ABSTRACT
This chapter considers the standard self-controlled case series (SCCS) models so far share the following characteristics: the age effect is assumed to be constant on pre-defined age groups; the exposure effect is also assumed to be piecewise constant on pre-defined risk periods; events are of a single type; and exposures are discrete. It discusses the constraints imposed by these characteristics, and introduces a wider range of models. In the semiparametric SCCS model, the age effect is left completely unspecified, other than it being non-negative and bounded. It turns out that the maximum likelihood estimator of the cumulative relative age effect is a non-decreasing step function, taking steps only at event times. The chapter considers SCCS models for events of several different types and presents a SCCS model for quantitative exposures. It also discusses SCCS models for environmental exposures, and their relationship with other Poisson models.
