Michael A. Proschan, National Institutes of Health

Two-Stage Designs to Adaptively Modify Sample Size in Clinical Trials

Sample size calculations are essential to ensure a well-designed clinical trial. Unfortunately, they depend on unknown nuisance parameters and the treatment effect. The nuisance parameter for the comparison of means with a continuous outcome is the variance, while for the comparison of proportions with a dichotomous outcome it is the overall event probability. Ideally, the treatment effect parameter is chosen as the minimum clinically relevant treatment difference, but this may also not be known, and in practice one often chooses a treatment effect that was similar to those seen in other trials. But those other trials may not be sufficiently similar, so why not use data from the current trial to estimate these parameters? This talk discusses two-stage clinical trials in which the first stage is used to estimate parameters upon which the ultimate sample size is based. The first part of the talk focuses on trials in which only the nuisance parameter must be estimated, and the second part considers trials in which the treatment effect is estimated.