ABSTRACT
Recurrent events arise frequently in studies of chronic diseases with examples including fractures in osteoporosis trials, respiratory exacerbations in chronic bronchitis, or falls in the studies of mobility in the elderly. This chapter gives a clear presentation of the most useful methods for rigorous analysis of the resulting data, as well as methods for the design of clinical trials. Intensity functions completely characterize the event process and facilitate inference based on likelihood functions. Rate functions, which do not condition on the history of the event process, have great appeal for use in clinical trials where marginal treatment comparisons and causal inference are required; estimating functions play a central role in robust inference in rate-based analyses. Formula for sample size calculation are provided based on mixed Poisson models, and methods for blinded sample size re-estimation are provided. The distinction between the effects of event-dependent random censoring and terminal events is made; inverse probability of censoring weights can yield consistent estimators of standard rate functions in the former case, but marginal rate functions are advocated in the case of terminal events. An illustrative analysis is provided involving data from a cystic fibrosis trial where the events represent exacerbations of symptoms.
