Comparison of Two Groups and Test for Non-Inferiority

Abstract

In situations, where inclusion of a placebo arm is not appropriate, the experimental treatment has to be compared with a therapy for which efficacy has already been demonstrated in the past. Frequently, the new treatment is not expected to be more efficacious than this active control but that there is an advantage with respect to another aspect, for example, a better safety profile or a more comfortable way of administration. Then, it is accepted that, due to this advantage, efficacy of the new therapy is allowed to be smaller by a clinically irrelevant amount. This concept leads to so-called non-inferiority trials. In this chapter, methods for sample size calculation are presented for these trials and the case of two-group comparisons. For normally distributed outcomes, the scenarios that the non-inferiority margin is expressed in terms of the difference and ratio of means, respectively, are considered. Furthermore, methods for continuous and ordered categorical data, binary, and time-to-event endpoints are given. For binary data, the methods cover asymptotic and exact unconditional tests as well as the cases that the risk difference, the risk ratio, or the odds ratio are used as effect measure.