Something for Nothing in Noninferiority/Superiority Testing: A Caution

Abstract

Clinical trials comparing two active treatments often are conducted to show that an investigational treatment is not inferior to a standard therapy. When noninferiority is shown, a conditional analysis can be used to establish whether the investigational treatment is superior to the standard. Alternatively when superiority is desired but not shown, a conditional analysis can be used to establish noninferiority. Such analyses can be used without adjustment to the type I error rate, under certain assumptions. This paper presents cautions in implementing such an analysis strategy. Equivalence and noninferiority analyses often use the per protocol group of subjects in the primary analysis, while analyses that want to show a difference generally use a wider group of subjects from the study. If the two analyses use different groups of subjects, the type I error rate may be inflated unless a stepwise procedure in the proper direction is instituted. Strategies for planning and interpreting such analyses are discussed.