ABSTRACT
In previous chapters, we discussed the fixed-size design without interim analysis, as well as sequential designs with interim analyses. The alphaspending function approach of the classical group sequential (GS) design is particularly useful in providing a flexible tool for monitoring the safety and efficacy aspects of clinical trials. In particular, this approach provides flexibility in the schedule and the number of interim analyses. The fundamental structure of the fixed-size design without interim analyses and the classical GS procedure with interim analyses relies on maximum, fixed, and prespecified information in the study protocol. The information times (fractions) used for interim analyses are based on this fixed, prespecified, maximum information. What if the prespecified maximum information is incorrect? Can it be altered later if it is found to be inadequate? These questions are often asked. Recall that in Chapter 4 we discussed how sample size calculation, study power, and values for design parameters such as treatment effect, withingroup variance, compliance rate, etc., had to be assumed, ideally based on data derived from prior studies with very similar design conditions, especially for Phase III confirmatory studies. However, we often find that the prior studies, if any, usually involve different patient populations (defined by the inclusion and exclusion criteria and different staging technologies), medical practices (e.g., allowable concomitant medications), study durations, or treatment regimens. An intriguing idea, therefore, would be to see if these assumptions could be verified or updated with interim data from the ongoing study. For example, after checking the interim data, we find that an extension of the (initial) maximum information is necessary to ensure the study power. With this finding, we would then need to modify the original sample size and perhaps also the conventional fixed-size test or the classical sequential procedure. In performing all these modifications, we would then need to ensure that the type I error rate remained protected. A recent development in this area is termed adaptive designs or flexible designs. In this chapter, we discuss three topics related to the adaptive or flexible designs: (i) sample size reestimation (SSR) with continuous or binary endpoints, (ii) monitoring trial duration for maximum information (for event-driven trials) or maximum duration (for duration driven trials) with survival endpoints, and (iii) modification of the classical GS alpha-spending function. In both SSR and monitoring trial duration, the methods involve only blinded data. For the modification of the alpha-spending function, we use unblinded interim data. SSR and monitoring trial duration with unblinded data remains an ongoing research topic and
is beyond the scope of this book. Overall, the recent development of adaptive designs offers trials great flexibility in making midcourse modifications of the maximum information or duration, but there could be some potential pitfalls. Examples are given to illustrate some vigilant points.
