ABSTRACT
Group sequential procedures are a mixture of trajectory analysis (i.e., they compute the likelihood that the test statistic would have followed the path it did if there was no treatment effect), and classical statistics (the answer is in terms of type I error rates). Thus, group sequential procedures construct a boundary for the test statistic at each monitoring point in the clinical trial. If the test statistic at any of these points is larger than the boundary at that point in time, we say that the monitoring
rule suggests that early termination be considered. Rules such as those of Lans– DeMets are very popular group sequential procedures [2,3].
