What is a non-inferiority trial?

According to the FDA guidance, a non-inferiority trial aims to show that the new drug falls short of the active control by no more than a pre-specified amount, the margin M. The null hypothesis is that control minus new drug is at least M; the test checks whether the upper bound of the two-sided 95% confidence interval for that difference is below M, which corresponds to a one-sided significance level of 0.025. According to Schumi and Wittes (2011), an equivalence trial instead requires both ends of the interval to lie within the margin.

The guidance separates two levels of margin. M1 is the whole effect of the active control over placebo, estimated from past trials and, unless the trial has a placebo arm, not measured in the current one; M2 is the largest clinically acceptable loss of effect and is usually chosen smaller than M1. Schumi and Wittes describe a ‘95-95’ approach: M1 taken from the lower limit of the 95% confidence interval of the control’s effect against placebo, and M2 set to preserve a fraction of that effect, for example 50% or 75%.

The non-inferiority and equivalence checklist on the CONSORT website (2006 version) asks, in the sample-size item, whether the calculation used a non-inferiority or equivalence criterion and for the margin and the rationale for its choice. Piaggio et al. (2012) present the updated extension based on CONSORT 2010.