ABSTRACT
In the standard self-controlled case series (SCCS) model, the observation period for each case is partitioned into non-overlapping segments on which the age and exposure effects are assumed to be constant. In the standard SCCS model, age and exposure effects are assumed to be piecewise constant. The age and exposure levels are represented by parameters, the values of which are estimated by maximising the SCCS likelihood. All SCCS models considered are proportional incidence models, in which the different components are combined multiplicatively. The standard SCCS model is a log-linear product multinomial model. Several techniques can be used to fit the model using standard statistical software. A common misconception about SCCS studies is that risk periods must be of short duration and transient. The main modelling issue confronting SCCS analyses of this type is potential confounding between age effects and exposure effects. Curtailing the observation period at a time determined by the exposure history is perfectly acceptable in SCCS studies.
