Some statistical considerations in the analysis of case-control studies when the exposure variables are continuous measurements.

This paper focuses on some statistical considerations in the estimation of dose-response in case-control studies when the exposure variables are continuous measurements. The first point is that the effects of differential variability in the exposure distributions over cases and controls cannot be differentiated from a true quadratic risk model. The second point is that when dealing with variables where zero denotes no exposure, it is important to treat the unexposed subjects separately from those who were exposed. Failure to do so can lead to differential variability among cases and controls and the resulting confounding with a quadratic risk model. Both of these points are illustrated by an example.