|
Abstract:
|
Sample size determination is critical in design of clinical trials. An adequately sized trial can have sufficient power to detect the treatment difference of interest yet without over-sizing the study. However, the traditional sample size approach based on a single point initial projection of the true treatment difference may fail to achieve this goal if this initial projection misses its target for various likely reasons. In this study, a new approach to the sample size problem via minimizing regret score is proposed, assuming that the true treatment difference, though fixed, is in a likely interval rather than a single value. The regret score is used to select a better design with uniformly sufficient power and a near optimal sample size with respect to all possible values of the true treatment difference in the chosen interval.
|
