Institute for Epidemiology and Prevention Research, Achterstrasse 30, D-28359 Bremen, Germany.
Stat Med. 2013 Jun 30;32(14):2335-48. doi: 10.1002/sim.5563. Epub 2012 Sep 18.
This paper addresses statistical issues in non-inferiority trials where the primary outcome is a fatal event. The investigations are inspired by a recent Food and Drug Administration (FDA) draft guideline on treatments for nosocomial pneumonia. The non-inferiority margin suggested in this guideline for the endpoint all-cause mortality is defined on different distance measures (rate difference and odds ratio) and is discontinuous. Furthermore, the margin enables considerable power for the statistical proof of non-inferiority at alternatives that might be regarded as clinically unacceptable, that is, even if the experimental treatment is harmful as compared with the control. We investigated the appropriateness of the proposed non-inferiority margin as well as the performance of possible test statistics to be used for the analysis. A continuous variant of the margin proposed in the FDA guideline together with the unconditional exact test according to Barnard showed favorable characteristics with respect to type I error rate control and power. To prevent harmful new treatments from being declared as non-inferior, we propose to add a 'second hurdle'. We discuss examples and explore power characteristics when requiring both statistical significance and overcoming the second hurdle.
本文针对主要结局为致命事件的非劣效性试验中的统计问题进行了研究。这些研究的灵感来自于最近食品和药物管理局（FDA）关于医院获得性肺炎治疗的指南草案。该指南中为全因死亡率终点建议的非劣效性边界是基于不同的距离度量（率差和优势比）定义的，并且是不连续的。此外，该边界使得在替代方案下进行非劣效性统计证明具有相当大的功效，而这些替代方案可能被认为在临床上是不可接受的，也就是说，即使与对照相比，实验治疗是有害的。我们研究了所提出的非劣效性边界的适当性以及可能用于分析的检验统计量的性能。FDA 指南中提出的边界的连续变体以及 Barnard 的无条件精确检验在控制 I 型错误率和功效方面表现出了良好的特性。为了防止有害的新疗法被宣布为非劣效，我们建议增加一个“第二个障碍”。我们讨论了例子，并探索了当需要同时具有统计学意义和克服第二个障碍时的功效特征。
