ASPECTS OF MISCLASSIFICATION OF CONFOUNDING FACTORS

Misclassification of exposure in epidemiologic investigations has been extensively studied and is now well understood. In contrast, misclassification of confounding factors has been much less investigated. First, we consider a situation with confounding by age, in which misclassification is introduced through stratification of this inherently continuous variable. This misclassification turns out to be benign: 75% of the original confounding is removed by stratification into two age classes and more than 90% by using three age classes. Second, we consider a situation with serious confounding and serious misclassification of the confounding factor but no misclassification of the exposure. In this situation, the misclassification turns out to be of importance. After stratification for the misclassified confounding factor, it appears as though the exposure has a stronger effect on the incidence than the confounder, which is the reverse of the true situation.