ABSTRACT

Despite a great effort, we often face a high degree of uncertainty about

parameters when designing a trial or justifying the sample size at the design

stage. This could involve the initial estimates of within-or between-patient

variation, a control group event rate for a binary outcome, the treatment

effect desired to be detected, the recruiting pattern, or patient compliance,

all of which affect the ability of the trial to address its primary objective

(Shih, 2001). This uncertainty can also include the correlation between the

measures (if a repeated measure model is used) or among different variables

(multiple endpoints, covariates). If a small uncertainty of prior information

exists, a classical design can be applied. However, when the uncertainty is

greater, a classical design with a fixed sample size is inappropriate. Instead,

it is desirable to have a trial design that allows for reestimation of sample

size in the middle of the trial based on unblinded data. Several different

algorithms have been proposed for sample-size reestimation, including the

conditional power approach and Cui-Hung-Wang’s approach based on the

ratio of observed effect size versus expected effect size.