ABSTRACT

This chapter develops a self-controlled case series (SCCS) model to handle event-dependent exposures. These are exposures that may be influenced by prior events, thus violating a key assumption. This takes us away from likelihood-based methods altogether, and into the realm of estimating equations. The chapter describes a SCCS model to handle event-dependent observation periods, specifically in the situation where occurrence of an event may precipitate the end of observation, thus violating another key assumption. The approach we propose requires an additional modelling step to obtain weights with which the SCCS likelihood is then adjusted. The chapter discusses more specifically the application of the SCCS method when the event of interest is death. In this situation both the observation period and the exposure are event-dependent in an extreme sense. The chapter provides heuristic accounts of the key ideas that underpin the mathematics which, as elsewhere in the book, are developed in starred sections which may be skipped.