Study on reseach design and statistical inferences for exploratory clinical researches

我们开发了两阶段和三阶段的设计，用于样本大小估计信息非常稀缺的情况。在这些设计中，在第一阶段之后重新估计样本量。第二阶段包括两个阶段，进行一次中期分析。在此阶段不重新估计样本量。经证明，这些设计的功率约为单级设计的85%至95%，平均样本量远小于单级设计。我们还调查了多个测试问题。当我们希望检验一个由几个基本假设组成的家族时，这些基本假设是根据它们之间的重要性和逻辑关系来组织的。我们得到了每个假设检验的校正p值的显式公式。此外，我们定义了多个平行/固定序列假设结构，该结构由多个有序处理和同等重要的多个端点组成。我们还为基本假设的调整p值开发了显式公式。

We developed two-phase and three-stage designs for a situation where information on sample size estimation was very scarce. In these designs sample size is re-estimated after the first phase. The second phase consists of two stages and one interim analysis is performed. Sample size is not re-estimated at this phase. Theses designs were proved to give about 85 to 95 percent of the power of the single stage design and much less average sample sizes than that design.We also investigated multiple testing problems. When we wish to test a family of several elementary hypotheses which are structured according to the importance and logical relationships among them. We obtained explicit formulae of adjusted p-values of each test of hypothesis. Furthermore, we defined multiple parallel/fixed sequence hypothesis structure, which consists of several ordered treatments and equally important multiple endpoints. We also developed explicit formulae for the adjusted p-values of elementary hypotheses.