Because air pollution, like many other environmental exposures, is distributed fairly uniformly over large geographic areas, there has been a tendency for epidemiologic research to rely heavily on ecologic inference. The limitations of this approach are well known, but largely unavoidable in this field. Although there is a considerable literature on this topic in sociology and other disciplines, the valid uses of ecologic methods are only recently beginning to be appreciated in epidemiology. In this project, we are developing methods of analysis for assessing exposure effects at the individual level in studies where group effects are also important and must be controlled or (in the case of group exposure variables) exploited.

Detailed personal exposure assessment is seldom possible for all subjects in a large epidemiology study, and is usually confined to substudies. For example, in the USC air pollution study, ambient levels are available for each community, together with questionnaire data on residence history and various exposure modifying factors (time spent outdoors, physical activity, household characteristics) on all subjects; in addition, personal dosimetry and daily activity diaries are being collected for a subsample, together with microenvironmental sampling data. Another aspect of this statistical project is the development of models for predicting personal exposures from the data that will be available on all subjects, and methods of analysis of exposure-response relationships allowing for the uncertainties in these exposure assignments (Navidi W, Lurman F, J Exp Anal Environ Epidemiol, 1995; 5: 111-124).

As part of EPA's review of the standard for particulate air pollution, Dr. Thomas was commissioned to write a report on the relevant statistical issues. This report included a simulation study of the small sample performance of the Generalized Estimating Equations (GEE) methods that have been widely used by investigators in this field, but have recently been criticized on the basis of the small number of years typically treated as independent observations. We are continuing to explore these issues while Dr. Thomas serves as a member of HEI's oversight committee for a major reanalysis of these studies by Drs. Jonathan Samet and Scott Zeger of Johns Hopkins. Dr. Navidi has also been involved on HEI as a study section member to review grants for further epidemiologic research and has been collaborating with Dr. Langholz as an investigator of the statistical foundations of the case-crossover design that is being frequently considered in this context.