health disorders.Note 48Note 50

Air pollution estimates were attached to CCHS respondents using geographic coordinates from PCCF+ postal code records. Although these coordinates are relatively accurate within larger cities, in rural areas the point estimates are typically positioned 5.2 km (interquartile range = 2.9 km to 8.7 km) away from the home’s actual location.Note 37 However, in this study, rural respondents represented only 19% to 27% of the sample. Self-reported postal codes do not always represent the respondent’s address; sometimes they represent a business address or a post office box. Despite inaccuracies in residential geolocation in rural areas, air pollutant estimates tend to be more homogeneous across the rural landscape, which may help to mitigate some of this geographic inaccuracy.


Three samples of respondents from the CCHS were spatially attached to PM2.5, NO2 and O3 air pollution data to assess associations with a validated measure of psychological distress. Distress was positively associated with exposure to PM2.5 in all three areas. Distress was also positively associated with NO2 in Ontario and Quebec, and with O3 in Quebec. After adjusting for individual and neighbourhood-level socioeconomic characteristics and behaviours, associations between distress and air pollution were observed only in Quebec. These findings suggest a possible epidemiological link between air pollution and distress, in line with toxicological evidence of stress axis activation. Future studies are warranted on how air pollution may lead to a distressed state and how distress might lead to greater susceptibility between air pollution and other health outcomes.

Funding source

This work was supported by Health Canada (Clean Air Regulatory Agenda, project 810577).

Date modified: