Menon Sandeep, Chang Mark
Pfizer, Inc., Cambridge, MA, USA.
J Biopharm Stat. 2012;22(4):641-61. doi: 10.1080/10543406.2012.676532.
The rising cost of clinical trials is impeding the development of new drugs. There is an acute need for critical evaluation and innovate thinking while designing the trial. Adaptive design has been repeatedly called upon in the last decade as one of the prescriptions for this intricate problem. From a pure statistical perspective, the adaptive design framework depends heavily on the appropriate selection of the type of test statistics and stopping boundaries. There are several methods proposed in the literature, based on different test statistics and stopping boundaries. All of these methods are rigorous in controlling type I error. In this paper, we group combination p-value methods into major categories along with their stopping boundaries. We review and compare these methods based on their operating characteristics, including average sample size and maximum sample size under null and alternative hypothesis, power, and early stopping probabilities. The optimal interim analysis timing and alpha spending function were used as the independent factors for this assessment. We propose an evaluation matrix and establish a framework to assess the most efficient design in order to assist in "one stop shopping."
临床试验成本的不断上升正在阻碍新药的研发。在设计试验时，迫切需要进行批判性评估和创新思维。在过去十年中，适应性设计作为解决这个复杂问题的方法之一，被反复提及。从纯粹的统计学角度来看，适应性设计框架在很大程度上依赖于检验统计量类型和停止边界的适当选择。文献中基于不同的检验统计量和停止边界提出了几种方法。所有这些方法在控制I型错误方面都很严格。在本文中，我们将组合p值方法及其停止边界分为主要类别。我们根据它们的操作特征对这些方法进行回顾和比较，包括在原假设和备择假设下的平均样本量和最大样本量、功效以及早期停止概率。最优期中分析时间和α消耗函数被用作该评估的独立因素。我们提出了一个评估矩阵，并建立了一个框架来评估最有效的设计，以协助进行“一站式购物”。
