Using Monte Carlo Methods to Assess the Impact of Uncertainties in Exposure on the Analysis of Dose-Response in Epidemiologic Studies. L. T. Stayner, S. J. Gilbert, and R. J. Smith, National Institute of Occupational Safety and Health, Cincinnati, OH; A. J. Bailer, National Institute of Occupational Safety and Health, Cincinnati, OH and Miami University, Oxford, OH; and E. Garshick, Brockton/West Roxbury VAMC and Harvard Medical School, Boston, MA
Estimates of exposure are generally treated as if they are known without error in dose-response analyses of occupational cohort mortality studies. In fact, it is generally well recognized that misclassification of exposure is a potentially large problem in most occupational cohort mortality studies and may introduce substantial uncertainty into exposure-response analysis. Our objective was to present a method for ascertaining the magnitude of this uncertainty and to illustrate this method using data from a previously conducted cohort mortality study of workers exposed to diesel exhaust. The data were taken from a study of 55,000 railroad workers exposed to diesel exhaust from the 1950s through 1976.1 There are at least the following 3 major sources of uncertainty in the estimates of exposure for this study: 1) the date of first exposure, 2) measurement errors for the exposure intensity, and 3) the impact of historical changes in diesel engines over the time period of the study. One thousand Monte Carlo simulations were run based on distributional assumptions for each of these sources of error, and Poisson regression models for lung cancer were fitted to each of these simulations. Models were run using categorical and continuous measures of exposure and different parametric model forms. The relative risk of lung cancer was found to be generally elevated in all categories of exposure, but did not increase monotonically with exposure. Estimates of exposure-response were found to vary appreciably in the simulations performed, suggesting substantial uncertainty in the analysis related to potential exposure misclassification. This work illustrates the usefulness of Monte Carlo methods for assessing the uncertainty in exposure-response analyses related to exposure misclassification in epidemiologic studies. It also demonstrates that this uncertainty may have had a large impact on the exposure-response observed in the diesel exhaust-exposure study discussed here.
Go to . . .