Confounder prevalence and stratum-specific relative risks: implications for misclassified and missing confounders.

Abstract

In this article, I describe a simple relation between stratum-specific risk ratios and the distribution of strata in disease-exposure categories. The relation does not seem well known, but it presents an alternative way to view how to make adjustments to biased stratum-specific relative risk estimates when bias is due to either a misclassified or missing confounding variable. For instance, it demonstrates that there would be no misclassification bias if categories of a confounder surrogate have the same prevalence as those of the true confounder, in each disease-exposure category, regardless of whether the surrogate is a good or bad approximation to the true confounder. It also provides an insight into the conditions required to ensure uniformity of stratum-specific relative risks.