Abstract
A key assumption of the self-controlled case series (SCCS) method is that exposures should not depend on the event of interest. However, treatments such as vaccines may be deferred after an adverse health event. One suggestion to handle such delayed exposures is to include a pre-exposure window in the SCCS model. We study the impact of such adjustments and of exposure deferment on the SCCS relative incidence estimates. We obtain explicit results in a simplified setting, and investigate more realistic scenarios by simulation. We develop some practical recommendations for sensitivity analyses: when the delayed exposures remain within the observation period, no adjustment is needed. When exposures are delayed beyond the end of the observation period, an adjustment may be required. In some circumstances the SCCS model for event-dependent exposures should be used rather than the standard SCCS model. These options are illustrated with three practical examples.