Estimating and correcting for confounder misclassification.

Confounder misclassification is as pervasive a problem in epidemiologic studies as misclassification of the primary exposure variables. Adjustment using imperfectly classified confounders has been shown to remove only part of the confounding effect of that variable, most important when studying an exposure postulated to have a modest effect in the presence of a strong confounder. The Per Cent Adjustment (amount of confounding bias removed using the misclassified measure relative to the total confounding bias present) is derived for several case-control scenarios, and is found to have a stable relation to the sensitivity and specificity of confounder classification. Figures are provided for estimation of the completely adjusted odds ratio given the results from the misclassified proxy and an estimate of the extent of misclassification. This offers a method for quantitative assessment of the impact of residual confounding that results from confounder misclassification, illustrated with studies of coffee consumption and bladder cancer in the presence of confounding by cigarette smoking.