In the short-term, exposure to air pollution has been shown to cause and aggravate a number of respiratory conditions, including bronchitis, asthma and chronic obstructive pulmonary disease (COPD). There have been a number of studies which have tried to quantify the impact of air pollution on our health. Government legislation, such as the UK Clean Air Act in 1993 and the UK Air Quality Strategy in 2007, which limit pollution levels, are informed by the results of such studies. Typically, the effects of air pollution on human health are estimated by regressing daily measurements of population health against air pollution concentrations and measures of meteorology, that relate to an extended urban area such as a city. Numerous pollutants, including CO, NO2, O3, PM2.5, PM10 and SO2, are measured daily by a network of fixed site monitors. For simplicity, the majority of studies only assess the health risks from a single pollutant, however, the air we breathe and hence are exposed to is made up of many pollutants including, but not limited to, those listed above. The standard modeling approach in a single pollutant analysis, for estimating a representative level of that pollutant across the study region, is to average across the monitors. However, this will give rise to spatially mis-aligned data as the health data relate to the entire region of interest whereas the pollution data are specific to the location of the monitors. The locations of the monitors are likely to have been chosen by preferential sampling, which favors sites with high pollution concentrations and leads to an overestimation of the average concentration over the study region. Further to this, monitors are categorized into different local environments, such as roadside and background sites, therefore averaging over both types of site is also likely to lead to an overestimation of the true study region average. This approach also fails to take into account the population density for the study region and gives equal weighting to the measurements from each monitor regardless of the local density. Finally, the desired pollution measure for a region is an unknown quantity and hence the uncertainty in the estimation of it should be allowed for when estimating its health effects. To address these factors we propose a three stage approach for estimating a representative value of overall air quality and subsequently its effects on human health. We apply our approach to data from London, using counts of respiratory related hospital admissions and a measure of overall air pollution, the latter of which was estimated from the six pollutants previously listed.
Keywords: Air pollution; Bayesian Geostatistical Models; Air Quality Indices
Biography: Helen Powell is currently in the third year of her PhD at the University of Glasgow where she is supervised by Dr. Duncan Lee. Her work has been aimed at improving the modeling approaches currently employed in the study of the short-term effects of air pollution on human health.