UW Madison UW School of Medicine and Public Health
UW Biiostatistics & Medical Informatics UW Biostatistics & Medical Informatics UW Madison UW Biostatistics & Medical Informatics Site Map


General Departmental Seminar Series

Modeling Treatment Effects on the Recurrent Event Mean
and Rate in the Presence of a Terminating Event

Doug Schaubel, Ph.D
Department of Biostatistics at the University of Michigan

Friday, Friday 8, 2008
12:00 pm
5275 MSC


Often in biomedical studies, the event of interest is recurrent (e.g., hospitalization) and the recurrent event sequence is subject to being stopped by a terminating event (e.g., death).  There are several methods in the existing literature for analyzing recurrent events in the presence of a terminating event.  For example, marginal methods estimate the recurrent event mean, averaging over the survival distribution, while conditional methods estimate the recurrent event rate given survival. In comparing treatment options, a marginal method is preferred when interest is focused on the overall public health impact of a treatment. If treatment affects survival, then the marginal treatment effect is likely to vary over time. In settings where the treatment effect is time-dependent, the cumulative treatment effect is usually of much greater interest to investigators than the instantaneous effect. We propose semiparametric methods for estimating the cumulative treatment effect on the marginal mean number of recurrent events. Three estimators are proposed; the first uses nonparametric estimators of the survival and conditional event rate, while the second and third estimate the mean directly. Each of the proposed methods can be applied to observational data, since imbalances in the treatment-specific covariate distributions are adjusted out through inverse weighting. Large-sample properties are derived, with their applicability in finite samples assessed through simulation. The proposed methods are applied to kidney failure data.

Return to seminar list