|
Abstract:
|
Overdispersed event-count outcomes are frequently modeled with a negative binomial distribution in clinical trials. Oftentimes in the design stage, estimates of the event rates of the various treatment arms are unavailable or characterized by large uncertainty. Therefore, it is advantageous to consider a design that allows for mid-course sample size adjustment based on estimates derived from interim data in order to reduce the chances of underpowering the study if erroneous treatment effect assumptions were made at the outset. In this talk, we propose a framework to perform unblinded sample size re-estimation in such settings, and illustrate the performance and operating characteristics of the proposed design via simulations.
|
