Incorporating individual-level uncertainties in exposure into epidemiologic analyses: An example using arsenic in drinking water and bladder cancer

Epidemiologic analyses traditionally rely on point estimates of exposure for assessing risk. In reality, however, exposure is characterized with uncertainty that can vary across individuals. Methods which incorporate uncertainty in exposure into epidemiologic analyses will yield a greater degree of confidence in estimates of risk. In this report we demonstrate a method using SAS (SAS Institute, Inc., Cary, NC) that uses Monte Carlo simulation to pull estimates of exposure from a normal distribution specified by a mean and uncertainty estimate for each individual. After the exposure estimate is pulled, the relationship between exposure and disease is evaluated using logistic regression. This process is repeated many times and a range of odds ratios and confidence intervals are generated. This method will be demonstrated in a case-control study of bladder cancer using estimates of exposure to arsenic in drinking water over the life-course and its associated uncertainty. The method is applied to time-varying and time-stable individual-level estimates of exposure and uncertainty. The resulting distribution of odds ratios and confidence intervals gives a more realistic estimate of risk from the exposure. Given the ease of implementation, this approach can be readily adopted by epidemiologists who possess quantitative measures of exposure and its associated uncertainty.