ABSTRACT
In Chapters 6 and 7, we discussed two-stage designs. Group sequential design is a generalization for multiple stage designs. The classical group sequential procedure (design, monitoring, and analysis) is probably the most commonly applied method to facilitate the conduct of interim analysis in clinical trials, especially trials with mortality or severe or irreversible morbidity as the primary endpoints. In clinical trials with such endpoints, the fundamental consideration is an ethical obligation to modify or terminate the study at the earliest opportunity to prevent administration of a less favorable (for efficacy or toxicity) therapy to additional patients. Conducting these clinical trials in stages, with group evaluation sizes predetermined by a certain schedule, provides logistical feasibility over fully sequential methods, because the endpoint in these trials can often take a long time to occur. However, the advantage of reducing the average sample number still applies. Many different methods for group sequential procedures are available and include different stopping boundaries, different type I error spending/use functions, conditional or predictive power approaches, Bayesian methods, etc. These procedures are based on a fixed maximum sample size. We call them classical group sequential methods to distinguish them from the more recent development in using adaptive group sequential methods, in which one can extend the maximum sample size or modify other design specifications, such as study population or endpoint, as the trial is ongoing. With more flexibility and potential of introducing bias by monitoring interim data, regulatory authorities have developed strict guidance, particularly for group sequential trials. The logistics of a group sequential trial are also complicated and require careful planning and execution.
