Design of clinical trials for recurrent events with periodic monitoring
Shigeyuki Matsui, Hideaki Miyagishi · Statistics in Medicine · 1999
This paper presents a design for randomized clinical trials in which incomplete data are collected on the occurrence of potentially recurrent events through periodic monitoring. In particular, events are assumed to arise according to a point process, but information is available at the times of monitoring only if one or more events has occurred since the preceding monitoring point. The event process is modelled via a piecewise Poisson process, and a proportional rates model is introduced to represent the difference in event rates between treatment groups. The design was developed on the basis of a Wald-type test derived from the generalized estimating equations of Liang and Zeger (Biometrika 73, 13-22 (1986)). Robustification of the variance of the estimator of the treatment effect was considered under a random effects model with a semi-parametric mixture distribution. The design was adopted to address issues which arose in an osteoporosis trial conducted in Japan.