477Long-term exposure to low PM2.5 and NO2 and all-cause mortality in Australia: a difference-in-differences analyses

Abstract Background Although many cohort studies have documented the long-term effects of ambient air pollution on mortality, but they suffer from residual confounding, being unable to control unmeasured confounders, and are often not population representative. A recently developed variant of difference-in-difference (DID) approach is promising to address these limitations. Methods We collected annual all-cause death data from 2,193 statistical areas level-2 (SA2) in Australia during 2001-2015. Area-level annual mean concentrations of PM2.5 and NO2 were derived from widely used grid (0.01°×0.01° and 0.1°×0.1°, respectively) datasets. The variant of DID method was used to evaluate the causal relationship between annual PM2.5 and NO2 and all-cause mortality. We further developed this method by considering non-linear associations and lag impacts using distributed lag non-linear model. Results The impacts of low PM2.5 (1.94-12.00 µg/m3) and NO2 (0-7.41 µg/m3) on all-cause mortality were non-linear and lasted for 0-3 year and 0-6 year, respectively. The moving average PM2.5 (0-3 year) and NO2 (0-6 year) showed non-significant impacts below the thresholds (4.44 µg/m3 and 1.10 µg/m3) observed, while every 1 µg/m3 increase above the thresholds were associated with 2.4% (95%CI: 1.6-3.3%) and 9.4% (95%CI: 7.9-10.8%) increase in all-cause mortality, respectively. We estimated that 3.0% (95%CI: 2.0-3.9%) and 9.9% (8.3-11.3%) deaths were attributable to PM2.5 and NO2, respectively. Conclusions We further developed the causal DID model and documented the deadly impacts of long-term exposure to low PM2.5 and NO2 with thresholds and lag periods. Key messages Long-term exposure to low PM2.5 and NO2 are still deadly but have thresholds.