You are here:
COMBINING EVIDENCE ON AIR POLLUTION AND DAILY MORTALITY FROM 20 LARGEST U.S. CITIES: A HIERARCHICAL MODELING STRATEGY
Citation:
Cox, L H. COMBINING EVIDENCE ON AIR POLLUTION AND DAILY MORTALITY FROM 20 LARGEST U.S. CITIES: A HIERARCHICAL MODELING STRATEGY. JOURNAL OF THE ROYAL STATISTICAL SOCIETY 163(3):294-295, (2000).
Description:
Environmental science and management are fed by individual studies of pollution effects, often focused on single locations. Data are encountered data, typically from multiple sources and on different time and spatial scales. Statistical issues including publication bias and multiple comparisons are often present but unaddressed. Policy makers must pool individual studies to infer and understand pollution-health effects relationships and trends and act upon them. Inference and pooling cannot rely on traditional design-based methods or meta-analytic techniques, and model-based, hierarchical, Bayesian methods are needed. The authors have contributed to environmental epidemiology and environmetrics by providing a general hierarchal methodology for pooling estimates of pollution health effects, demonstrated for the important case of particulate mortality effects.
Contributions of this work are to enable scientific investigation on a regional or national scale, and to borrow strength between multiple studies, some of which may be too small (in observations) or too coarse (in frequency) to provide reliable estimates at all locations. Potential advantages are to enable critical examination of the problem at single locations, e.g., to assess bias or to explore additional covariates. The latter is particularly important for particulate matter health effects, as questions remain regarding the relative mortality effects of size, chemical composition, shape and number of particles. Effects of multiple pollutants are also important, including fine (2.5 microns or less) and coarse (2.5-10 microns) particles.
Some observations: The examination of bias introduced by measurement error is important and was raised in a National Research Council report on setting research priorities for particulate matter studies (National Academy of Sciences 1998). The authors selected a two-stage, not fully Bayesian, model. Pooled meteorological adjustments might capture otherwise ignored regional effects and improve estimates of local relative mortality effects. Some regional effects are captured by the spatial analysis, based on distance. Direction may also be influential; geometric anisotropy (Cressie, 1994, p. 64) could be incorporated into the spatial model. A fully Bayesian approach enables assessing uncertainty due to model selection (Clyde 1998). Despite evidence of negligible difference between the two approaches for these data, a general methodology for national surveillance of effects of air pollution and weather on public health would likely benefit from a fully Bayesian approach. The authors did not set out to compare U.S. national air quality standards for particulate matter with epidemiological evidence of particulate matter health effects, but observe that estimated relative effects obtain above and below the regulatory standard. A more meaningful regulatory comparison would be based on maximum daily concentration.
Views expressed are solely those of the author and should not be interpreted as representing the policies or practices of the U.S. Environmental Protection Agency.