Integrated planning of pilot studies and confirmatory studies in clinical research

在临床研究中，经常进行试点研究，以获得有关新疗法的治疗效果的信息。根据结果，决定是否进行后续的确证性试验，如果答案是肯定的，则假设哪种治疗效果用于样本量计算。目前用于确定试点研究样本量的方法存在严重缺陷。有一类程序是基于没有量化标准证明合理的临时规则。作为一种替代方法，初步研究的样本量与确证性试验的样本量类似，但出于可行性考虑，I型和II型错误率的值更高。然而，这与试点研究的非证实性性质严格矛盾。另一个不足之处在于，在确定试点样本量时，文献中提出的方法都没有考虑到随后的验证性试验和项目的总体样本量。计划中的项目旨在为广泛的设计方案制定试点和验证性研究的综合规划方法。新方法允许根据定义的参数值和样本量确定整个项目（初步研究+验证性研究）成功的概率。此外，效用函数的构造，包括所需的资源，用于执行程序，以及在一个成功的程序的情况下的收益。通过最大化期望效用，最优决策规则和最优样本量的试点研究。应用这项研究的结果将允许学习计划的规划，最佳平衡成本和收益。

In clinical research, pilot studies are frequently performed to obtain information about the treatment effect of a new therapy. On the basis of the results, it is decided whether to perform a subsequent confirmatory trial and, if the answer is positive, which treatment effect is assumed for sample size calculation. Currently available methods for determining the sample size of a pilot study show severe disadvantages. One class of procedures is based on ad hoc rules that are not justified by quantitative criteria. As an alternative approach, the sample size of the pilot study is determined analogously to that of confirmatory trials but, due to feasibility considerations, with higher values for the type I and type II error rate. This, however, strictly contradicts to the non-confirmatory nature of pilot studies. A further deficit lies in the fact that none of the methods proposed in the literature takes the subsequent confirmatory trial and the overall sample size for the program into account when fixing the sample size of the pilot. The planned project intends to develop methods for an integrated planning of pilot and confirmatory study for a broad range of design scenarios. The new methods allow determining the probability of a successful overall program (pilot study + confirmatory study) depending on defined parameter values and sample sizes. Furthermore, utility functions are constructed including the required resources for performing the program as well as the gains in case of a successful program. By maximizing the expected utility, optimal decision rules and optimal sample sizes for pilot studies are derived. Applying the results of this research will allow for a planning of study programs that optimally balances costs and gains.