Abstract
In a noninferiority trial without a placebo arm, a basic and vital design specification is the noninferiority margin, which is required to be smaller than or no greater than the effect of the selected active control under the noninferiority trial setting. The FDA noninferiority draft guidance document stipulates that the noninferiority margin is determined from a two-step process: (1) derive a statistical margin, relying on the guidance obtained from the selected relevant historical trials that provide the estimates of the effect of the active control and the variances associated with the estimates and taking into account unverifiable assumptions such as the constancy assumption, and (2) use clinical judgment to determine whether the statistical margin needs to be further shrunk to generate a clinical margin reflecting “noninferiority.” All the methods for margin determination rest upon subjective judgment and unverifiable assumptions. Additional issues such as multiplicity and covariate adjustments need to be carefully considered in the process of deriving the margin. Synthesis methods might help at Step 2 to generate a clinical margin. When there are multiple endpoints and interim analyses involved in such noninferiority trials, additional complexity resulting from multiple testing needs further attention in handling the statistical significance level—alpha. The long-standing controversy between intent-to-treat analysis and per-protocol analysis remains, though, in some cases, on-treatment analyses may be considered as an alternative to the intention-to-treat analysis to mitigate the impact of unfavorable bias on statistical inference for noninferiority.
Acknowledgment
The authors thank Dr. Brian Wiens for inviting us to contribute this article in honor of Dr. Robert O’Neill for his great contributions to regulatory science statistics over the years.