Modelling spatial variability in concentrations of single pollutants and composite air quality indicators in health effects studies



The health impact resulting from a few days of elevated air pollution concentrations has been the focus of much recent research, most of which assesses the effects of single pollutants rather than composite air quality indicators. The average concentrations of these pollutants across the study region are typically estimated by averaging the measurements from the available network of monitors, and this simplistic approach has several deficiencies. Firstly, it is unlikely to be the average concentration across the region under study, owing to the likely non-random placement of the monitoring network. Secondly, the true spatial average is an unknown quantity, and hence the uncertainty in any estimate should be allowed for when estimating its health effects. This paper proposes a novel Bayesian hierarchical framework for addressing these problems, which consists of statistical models for estimating spatially representative measures of single pollutants and composite air quality indicators, and the health effects of these pollution measures while correctly allowing for their uncertainty. This methodological development is motivated by an epidemiological study of the effects of air pollution on respiratory mortality in Greater London, England, between 2003 and 2005. The key findings from this study are that traditional approaches are likely to underestimate the uncertainty in the health effects of air pollution compared with the approach proposed here and increased risks of between 1.4% and 3.1% are associated with 1-standard-deviation increases in the concentrations of ozone, particulate matter (PM10) and the composite air quality indicator that is adopted here.