Misclassification in case-control studies with two dichotomous risk factors.

The effects of misclassification are considered for a case-control study in which the effects of two dichotomous risk factors on disease risk are under investigation. When only one of the risk factors is misclassified, the summary odds ratio associated with that factor based on expected cell counts is biased towards unity, and the magnitude of the apparent interaction between the two variables is altered. When the misclassified variable is a confounder, the summary odds ratio for the exposures of interest remains partly confounded. Extensive tabulations of the results are given, for a range of prevalences, relative risks and misclassification rates of the factors. Even misclassification rates as low as 10% can prevent adequate control of confounding. The implications of this finding for the interpretation of epidemiological studies are examined.