The effect of misclassification in matched-pair case-control studies.

It has been noted that nondifferential misclassification of the study exposure can produce more severe bias and power loss in pair-matched case-control studies than in unmatched studies, and that the "excess bias" is a direct function of the closeness of matching on correlates of exposure. This paper illustrates the extent of this excess bias for several case-control situations. It is demonstrated that when the misclassification is differential, bias may be either increased or decreased by matching. A method for correcting for misclassification bias in matched-pair studies is also presented.