Bias Due to Within-Subject Exposure Dependency With or Without Bias Due to Lack of Pairwise Exchangeability When...

The case-crossover study design has been proposed as a suitable design for use when a brief exposure causes a transient change in risk of an acute-onset disease. In pharmacoepidemiology, the condition of "brief exposure" is rarely satisfied because medication use is often chronic or successive, which may result in bias due to within-subject exposure dependency. Here we describe a simulation of a case-crossover study conducted within a cohort, where patients successively used a drug for 60 or more days and the rate ratio for the outcome occurrence was 4.0. Standard conditional logistic regression for the analysis produced overestimated odds ratios ranging up to 7.8. This bias due to within-subject exposure dependency from chronic use can be removed by the Mantel-Haenszel method or by our recently proposed weighting method. We also show that when some patients are censored after switching to another drug, a lack of pairwise exchangeability causes bias which is similar to bias due to an exposure time trend. This bias can be removed by using the case-time-control study design. We show that bias due to within-subject exposure dependency and lack of pairwise exchangeability occur independently and can occur separately or simultaneously, and we demonstrate how to detect and remove them.