A Direct Approach to Simultaneous Tests of Superiority and Noninferiority with Multiple Endpoints

In clinical trials it is often desirable to test for superiority and non-inferiority simultaneously. The hypotheses are formulated to test whether a new treatment is superior to a control on at least one endpoint, and non-inferior on all endpoints. The simulation studies of \citet{logan2008superiority} shows that most of the current testing methods are very conservative, especially when the non-inferiority margins are close to zero. In this paper we propose a method for the superiority and non-inferiority problem based on the lower one-side confidence intervals. Theoretically, we prove that our testing method can control the type I error at a pre-specified level $\alpha$, such as 0.05, 0.025 or 0.01, which is also demonstrated in our simulation study. Meanwhile, the simulation study show that our method has a higher power than several alternative methods when the non-inferiority margins are close to zero or when both endpoints have a positive value, hence, the proposed method successfully avoid the deficiency of being conservative and has a higher power. A real example about the efficacy and toxicity of an inhaled drug for asthma compared to placebo is used to illustrate the proposed method.