Re‐use of case–control data for analysis of new outcome variables

Abstract

Case-control studies are usually defined to investigate risk factors for a single disease of interest. However, subsequent to data collection, investigators may wish to examine as an 'outcome' a variable that was an exposure in the original study. A naive analysis that disregards the sampling strategy that gave rise to the data is clearly prone to bias. We present here a simple approach to the analysis of such data using an appropriately weighted regression model. Viewing the problem as a two-stage design provides a unified framework for recognizing and defining the necessary weights when confronted with a variety of real data problems that at first seem unrelated. We provide illustrations that highlight the generality of the approach and demonstrate that the method gives essentially the same results as more specialized methods that require non-standard tools for analysis.