Many different criteria have been proposed for the selection of a stopping rule for group sequen- tial trials. These include both scientific (e.g., estimates of treatment effect) and statistical (e.g., frequentist type I error, Bayesian posterior probabilities, stochastic curtailment) measures of the evidence for or against beneficial treatment effects. Because a stopping rule based on one of those criteria induces a stopping rule on all other criteria, the utility of any particular scale relates to the ease with which it allows a clinical trialist to search for sequential sampling plans having de- sirable operating characteristics. In this paper we examine the use of such measures as conditional power and predictive power in the definition of stopping rules, especially as they apply to decisions to terminate a study early for “futility”. We illustrate that stopping criteria based on stochastic curtailment are relatively difficult to interpret on the scientifically relevant scale of estimated treat- ment effects, as well as with respect to commonly used statistical measures such as unconditional power. We further argue that neither conditional power nor predictive power adhere to the stan- dard optimality criteria within either the frequentist or Bayesian data analysis paradigms. Thus when choosing a stopping rule for “futility”, we recommend the definition of stopping rules based on other criteria and careful evaluation of the frequentist and Bayesian operating characteristics that are of greatest scientific and statistical relevance.
Emerson, Scott S.; Kittelson, John M.; and Gillen, Daniel L., "On the Use of Stochastic Curtailment in Group Sequential Clinical Trials" (March 2005). UW Biostatistics Working Paper Series. Working Paper 243.