ORACLE II – Optimal Rules for Adaptive Designs with reCalculation of sampLE size

点击翻译按钮获取中文摘要

In standard clinical trial designs, all design parameters must be fixed during the planning stage. Thereby, in particular the choice of the sample size is a crucial issue. It is based on the assumed treatment effect size, the chosen level of significance, the desired power, and the distribution of the applied test statistic. However, the common uncertainty on the true underlying effect size in the planning stage of a trial may result in wrong planning assumptions and, therefore, imply misspecification of the sample size. Adaptive designs provide the possibility to modify the trial design during the ongoing trial and allow correcting for wrong planning assumptions. At the interim analysis, it is decided whether the trial should be stopped for futility or efficacy, respectively, or whether the trial enters the second stage. In the latter case, many design parameters can be adapted, in particular the required sample size. Thereby, the second stage sample size is usually determined by predefined rules and is usually provided as a function of the interim result.The question of a sensible choice of the specific adaptive trial design arises intuitively. There are several challenges related to adaptive designs. The overall aim of this joint proposal between Heidelberg and Berlin is to continue the successful work of the previous DFG-funded project ORACLE. Within the ongoing ORACLE project, recalculation rules were improved and optimized with respect to different optimization and performance criteria. Within the new ORACLE II project, the results and methods of the previous project should be further improved an extended, in particular with respect to non-normally distributed outcomes, designs with more than one interim analysis, and designs with more than two groups. Whereas the ORACLE project exclusively focused on the aspect of sample size in adaptive designs, the focus of the new project ORACLE II is broader: An adequate sample size recalculation rule only has a relevant impact for practical applications if other challenges related to adaptive designs are addressed simultaneously. The most prominent requirements are adequate, unbiased treatment effect estimators and confidence intervals along with corresponding p-values. The above objectives will be addressed within the following work packages and time frames:• Work package 1 (Month 1-12, Berlin): Assessing performance of sample size recalculation • Work package 2 (Months 1-12, Heidelberg): Point estimators, confidence intervals, p-values for adaptive designs with sample size adaptions• Work package 3 (Months 13-24, Berlin & Heidelberg): Sample size recalculation in designs with more than two arms or stages• Work package 4 (Months 25-36, Berlin & Heidelberg): Extensions to binary and time-to-event endpoints