Short-term health impacts of air pollution in Canada: a nonlinear multi-pollutant Bayesian case-crossover framework

Abstract

Abstract Background Most short-term air pollution studies rely on linear, single-pollutant models that cannot adequately describe the nonlinear and interdependent nature of pollutant-health associations. Methods We developed a nonlinear multi-pollutant Bayesian case-crossover model to jointly evaluate short-term health effects of PM 2.5, NO2, and O3. The model accommodates overdispersion, propagates exposure uncertainty from imputed monitoring data, and pools Census Division (CD)-specific exposure-response functions through a Bayesian hierarchical framework. We applied this approach to cause-specific daily mortality (2001–2015) and hospitalization (2001–2018) data across 20 Canadian CDs, covering more than half of the national population. Results The estimated exposure-response relationships varied by pollutant and health outcome. The combined PM 2.5+NO2 metric showed the clearest positive associations with all non-accidental mortality and hospitalization. Respiratory hospitalization estimates were higher among seniors, but evidence of an age difference remained uncertain. An approximate 10-unit contrast in combined PM 2.5+NO2 corresponded to an estimated 1.5% (95% credible interval [CrI]: 0.3, 2.2) increase in all non-accidental mortality and a 0.6% (95% CrI: 0.0, 1.3) increase in all non-accidental hospitalization, while the corresponding estimate for senior respiratory hospitalization was 2.7% (95% CrI: 0.9, 4.5). Several PM 2.5 and NO2 curves were nonlinear, whereas O3 associations were closer to linear over much of the observed range. Conclusion Our study introduces a flexible epidemiologic modelling framework for quantifying the short-term health impacts of correlated pollutants. The results highlight the value of accounting for nonlinear and joint exposure patterns in multi-pollutant settings, while uncertainty remains substantial for several cause- and age-specific estimates.