Continuously monitored stopping boundary methodologies: the issues of sensitivity, association and trial suspension

Thall et al. consider a continuous monitoring strategy for multiple discrete outcomes to determine whether a trial should terminate early. We evaluate important issues raised in the application of a continuous monitoring strategy for multiple outcomes. Specifically, we evaluate: (i) the sensitivity of such a methodology to small perturbations in the stopping boundaries; (ii) the need to employ accrual buffers when a trial approaches a stopping boundary-a large buffer implying that temporary suspension is unwarranted; and (iii) the role of association among the multiple outcomes of interest. Simulation studies demonstrate that the methodology is sensitive to small perturbations in the stopping boundaries, that the size of an accrual buffer can vary widely over the course of a trial, and that the extent of association among multiple outcomes plays a large role in determining the stopping properties of a trial. We illustrate these issues using the HLA non-identical donor bone marrow transplant trial, with two discrete outcomes.