Abstract:
|
Noninferiority (NI) studies are gaining popularity in clinical trials. The objective of NI trials is to explore potential new treatments for existing treatments (reference treatments) by verifying that new treatments maintain a sizable portion of the efficacy of reference treatments. The loss of efficacy of new treatments, as compared to reference treatments, can be compensated by its other benefits, such as alleviating side effects, lowering costs, and simplifying intricate treatment regimens. Statistical methods have been developed for the simultaneous testing of NI of multiple new treatments against multiple reference treatments. However, these procedures are based upon the assumption of equal variances in all treatments. In this paper, we use a simulation study to explore the undesirable effects of a violation of the homogeneous variance assumption of using these procedures on the familywise Type I error rate (FWER). To remedy the problem, we propose procedures that are more appropriate because these procedures are able to control the FWER. A power study is then conducted to compare the different procedures, and a clinical example is given for illustrative purposes.
|