Using GIS to Estimate Environmental Levels of Target Contaminants in an Epidemiologic Study
Exposure is a function of the concentration of target contaminant in the environment of the study population. The optimal method for quantifying levels of the target agent is the measurement of the environmental media associated with each potential route of exposure during the critical time period for exposure. However, rarely is there an opportunity to make such measurements. Alternatively, predicted environmental levels of the target agent can be estimated using source-receptor modeling. Computer-based models designed to predict levels of contaminants due to point sources (smokestack) or nonpoint sources (drift from aerial spraying of pesticides) are available. Often these predictions are used as a surrogate for exposure in the epidemiologic analysis. In either case, validation of the estimates is important to understand the results of the epidemiologic study. Validation is often overlooked in the exposure assessment process. It is also important that the environment depicted in the modeling period be representative of the environment during the exposure period necessary for the epidemiologic study. Generally, the degree to which validation can be accomplished is a function of measurement data available for the time period of interest. Most source-receptor models require some measurements for constructing (calibrating) the predictive algorithms.

Example: The Lung Cancer in Stockholm Study (Bellander et al. 2001; Nyberg et al. 2000)
A population-based case–control study, the Lung Cancer in Stockholm Study (LUCAS), was designed to investigate whether urban air pollution increases lung cancer risk. Previous studies had commonly used crude surrogates for individual exposure, limiting the power of detecting any risk associated with air pollution. The LUCAS study used advanced modeling techniques to assess individual exposure for relevant time periods (several decades before diagnosis). Detailed emission data, dispersion models, and GIS were used to assess historical exposure to several components of ambient air pollution. The study base consisted of all men 40–75 years of age who lived in Stockholm County at any time between 1985 and 1990 and who had lived in the county since 1950, with a maximum of 5 years of residence outside the county. A total of 1,042 lung cancer cases diagnosed between 1985 and 1990 were included, as well as 2,364 controls. Information on residence from 1955 to the end of follow-up for each individual, 1990–1995, was collected using a questionnaire. Nitrogen oxides (NOx and NO2) and sulfur dioxide (SO2) were chosen as indicators of air pollution from road traffic and residential heating, respectively.
Ambient air concentrations of NOx , NO2, and SO2 were assessed throughout the study area for three points in time (1960, 1970, and 1980) using reconstructed emission data for these index pollutants together with dispersion modeling (Figure 5). The modeled NO2 estimates for 1980 were validated with available measurement data. Linear intra- and extrapolation were used to obtain annual estimates for the remainder of the exposure period (1955–1990). Individual addresses were geocoded with an estimated error of < 100 m for 90% of the addresses. Annual air pollution estimates were then linked to residence coordinates, yielding cumulative residential exposure indices for each individual. There was a wide range of individual long-term average exposure, with an 11-fold interindividual difference in NO2 and an 18-fold difference in SO2
The detailed individual exposure assessment made it possible to assess relative risk potentially associated with road traffic. Average traffic-related NO2 exposure over 30 years was associated with a relative risk of 1.4 and a 95% confidence interval 1.0, 2.0 for the top decile of exposure, adjusted for tobacco smoking, socioeconomic status (SES), residential radon, and occupational exposures, and taking into consideration a latency period of 20 years (Nyberg et al. 2000). The signifi-cance of these results was recognized in an accompanying editorial as being the first study that had used this advanced exposure assessment, making the detailed analysis possible (Rothman and Cann 2000).
The results indicate that GIS can be useful for exposure assessment in environmental epidemiology studies, provided that detailed geographically related exposure data are available for relevant time periods.