Other Models and Topics
The central theme of the previous eight chapters is modeling survival time when there is a single event of interest that terminates observation. This situation, in which each subject may experience the event of interest only once, describes the majority of applied settings when follow-up time is observed. There are situations, however, when the event of interest may occur more than one time for each subject. Another is a setting in which the disease process progresses through several stages, each with its own event time. For example, a subject’s cancer may be treated and then recur some time later with treatment and recurrence happening several times. The follow up may ultimately end due to death. We discuss several approaches to modeling recurrent event data in Section 9.2.
Another extension of the standard modeling situation occurs when subgroups of responses are correlated due to study design. This lack of independence of response can occur in any setting in which survival time is influenced by unmeasured factors that are the same within groups of subjects and are thought to have significant group-to-group variability. When these factors are present in the usual normal errors linear model setting, they are called random effects. Survival analysis models incorporating such factors are called frailty models. These models have also been suggested for use in the recurrent event setting. We discuss frailty survival-time models in ...