groupsequential

Group-sequential (often written groupsequential) refers to a class of statistical designs used mainly in clinical trials and other experiments that allow multiple looks at accumulating data. In a group-sequential design, interim analyses are scheduled after prespecified amounts of information have been collected, and the trial may be stopped early for efficacy, futility, or safety. The overall Type I error rate is preserved by applying stopping boundaries at each look.

A central idea is alpha spending: the total Type I error used across all looks is allocated according to a boundary function, so that the probability of a false positive remains controlled. Commonly used boundaries include the Pocock boundary, which applies a nearly constant threshold across looks, and the O’Brien-Fleming boundary, which requires a very stringent boundary early on and becomes more permissive later. Flexible approaches, such as Lan-DeMets spending functions, allow boundaries to adapt to the actual timing and number of looks.

Key concepts include information time, which measures how much information (often related to sample size or number of events) has been accumulated at an interim analysis, and stopping rules for efficacy or futility. Group-sequential methods reduce expected sample size and can shorten trials when strong interim evidence emerges, but require careful planning and data monitoring to avoid operational bias. They are widely used in regulatory settings, with established guidance from authorities and a rich history of development by researchers such as Pocock, O’Brien, Fleming, and Lan and DeMets.