Methods for planning and analysis of clinical phase II trials in oncology

在项目的当前资助期内，为肿瘤学临床II期研究开发了两阶段自适应设计，在试验过程中，允许在控制显著性水平的同时对计划阶段固定的初始样本量进行任意（特别是：数据驱动）修改。这些设计允许处理在研究过程中发生的不可预见或不可预测的变化，而不会影响指定的I型错误率。例如，在计划阶段所作的关于治疗效果的假设可能会受到中期分析结果的质疑。因此，减少或增加样本量将是一个有希望的选择。然而，目前，在这种情况下，各种可能的样本量重新计算策略中哪一种是“最优”的仍然是一个悬而未决的问题。这个问题将在项目的继续阶段进行研究。根据这些调查的结果，应提出建议，在各自的情况下选择适当的设计或样本量重新计算策略。此外，还应制定方法，以充分处理为中期分析计划的样本量的过剩或不足。这种情况在实践中经常出现，例如由于患者招募停止得太晚或太早。所有现有的方法都显示出严重的缺陷；除此之外，它们不控制预先指定的显著性水平。此外，目前还没有方法可用于计算具有适应性设计的II期肿瘤试验的有效p值、点估计和置信区间。众所周知，为各自的单阶段设计开发的方法的应用会导致扭曲的结果，从而导致对研究的有偏见的解释。当我们的项目继续进行时，我们的目标是开发具有适应性设计的II期肿瘤试验的分析方法，考虑到研究设计的顺序和灵活性，从而获得公正的结果。将审查所创建方法的特征，并就哪种方法是首选方法给出建议。开发的方法将在一个具有用户友好界面的R软件包中实现。这将导致项目成果直接转化为实践。

Within the current funding period of the project, two-stage adaptive designs were developed for clinical phase II studies in oncology that, during the course of the trial, allow for arbitrary (especially: data-driven) modifications of the initial sample size fixed in the planning phase while controlling the significance level. These designs allow for addressing unforeseen or unpredictable changes occurring during the study conduct without compromising the specified type I error rate. For example, assumptions regarding the treatment effect made in the planning stage might be put into question by the results of an interim analysis. It would then be a promising option either to reduce or to increase the sample size. Currently, however, it is still an open question which one of the various possible strategies for sample size recalculation is "optimal" in such a situation. This problem will be examined in the continuation phase of the project. Based on the results of these investigations, recommendations shall be provided about the choice of an appropriate design or sample size recalculation strategy in the respective situation at hand. Moreover, methods are to be developed that allow to adequately deal with over- or underrunning of the sample size planned for the interim analysis. Such a situation arises quiet often in practice, for example due to a too late or too early stop of patient recruitment. All currently available methods show serious deficits; amongst others, they do not control the pre-specified significance level.Furthermore, there are currently no methods available for calculating valid p-values, point estimates andconfidence intervals in phase II oncology trials with an adaptive design. It is well-known that an application of methods developed for the respective one-stage designs lead to distorted results and, consequently, to a biased interpretation of the study. When our project is being continued, it is our aim to develop analysis methods for phase II oncology trials with adaptive design that take into account the sequential and flexible nature of the study design leading to unbiased results. The characteristics of the created methods will be examined, and recommendations will be given about which of the methods are to be preferred.The developed methods will be implemented in an R software package with a user-friendly user interface. This will lead to a direct transfer of the project results into practice.