The use of interim analysis for sample size adjustment

A frequent problem in designing a clinical trial is the need to rely on guesses of the magnitude of parameters, such as means and variances, needed to perform the calculations of required sample size. A sample size adjustment design incorporates a protocol-specified interim analysis conducted not to test a hypothesis but rather to adjust the sample size of the trial. This paper extends previous work on internal pilot studies for normally distributed outcome variables to the case of a binomial outcome variable. Through a specific example from a cancer clinical trial, the effect of the sample size adjustment process on significance level, power, and average sample size is addressed. Design issues such as timing of the interim analysis, potential for unblinding, and the effect of disparity of initial guesses of parameter values and estimates derived from interim analysis are also discussed. The sample size adjustment process is found to have negligible effect on significance level and to deliver reasonable power. Through proper use of a data monitoring board the potential for unblinding is negligible. Considerable savings in sample size can be achieved with this design over a traditional fixed sample size design. Sample sizes larger than originally thought necessary are automatically provided by the adjustment process when needed.