highly developed, standardized coding systems. Air-pollution data have been linked to mortality data on a local basis in numerous cross-sectional studies (Boyd, 1960; Buck and Wicken, 1967; Glasser and Greenburg, 1971; Lave and Seskin, 1973). Time-series analyses may require control for other variables, such as seasonality and autocorrelation (Ozkaynak et al., 1986; Mazumdar et al., 1982; Schwartz and Marcus, 1990). These approaches are ecologic; that is, a measure of the distribution of pollutant concentrations in an area is correlated with a measure of the distribution of health status in the area such as death rates.

An alternative is to study health measures in cohorts of known exposure status, such as mortality in occupational cohorts (Fraser et al., 1982; Wingren et al., 1991). The subjects can be persons at risk of exposure to toxic materials because they live near toxic-waste sites or for other reasons, e.g., inclusion in the National Exposure Registry of ATSDR.

Other assessments of the health of a population are based on hospital admissions, emergency-room visits, and calls for ambulances. Pope (1991) used regression methods to study the association between hospital admissions for respiratory conditions and measurements of PM10 in local areas, including control for temperatures based on month of admission. Although characteristics of individual subjects were not available, Pope noted strong associations between indicators of respiratory health, particulate pollution, and the operation of a nearby steel mill.

Studies with Information on Health Status and Nonconcentration Measures of Environmental Status:

The studies discussed above used ambient-concentration data from monitoring stations mainly as a surrogate for the probability that personal exposures were sufficient to affect health, but models can use other sources of environmental data. For instance, Frank et al. (1986) used the NHIS to evaluate the relation between chronic cardiovascular illness and exposure to carbon monoxide in the workplace. Information on occupational exposure from the National Occupational Hazards Survey was used to estimate the probability that people in specified jobs were exposed to carbon monoxide. Thus, the exposure data did not include direct measures. Information for individual subjects was linked to exposure probabilities on the basis of current occupation and humidity, based on county of residence. Additional risk factors evaluated in multivariate analyses included age, obesity, sex, demographic variables, and smoking status. Because information on health status is predicated on a subject's reporting that a medical professional had diagnosed a condition, Frank et al. incorporated economic variables—such as availability of health care and ability to afford health care—into the analyses.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement