ABSTRACT
A non-inferiority analysis is frequently conducted based on the determination of a non-inferiority margin or threshold. The choice of the margin should depend on prior experience of the estimated effect of the active control in adequate, well-controlled trials, and account for regression-to-themean bias, effect modication, and clinical judgment. The non-inferiority margin must be small enough to preclude that a placebo (or a treatment that is no better than placebo on a given endpoint) is noninferior to the active control. Other concerns about the non-inferiority margin might make the margin even smaller, but it should not be larger than the smallest anticipated difference between a placebo and the active control in the setting of the noninferiority trial.
