Interpretation and Analysis of Differential Exposure Variability and Zero-Exposure Categories for Continuous Exposures

We reconsider the situations discussed by Robertson et al1 and reach different conclusions. First, differential exposure variability among cases and controls is not a physical cause of quadratic dose-response relations; rather, differential variability is an effect of either nonlinear exposure effects, bias, or both. As such, it is not a methodologic problem, and attempts to induce comparable case and control exposure variances can produce bias. Second, deleting unexposed subjects from the doseresponse analysis is not always superior to including them. When there is no definitive basis for deleting or including the unexposed subjects, it is advisable to perform and compare analyses with and without those sub? jects.