BIAS FROM NONDIFFERENTIAL MISCLASSIFICATION IN CASE-CONTROL STUDIES WITH 3 EXPOSURE LEVELS

In case control studies with three exposure levels where disease risk is associated with exposure, the direction of bias in the odds ratios (ORs) from nondifferential misclassification depends on the risk level, misclassification rates, and exposure distributions. To extend these generalizations, we present a graphical analysis of bias from nondifferential misclassification assuming linear and nonlinear monotonic increasing exposure-risk patterns. In both middle and upper exposure levels, bias is usually toward the null, increasing in magnitude as the misclassification rates increase and as the skewness of the exposure distribution increases. In the middle exposure level, bias away from the null may occur when the misclassification rate is low in the reference level and moderate to high in the upper exposure level, and risk increases with exposure. Bias away from the null does not occur in the upper exposure level. In both excess risk levels, crossover bias (that is, a reversal of the OR) may occur when exposure classification is worse than chance. The magnitude of bias away from the null is constrained by the unbiased OR of the upper exposure level, whereas that of crossover bias is constrained by the inverse of the unbiased OR of the upper exposure level.