You are here

Comparing the estimated health impacts of long-term exposure to traffic-related air pollution using fixed-site, mobile, and deep learning models

Principal Investigator: 

McGill University, Canada

This study will evaluate health impacts of long-term exposures to traffic-related air pollution using exposure estimates from fixed-site and mobile measurement campaigns, as well as deep learning models, in Toronto and Montreal, Canada. The investigators will compare exposure estimates generated by these models to present-day and historical measurements, and to each other. 

Funded under
In review

Poster abstract for HEI Annual Conference 2023

Long-term Exposure to Outdoor Ultrafine Particles and Nonaccidental Mortality in a Large Population-Based Cohort: Persistent Associations Across Different Approaches to Exposure Assessment

Marshall Lloyd1, Olaniyan Toyid2, Arman Ganji3, Junshi Xu, Alessya Ventura1, Leora Simon1, Mingqian Zhang3, Milad Saeedi3, Shoma Yamanouchi3, An Wang3, Alexandra Schmidt1, Kris Hong4, Hong Chen5, Paul Villeneuve6, Joshua Apte7, Eric Lavigne8, Rick Burnett9, Michael Tjepkema2, Marianne Hatzopoulou3, Scott Weichenthal1

1McGill University, Montreal, Canada; 2Statistics Canada, Ottawa, Canada; 3University of Toronto, Toronto, Canada; 4Consultant, Manchester, United Kingdom; 5Health Canada, Ottawa Canada; 6Carleton University, Ottawa, Canada; 7University of California Berkeley, Berkeley, California, USA; 8University of Ottawa, Ottawa, Canada; 9Consultant, Ottawa, Canada

BACKGROUND: Outdoor ultrafine particles (UFP; particulate matter smaller than 100nm) have been associated with adverse health outcomes including cardiovascular disease and brain tumour incidence. Assessing long-term UFP exposures is a challenge and health effect estimates may be sensitive to exposure assessment approach. We applied estimates from several novel UFP exposure models to a population-based cohort of 1.9 million people and investigated associations with nonaccidental mortality.

METHODS: This analysis followed adults from the Canadian Census Health and Environment Cohort in Montreal and Toronto over a 15-year period (2001-2016) pooled across multiple census years (1991, 1996, 2001, 2006). Cox proportional hazard ratios (HR) were estimated for nonaccidental mortality (ICD-10: A through R) and long-term exposure to outdoor UFP concentrations at residential address. UFP exposures were estimated using deep learning models, generalized additive models (GAM), and models combining deep learning and GAM approaches (i.e., combined models). Exposure estimates were projected into the past (i.e., backcasted) using historical traffic data to account for long-term trends in outdoor concentrations. Exposures were also weighted using travel demand surveys to account for population mobility. HRs were estimated using UFP estimates from each exposure modelling approach both with and without mobility weighting as well as backcasting in order to investigate how choice of exposure assessment approach may impact estimated HRs. Models controlled for relevant sociodemographic factors and outdoor concentrations of fine particulate matter (PM2.5), black carbon, and oxidant gases (NO2 and O3) as well as mean UFP size.

RESULTS: For each exposure assessment approach, an interquartile increase in UFP concentrations (~5000 pt/cm3) was associated with nonaccidental mortality. The UFP combined model resulted in an HR of 1.034 (95% CI 1.027-1.042). Backcasting resulted in an HR of 1.030 (95% CI 1.023-1.037) and for mobility-weighting it was 1.023 (95% CI 1.015-1.031). Backcasting and mobility-weighting together led to an HR of 1.013 (95% CI 1.006-1.016). Using deep learning models and GAMs on their own resulted in HRs of 1.046 (95% CI 1.039-1.053) and 1.010 (95% CI 1.005-1.015), respectively. Applying mobility weights and backcasting to the deep learning and GAM exposure estimates gave similar results.

CONCLUSION: Long-term exposure to outdoor UFPs was associated with nonaccidental mortality regardless of exposure modelling approach (i.e., deep learning, GAM, or combined models) and both with and without adjustments for population mobility or historical traffic patterns. This persistent association suggests that long-term exposure to elevated outdoor UFP concentrations is harmful to human health.