1. ¹Departments of Biostatistics and Environmental Health, University of Washington, Seattle, Washington 98195-7232
2. ²Department of Epidemiology, University of Washington, Seattle, Washington 98195-7232
3. ³Departments of Environmental Health and Epidemiology, University of Washington, Seattle, Washington 98195-7232

Correspondence: Dr. Lianne Sheppard, PhD, Box 357232, Departments of Biostatistics and Environmental Health, University of Washington, Seattle, WA 98195-7232. Tel.: +1-206-616-2722. Fax: +1-206-616-2724. E-mail: sheppard@biostat.washington.edu

Received 26 April 2000; Accepted 21 August 2000.

Abstract

A limitation of most air pollution health effects studies is that they rely on monitoring data averaged over one or more ambient monitors to represent daily air pollution exposures for individuals. Such data analyses therefore implicitly require the assumption of a homogeneous spatial distribution for particulate matter (PM). This assumption may be suspected in the Pacific Northwest because of its hilly topography and local variations in wood burning. To examine the bias from substituting regional PM (i.e., the average of three ambient monitor measurements) for individual PM exposure, we conducted an exposure substudy to identify the influence of location factors, specifically urban versus suburban classification and topographic features ("upstream" versus "downstream"), on local ambient measurements. Using nephelometer measurements collected over 1 year in four locations, we developed regression models to predict local PM as a function of regional PM, atmospheric stagnation, temperature, and location. We found a significant interaction between atmospheric stagnation and topography, with the most upstream site having reduced PM levels on high stagnation days after controlling for regional PM. We also found a significant interaction with temperature at one downstream site thought to be heavily exposed to wood smoke in the winter. These results are consistent with the physics of surface radiation inversions. The interactions reordered the index versus referent exposures in a case–crossover analysis of out-of-hospital primary cardiac arrest for subjects living in specific locations, but did not meaningfully change the associations with PM from the analysis using regional PM as the exposure. The lack of change in these results may be due to limitations in the data used to correct the exposure estimates or to the absence of a PM effect among persons without prior heart disease who experienced a primary cardiac arrest.