Statistical analysis of multisite time series data for estimating health effects of environmental exposures

Multisite time series studies provide a rich data source for estimating health effects of widespread environmental exposures, such as heat waves and air pollution. In this work we develop statistical methodology motivated by three research priorities in the field of environmental epidemiology. First, we propose a framework to methodically assess future health impacts under global climate change. We apply this framework to a case study to predict future mortality attributable to heat waves in Chicago under a range of climate models and climate change scenarios. Second, we investigate the role of model uncertainty underlying heat wave risk estimates and develop a Bayesian model averaging approach to systematically account for the model uncertainty. This approach is applied to estimate the relative risk of mortality associated with heat waves in 105 US cities during the period 1987--2005. Third, we develop methodology to estimate the health effects of simultaneous exposure to multiple pollutants. While many studies have estimated risks associated with exposure to individual pollutants, this may not realistically capture true exposure to complex mixtures. We propose the reduced Bayesian hierarchical model as a new statistical approach for combining information across locations on the parameter of interest (e.g. excess number of deaths attributed to simultaneous exposure to high levels of many pollutants) when the within-location model has a large number of nuisance parameters. We apply this approach to estimate location-specific and overall relative risks of cardiovascular hospital admissions associated with simultaneous exposure to elevated levels of particulate matter and ozone in 51 US counties during the period 1999--2005.