The 131st Annual Meeting (November 15-19, 2003) of APHA

Abstract #65561

Anne Jurek, MS, George Maldonado, PhD, and Timothy R. Church, PhD. Department of Environmental Health, School of Public Health, University of Minnesota, 420 Delaware St SE, MMC 807, Minneapolis, MN 55455, 612-623-3394, jure0007@umn.edu

Many investigators believe that nondifferential exposure misclassification inevitably leads to reduction in the strength of an exposure-disease association estimate. However, Sorahan and Gilthorpe (Occup Environ Med, 51, 1994, 839-40) and Thomas (Am J Epidemiol, 142, 1995, 782-3) showed that nondifferential exposure misclassification can spuriously increase the strength of a relative-risk estimate. In response to Thomas (1995), Weinberg et al. (Am J Epidemiol, 142, 1995, 784) suggested additional interesting results to be investigated. Here we present these additional results. In addition, we simulated a wider range of situations than did Sorahan and Gilthorpe (1994) or Thomas (1995). Using a binary exposure variable and various values for the true incidence proportion ratio (IPR_T) and outcome occurrence rates, we generated from paired replicate data sets two estimates, one with exposure misclassified (IPR_M) and one with exposure correctly classified (IPR_C). For all situations we examined, random error alone caused departures away from the true value (IPR_C > IPR_T) in roughly half of the simulation trials, as expected. Except in the most extreme cases, a reversal from causative to protective (IPR_M < 1) was observed in some of the trials. The percentage of simulations where IPR_M was less than IPR_T varied by the underlying IPR_T value and misclassification rates. Contrary to the apparent belief of many investigators that IPR_M is always less than IPR_T in the face of nondifferential exposure misclassification, only in the most extreme cases does this hold true.

Learning Objectives:

• Know that qualitative evaluation of exposure-measurement error does not necessarily work.
• Recognize the importance of quantitatively evaluating the effect of exposure-measurement error (and all systematic error) in study results.
• Identify one method to quantify the effect of exposure-measurement error in study results.

Keywords: Epidemiology, Methodology

Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.