Abstract
BACKGROUND: Cohort studies have reported positive associations between long-term exposure to ambient fine particles (PM(2.5)) and mortality. However, there is heterogeneity across results that may be due to unmeasured and residual confounding. We aim to compare two distinct types of analysis to examine whether long-term exposure to air pollution is associated with all-cause and ischemic heart disease (IHD) mortality: (i) a traditional survival analysis that contrasts different individuals at similar times, and (ii) a self-controlled design that controls for time-invariant confounders by contrasting the same person at different times. METHODS: We used an open population-based cohort created from health administrative databases. The cohort included all adults older than 20 years living in the province of Quebec, Canada, between 2002 and 2017. We assessed long-term exposure to ambient PM(2.5) using annual mean concentrations estimated from satellite-based model. We assigned time-varying annual exposures to individuals based on their residential postal code. For both types of analyses, we estimated the association with all-cause and IHD mortality using Cox proportional hazard models adjusted for age, time-varying neighborhood socioeconomic status, and current year. RESULTS: The cohort included 7,506,027 individuals, with 996,665 and 231,376 deaths from all causes and IHD, respectively. In the between-subjects analysis, hazard ratios from linear models were 1.03 (95% confidence interval [CI]: 1.02, 1.03) for all-cause and 1.04 (95% CI: 1.03, 1.05) for IHD per interquartile range (3.3 µg/m(3)). In the within-subjects analysis, hazard ratios were 1.06 (95% CI: 1.04, 1.08) for all-cause and 1.06 (95% CI: 1.00, 1.11) for IHD per interquartile range increase in the difference between event and referent years (1.9 µg/m(3)). However, we found evidence of nonlinearity, with a steeper slope at lower concentrations. CONCLUSIONS: Consistently across the two designs, we found positive associations between annual mean exposure to low level of ambient PM(2.5) and mortality. However, the magnitude of the association varied depending on the statistical design.