Detection of temporal trends in the short-term health effects of air pollution

This paper proposes a new method to detect trends in the association between a single pollutant and single health outcome. We model air pollution as a time-varying second-order random walk, which gives 2 ways to characterize trend—momentum for direction and stability for variability. We fit this model to circulatory and pulmonary mortality data from 6 Canadian cities across 2 age groups (1–65 vs. 66+) over 17 years (1996–2012) and find that 2 outcomes have trends—St. John’s has a decreasing trend for circulatory mortality, and Toronto has an increasing trend for pulmonary mortality, both for ages 66+. We make 3 additional observations. First, although the posterior means of the air pollution effect are smooth and are relatively stable over time, their estimates often have large uncertainty. Second, pulmonary mortality has more time-varying associations than circulatory mortality. This may be related to seasonality in pulmonary mortality, due to factors such as the flu. Third, momentum and stability cannot indicate changes in trend for non-linear effects, an extension that warrants further research.