Analysis of recurrent events: application to a clinical trial of colony stimulating factor with the endpoint of febrile neutropenia.

This paper discusses the analysis of randomized clinical trials where the primary endpoint is a recurrent event observed during a time period which can vary among patients. In particular it describes a method of estimating robust confidence limits for clinically relevant parameters. As an illustrative example, a clinical trial of a macrophage colony stimulating factor (M-CSF) for patients with acute myeloblastic leukaemia (AML) is introduced where the endpoint is the incidence and duration of febrile neutropenia. The causal effects in the framework of Rubin's deterministic model are defined as estimands. A non-parametric method based on the permutation test, using an accelerated search procedure proposed by Garthwaite, is compared with other standard methods by simulation and by application to the AML trial data.