Tackling control risk problems in non-inferiority trials.

Abstract

Non-inferiority trials aim to show that major disease related outcomes with a new intervention are not importantly worse than with standard care. These trials are useful when the new intervention has some advantages over standard care (eg, toxicity, convenience, or cost). The ability to show non-inferiority, however, is sensitive to the control risk, the outcome frequency under standard care. Two control risk problems are described that can make non-inferiority trials underpowered or uninterpretable, and two ways of tackling these problems are outlined. Firstly, the choice of effect measure used to express the non-inferiority margin is critical: the effect measure must be based on understanding both the clinical setting and the implications for sample size. Which effect measures can lead to smaller or larger sample sizes is shown. Secondly, investigators need to consider, and potentially plan for, the possibility that the observed control risk might differ from the anticipated risk at the design stage of the trial. How the non-inferiority margin can be adapted in the trial analysis in a statistically principled manner is shown.