Description:
A valid sample size calculation is a key aspect for ethical medical research. While the sample size must be large enough to detect an existing relevant effect with sufficient power, it is at the same time crucial to include as few patients as possible to minimize exposure to study related risks and time to potential market approval. Different parameter assumptions, like the expected effect size and the outcome's variance, are required to calculate the sample size. However, even with high medical knowledge it is often not easy to make reasonable assumptions on these parameters. Published results from the literature may vary or may not be comparable to the current situation. Adaptive designs offer a possible solution to deal with those planning difficulties. At an interim analysis, the standardized treatment effect is estimated and used to adapt the sample size. In the literature, there exists a variety of strategies for recalculating the sample size. However, the definition of performance criteria for those strategies is complex since the second stage sample size is a random variable. It is also known since long that most existing sample size recalculation strategies have major shortcomings, such as a high variability in the recalculated sample size. Within Thesis Article 1, me and my coauthors developed a new performance score for comparing different sample size recalculation rules in a fair and transparent manner. This performance score is the basis to develop improved sample size recalculation strategies in a second step. In Thesis Article 2, me and my supervisor propose smoothing corrections to be combined with existing sample size recalculation rules to reduce the variability. Thesis Article 3 deals with the determination of the second stage sample size as the numerical solution of a constrained optimization problem, which is solved by a new R-package named adoptr. To illustrate the relation of the three Thesis Articles, all new approaches are applied to a clinical trial example to show the methods' benefits ...