An Assessment of Air Pollution Exposure Information for Health Studies

Most studies of air pollution health effects are based on outdoor ambient exposures, mainly because of the availability of population-based data and the need to support emission control programs. However, there is also a large body of literature on indoor air quality that is more relevant to personal exposures. This assessment attempts to merge these two aspects of pollution-related health effects, emphasizing fine particles. However, the basic concepts are applicable to any pollutant. The objectives are to examine sensitivities of epidemiological studies to the inclusion of personal exposure information and to assess the resulting data requirements. Indoor air pollution results from penetration of polluted outdoor air and from various indoor sources, among which environmental tobacco smoke (ETS) is probably the most toxic and pervasive. Adequate data exist on infiltration of outdoor air but less so for indoor sources and effects, all of which have been based on surveys of small samples of individual buildings. Since epidemiology is based on populations, these data must be aggregated using probabilistic methods. Estimates of spatial variation and precision of ambient air quality are also needed. Hypothetical personal exposures in this paper are based on ranges in outdoor air quality, variable infiltration rates, and ranges of indoor source strength. These uncertainties are examined with respect to two types of mortality studies: time series analysis of daily deaths in a given location, and cross-sectional analysis of annual mortality rates among locations. Regressions of simulated mortality on personal exposures, as affected by all of these uncertainties, are used to examine effects on dose-response functions using quasi-Monte Carlo methods. The working hypothesis is that indoor sources are reasonably steady over time and thus applicable only to long-term cross-sectional studies. Uncertainties in exposure attenuate the simulated mortality regression coefficients; correlations between “true” and hypothesized exposures are used to compare their effects. For a given exposure uncertainty level, attenuation of regression coefficients is similar for both types of simulated mortality studies, but since cross-sectional studies involve indoor sources they are more sensitive, to the point where regression coefficients may be driven to zero. The most pressing need for confirming data is the distribution of indoor sources among cities, especially for ETS