Effect of Nondifferential Misclassification on Estimates of Odds Ratios with Multiple Levels of Exposure

Abstract

Nondifferential misclassification of exposure status with a dichotomous exposure will produce biased estimates of odds ratios such that the misclassified odds ratio is always biased toward the null value. However, when an exposure classification has more than two levels, empirical data indicate that the direction of bias is less predictable. Analysis of an algebraic model of multi-level exposure misclassification reveals that all odds ratios based on the misclassified data are constrained between the nonmisclassified odds ratio for the most extreme category and the inverse of this value. This implies that the misclassified odds ratio for the most extreme exposure level will be biased toward the null but that odds ratios for intermediate levels of exposure could be biased away from the null value. Further, the amount of bias depends not only on the misclassification rates but also on the distribution of subjects across exposure levels. If it is assumed that misclassification is likely to occur only between adjacent categories, the range of possible misclassified odds ratios is reduced but is still sufficient to permit serious distortion of an exposure-response relation. In general, biases away from the null occur only for intermediate levels of exposure. Reversal of an exposure-response relation is more likely to occur when misclassification rates are high (especially between nonadjacent levels) and when the number of exposure levels is low.

Pooled reanalysis of cancer mortality among five cohorts of workers in wood-related industries

“…Within studies, misclassification will have occurred because persons were inaccurately assigned to exposure categories due to the limited exposure information available. Since the exposure information used was obtained prior to the occurrence of disease, the misclassification would be nondifferential and would tend to obscure potential relationships between wood dust and disease and could distort dose-response relationships (26,27). Further misclassification may have occurred between studies because differences in the detail of the exposure information available could have caused the accuracy of exposure assignment to vary by study.…”

The effects of recall errors and of selection bias in epidemiologic studies of mobile phone use and cancer risk

“…They also increase the uncertainty of risk estimates, making it more likely that real associations are not detected (Armstrong, 1998). However, for comparisons of exposure categories within polytomous exposures, bias can be in either direction (Dosemeci et al, 1990;Birkett, 1992;Dosemeci and Stewart, 1996;Armstrong, 1998). Systematic under-or overestimation of phone use leads to a bias of risk estimates upwards or downwards: if all subjects overestimate, measured relative risk estimates will be lower than the true relative risk, and vice versa.…”

Stability of Wertheimer–Leeper wire codes as a measure of exposure to residential magnetic fields over a 9- to 11-year interval

“…Statistical simulation studies have found that under extreme misclassification with three exposure levels, dose -response relationships can be inverted or distorted when observed odds ratios shift away from unity in intermediate exposure categories ( Dosemeci et al, 1990;Birkett, 1992;Correa -Villasenor et al, 1995 ). This occurred for the observable OHCC risk estimate in the threshold dose -response curve, slightly distorting the true shape.…”