Measurement Error Estimation and Correction Methods to Minimize Exposure Misclassification in Epidemiological Studies: Project Summary

Exposure to DBPs has been linked to adverse health outcomes including carcinogenic effects and adverse reproductive and developmental outcomes. Many epidemiologic studies of drinking water contaminants use surrogate ambient measures to estimate underlying town-level mean concentrations. These metrics are often used to estimate individual-level exposures for study participants. Using town-level ambient metrics to estimate individual-level exposures is a common source of measurement error in epidemiologic studies of DBPs because routinely collected monitoring data may not adequately characterize temporal and spatial variability in DBP concentrations experienced in many water systems or inter-individual differences in water use. The use of these surrogate measures can lead to two types of measurement error (Berkson error and classical error). Measurement error can result in biased effect estimates due to misclassification of exposures and outcomes in epidemiologic studies. This can also lead to increased variability of standard errors of effect estimates, distortion of exposure-response relationships and reduced statistical power to detect associations that may be present.

This report summarizes some of the recent progress in this area that has been conducted by the National Center for Environmental Assessment (NCEA). The primary objective of this research was to examine the utility of routinely collected monitoring data to estimate individual-level exposures in epidemiological studies. Another specific aim was to estimate the potential for exposure misclassification bias due to a variety of sources including (1) unmeasured spatial variability in town average surrogate measures and (2) lack of integration of individual water use practices data including exposure-modifying factors. In addition to quantifying potential exposure misclassification bias, we developed an approach to correct for measurement error that may cause this bias. This research provides valuable information to the risk assessment community in their efforts to quantify the potential impact of using indirect exposure assessment metrics in epidemiologic studies.

This is the final report. Final manuscript will be published in the future.

Citations:
Bateson T.F. and J.M. Wright.  2007.  Regression calibration to ameliorate classical measurement error bias in disinfection by-product studies. Am. J. Epidemiol. 165(Suppl):S37.

Wright J.M. and T.F. Bateson.  2005.  A sensitivity analysis of bias in relative risk estimates due to disinfection by-product exposure misclassification. J. Expos. Anal. Environ. Epidem. 15:212–216.

Wright J.M., P.A. Murphy, M.J. Nieuwenhuijsen and D.A. Savitz.  2006.  The impact of water consumption, point-of-use filtration and exposure categorization on exposure misclassification of ingested drinking water contaminants.  Sci. Total Environ. 366:65–73.

U.S. EPA. Measurement Error Estimation and Correction Methods to Minimize Exposure Misclassification in Epidemiological Studies: Project Summary. U.S. Environmental Protection Agency, Washington, DC, 2008.