You are here

Identifying the shape of the association between long-term exposure to low levels of ambient air pollution and the risk of mortality: An extension of the Canadian Census Health and Environment Cohort using innovative data linkage and exposure methodology

Principal Investigator: 

University of British Columbia, Canada

This study will investigate mortality effects of low levels of air pollution in Canada using Canadian Census data from about 6 million people. The shape of the exposure-response function will be characterized using newly developed flexible non-linear exposure-response functions. See also this Program Summary of HEI's research program on low levels of air pollution. 

Funded under
Abstracts for the 2018 HEI Annual Conference. Please scroll down to view the three abstracts and posters. 

MAPLE: Mortality-Air Pollution Associations in Low Exposure Environments

Michael Brauer1, Jeffrey R. Brook2, Paul Bissonnette3, Richard T. Burnett4, Tanya Christidis5, Daniel L. Crouse6, Anders Erickson1, Perry Hystad7, Chi Li3, Lauren Pinault5, Randall V. Martin3, Michael Tjepkema5, Aaron van Donkelaar3, Crystal Weagle3,Scott Weichenthal8

1 The University of British Columbia, Vancouver, British Columbia, Canada; 2 Environment Canada, Toronto, Ontario, Canada; 3 Dalhousie University, Halifax, Nova Scotia, Canada; 4 Health Canada, Ottawa, Ontario, Canada; 5 Statistics Canada, Ottawa, Ontario, Canada; 6 University of New Brunswick, Fredericton, New Brunswick, Canada; 7 Oregon State University, Corvallis, OR, USA; 8 McGill University, Montreal, Quebec, Canada

Background. Fine particulate matter (PM2.5) has been associated with mortality in many studies across the globe. However, the range in concentration for which this association is present is not known. Since nearly the entire population of Canada lives in areas with ambient concentrations below 12 µg/m3, and studies repeatedly demonstrate associations with mortality in this population, it is an ideal environment to study the mortality impacts of low PM2.5 concentrations.

Objectives. To apply novel satellite-based estimates of exposure to PM2.5 to several large population-based cohorts, and to characterize the shape of the relationship between PM2.5 exposure with cause-specific mortality.

Methods. We developed annual satellite-based PM2.5 exposure estimates at 1 km resolution across Canada for 1998-2016. Estimates were back-casted to 1981 using remote sensing, chemical transport models and historical ground monitoring data. Historical estimates for NO2 and ozone were also developed for the same period. Further refinements will incorporate new information on the relationship between aerosol optical depth (AOD) and PM2.5 based on measurements at five sites across Canada.

Exposure estimates were applied to four large, population-based, cohorts: ~8.5 million subjects who completed the 1991, 1996, and 2001 census long forms, and 389,000 participants of the Canadian Community Health Survey (CCHS) 2001, 2003, 2005, and 2007/2008 panels. All subjects are being linked to annual tax records, to establish residential histories, and mortality until 2016.

We analyzed the 2001 census cohort (CanCHEC 2001) linked to a mortality dataset including all International Classification of Disease (ICD-10) codes listed on the death certificate to evaluate the use of contributing cause of death information in informing cause-specific mortality analyses. In addition, we evaluated the application of indirect adjustment methods using the CCHS for behavioral risk factors not recorded in the census cohorts and compared exposure profiles between the CCHS and the CanCHEC 2001 cohort.

Results and Conclusions. Satellite-based PM2.5 estimates were highly correlated with ground monitors (R2 = 0.82) across North America. Root-mean-squared-error (1.5 µg/m3 for the full dataset) decreased slightly when higher PM2.5 concentrations were excluded. Initial results from filter analysis indicate variation in PM2.5 /AOD relationships across sampling sites, suggesting potential for further reductions in RMSE. Mention of diabetes was identified as an important contributing cause of cardiovascular disease mortality linked to PM2.5 exposure. In the 2001 CanCHEC, co-mention of diabetes increased magnitude of the association compared to CVD mortality without diabetes, suggesting that restricting analyses to the primary cause of death likely underestimates the role of co-morbidities such as diabetes on air pollution-related mortality.


Poster by Brauer et al, 2018 Annual Conference

Exposure Estimation for MAPLE: Mortality-Air Pollution Associations in Low Exposure Environments

Randall V. Martin1 (Presenter), Paul Bissonnette1, Jaqueline Burke1, Robyn Latimer1, Chi Li1, William Russell1, Graydon Snider1, Emily Stone1, Aaron van Donkelaar1, Crystal Weagle1, Perry Hystad2, Jeffrey R. Brook3, Alain Robichaud3, Richard Menard3, Scott Weichenthal4, Richard T. Burnett5, Daniel L. Crouse6, Anders Erickson7, Lauren Pinault8, Michael Tjepkema8, and Michael Brauer7

1 Dalhousie University, Halifax, NS, Canada; 2 Oregon State University, Corvallis, OR, USA; 3 Environment Canada, Toronto, ON, Canada; 4 McGill University, Montreal, QC, Canada; 5 Health Canada, Ottawa, ON, Canada; 6 University of New Brunswick, Fredericton, NB, Canada; 7 University of British Columbia, Vancouver, BC, Canada; 8 Statistics Canada, Ottawa, ON, Canada

Background Uncertainty remains in the association between mortality and long-term exposure to ambient fine particulate matter (PM2.5) at low concentrations. A paucity of air quality monitors in regions with low concentrations inhibits exposure assignment from ground-based monitors alone. Reduced ground-based monitoring for historical time periods also poses challenges for exposure assignment in cohort studies.

Objectives To develop PM2.5 concentration estimates across Canada by combining satellite remote sensing, chemical transport modeling, and ground-based monitoring.

Methods We developed estimates of PM2.5 concentrations at 1km by 1km resolution across North America for each year from 1981 to 2012 (and ongoing out to 2016). The estimates were based on a combination of satellite remote sensing of aerosol optical depth (AOD), relating AOD to PM2.5 using the GEOS-Chem chemical transport model, and integrating of these concentrations with ground-based monitoring data through geographically weighted regression. Estimates prior to 1998 included additional PM10 and total suspended particles (TSP) ground-based monitoring information.

We deployed targeted ground-based monitoring of the relationship between AOD and PM2.5.  This monitoring is being used to evaluate and improve the GEOS-Chem calculation of the relation between AOD and PM2.5.

Annual concentration estimates are also developed for O3 and NO2 using a combination of ground-based monitoring, chemical transport modeling, and for NO2 land use information and satellite remote sensing.

Results Satellite-based PM2.5 estimates were consistent with ground-based monitors (R2 = 0.82, slope=0.97) across North America even when large fractions were withheld for cross-validation (excluding up to 70% of monitors decreased R2 to 0.78). Root-mean-squared-error decreased slightly with decreasing PM2.5 concentrations from 1.5 µg/m3 for the full dataset to 1.3 µg/m3 for concentrations < 8 µg/m3. Historical PM2.5 surfaces benefitted from PM10 and TSP data.

Initial targeted ground-based monitoring identified the mass scattering efficiency (the relationship between PM2.5 and scatter) as a driving factor of the relation between AOD and PM2.5. Development of the representation of aerosol size and aerosol hygroscopicity in the GEOS-Chem model improved the simulation of mass scattering efficiency.

Conclusions A combination of satellite remote sensing, chemical transport modeling, and targeted ground-based monitoring offers valuable information about ambient air quality at low concentrations.


Poster by Martin et al, 2018 Annual Conference

Evaluation of a method to indirectly adjust for unmeasured covariates in large administrative data cohort analyses: An analysis of associations between fine particulate matter and mortality in the 2001 Canadian Census Health and Environment Cohort (2001 CanCHEC)

Anders C Erickson1 (Presenter), Michael Brauer1, Lauren L Pinault2, Daniel L Crouse3, Scott Weichenthal4,5, Randall V Martin6,7, Perry Hystad8, Jeffrey R Brook9, Michael Tjepkema2, Richard T Burnett5

1 The University of British Columbia, Vancouver, British Columbia, Canada; 2 Statistics Canada, Ottawa, Ontario, Canada; 3 University of New Brunswick, Fredericton, New Brunswick, Canada; 4 McGill University, Montreal, Quebec, Canada; 5 Health Canada, Ottawa, Ontario, Canada; 6 Dalhousie University, Halifax, Nova Scotia, Canada; 7 Harvard-Smithsonian Center for Astrophysics, Cambridge, MA, USA; 8 Oregon State University, Corvallis, OR, USA; 9 Environment Canada, Toronto, Ontario, Canada

Background. The method of indirect adjustment for unmeasured confounding variables has become a promising technique in large environmental epidemiological cohort studies. This method has not, however, been formally evaluated or tested for application in non-linear survival models.

Objectives. To describe and evaluate the indirect adjustment method using a large national longitudinal cohort, the 2001 Canadian Census Health and Environment Cohort (CanCHEC, N=2.4 million), and several pooled cycles of the Canadian Community Health Survey (CCHS, N=450,000) as the representative matching dataset with detailed risk factor information (e.g. smoking, body mass index, exercise).

Methods. First, we assessed the distribution of fine particulate matter (PM2.5) among subjects across characteristics (age, sex, etc.) for both the cohort and the health survey to compare for consistency. Next we examined the direction and magnitude of correlations amongst the variables available for both the cohort and health survey. We implemented validation tests to assess the performance of the indirect adjustment method on non-linear Cox proportional hazard models using only the CanCHEC and indirectly adjusting for known variables. To evaluate the application of the CCHS for the indirect adjustment of the CanCHEC, we assessed survival models wherein specific variables available in both datasets were excluded (e.g. education, income), applied indirect adjustment, and then returned to assess the amount of bias correction.

Results. Comparisons of the cohorts at baseline (2001) showed very similar PM2.5 distribution profiles across population characteristics, although PM2.5 levels for CCHS participants tended to be consistently 1.8-2.0 µg/m3 lower than in the CanCHEC cohort. This finding is likely due to sampling differences between urban and rural areas for the census and health survey. Applying a sample-weighting scheme to the CCHS largely corrected for this discrepancy in mean PM2.5 levels. Correlations among variables within the two cohorts were consistent. Results for validation tests are on-going.

Conclusions. A thorough and formal evaluation of the indirect adjustment method for health outcome data using a large longitudinal mortality cohort and representative health survey will help establish protocols that other jurisdictions can use to assess the viability of this method and possibly correct for differences in their own cohort datasets when information on potential confounding variables are not available.


Poster by Erickson et al, 2018 Annual Conference


Pinault LL, Weichenthal S, Crouse DL, Brauer M, Erickson A, Donkelaar AV, Martin RV, Hystad P, Chen H, Finès P, Brook JR, Tjepkema M, Burnett RT. Associations between fine particulate matter and mortality in the 2001 Canadian Census Health and Environment Cohort. Environ Res. 2017 Nov;159:406-415. doi: 10.1016/j.envres.2017.08.037. Epub 2017 Sep 18.

Pinault L, Brauer M, Crouse DL, Weichenthal S, Erickson A, van Donkelaar A, Martin RV, Charbonneau S, Hystad P, Brook JR, Tjepkema M, Christidis T, Ménard R, Robichaud A, Burnett RT. Diabetes status and susceptibility to the effects of PM2.5 exposure on cardiovascular mortality in a national Canadian cohort. Epidemiology. 2018 Aug 1. doi: 10.1097/EDE.0000000000000908.