Institute of Medical Biometry and Informatics, University of Heidelberg, Heidelberg, Germany.
Stat Med. 2012 Dec 30;31(30):4352-68. doi: 10.1002/sim.5585. Epub 2012 Aug 29.
In the development of a new treatment in oncology, phase II trials play a key role. On the basis of the data obtained during phase II, it is decided whether the treatment should be studied further. Therefore, the decision to be made on the basis of the data of a phase II trial must be as accurate as possible. For ethical and economic reasons, phase II trials are usually performed with a planned interim analysis. Furthermore, the decision about stopping or continuing the study is usually based on a short-term outcome like tumor response, whereas secondary endpoints comprise stable disease, progressive disease, toxicity, and/or overall survival. The data obtained in a phase II trial are often analyzed and interpreted by applying the maximum likelihood estimator (MLE) without taking into account the sequential nature of the trial. However, this approach provides biased results and may therefore lead to wrong conclusions. Whereas unbiased estimators for two-stage designs have been derived for the primary endpoint, such estimators are currently not available for secondary endpoints. We present uniformly minimum variance unbiased estimators (UMVUE) for secondary endpoints in two-stage designs that allow stopping for futility (and efficacy). We compare the mean squared error of the UMVUE and the MLE and investigate the efficiency of the UMVUE. A clinical trial example illustrates the application.
在肿瘤学新治疗方法的开发中，II 期临床试验起着关键作用。基于 II 期获得的数据，决定是否应该进一步研究该治疗方法。因此，基于 II 期试验数据做出的决策必须尽可能准确。出于伦理和经济原因，II 期试验通常采用计划的中期分析进行。此外，停止或继续研究的决定通常基于肿瘤反应等短期结果，而次要终点包括疾病稳定、疾病进展、毒性和/或总生存。II 期试验获得的数据通常通过应用最大似然估计器（MLE）进行分析和解释，而不考虑试验的顺序性质。然而，这种方法提供有偏结果，因此可能导致错误的结论。虽然已经为主要终点推导了两阶段设计的无偏估计量，但目前尚无针对次要终点的无偏估计量。我们提出了用于两阶段设计的具有通用性的最小方差无偏估计量（UMVUE），这些设计允许因无效性（和疗效）而停止。我们比较了 UMVUE 和 MLE 的均方误差，并研究了 UMVUE 的效率。一个临床试验实例说明了该应用。
