Alternatives for discounting in the analysis of noninferiority trials.

Determining the efficacy of an experimental therapy relative to placebo on the basis of an active-control noninferiority trial requires reference to historical placebo-controlled trials. The validity of the resulting comparison depends on two key assumptions: assay sensitivity and constancy. Since the truth of these assumptions cannot be verified, it seems logical to raise the standard of evidence required to declare efficacy; this concept is referred to as discounting. It is not often recognized that two common design and analysis approaches, setting a noninferiority margin and requiring preservation of a fraction of the standard therapy's effect, are forms of discounting. The noninferiority margin is a particularly poor approach, since its degree of discounting depends on an irrelevant factor. Preservation of effect is more reasonable, but it addresses only the constancy assumption, not the issue of assay sensitivity. Gaining consensus on the most appropriate approach to the design and analysis of noninferiority trials will require a common understanding of the concept of discounting.