Effects of Ambient Fine and Coarse Particles on Mortality in Phoenix, Arizona

Considerable attention has been given to the health effects of ambient air borne particulate matter as the Environmental Protection Agency (EPA) revises the National Ambient Air Quality Standards (NAAQS). Much of the recent focus has been on the effects of fine particles, with the establishment of additional monitoring platforms to measure both fine and coarse particles for epidemiological studies. Much of the evidence supporting the 1997 standards is based on statistical models using generalized additive models for time series data. Among the statistical concerns raised by National Research Council is the issue of whether observed statistical associations are a result of multiple testing and selection effects due to model selection. We propose a method based on Bayesian Model Averaging to estimate relative risks that incorporates both uncertainty due to estimation but also uncertainty due to model choice. This incorporates uncertainty regarding the choice of confounding variables, choice of pollutant (fine and coarse particles), and which lags of all variables should be included in the model. This approach is illustrated using fine and coarse particulate matter data from the National Exposure Research Laboratory research monitoring platform in Phoenix, AZ for the time period of May 1995 to June 1998. We consider elderly non-accidental mortality for three regions of increasing size in the Phoenix metropolitan area, as well as using accidental mortality for all ages as a control population. We find a weak, but suggestive, particulate matter effect on elderly mortality only for the geographic region where fine particles are expected to be spatially homogeneous; posterior probability of a PM effect is 0.91 and 90% probability interval for the relative risk (RR) with a 1 IQR increase in both fine and coarse particulate matter levels is (1 < RR A 1.04). While previous scientific information would suggest that fine particles should have a larger effect than coarse particles on health outcomes, we find instead that the effect of coarse particles on mortality is stronger, with very little support for models that include fine particles, but not coarse particles.