CMStatistics 2016: Start Registration
View Submission - CMStatistics
B1546
Title: Some statistical aspects of clinical trials for personalized medicine Authors:  Tim Friede - University Medical Center Goettingen (Germany) [presenting]
Marius Placzek - University Medical Center Goettingen (Germany)
Abstract: In personalized medicine patient populations are stratified with a view to improve treatment outcome in terms of efficacy and tolerability. Stratification is often carried out by biomarkers. A particular case are nested subgroups that might arise from using several thresholds of a continuous marker. Efficient testing strategies for normal data are derived under homoscedasticity and heteroscedasticity assumptions across the subgroups. Furthermore, procedures for sizing a study with several nested subgroups are presented. These depend among other quantities on nuisance parameters such as the variances of the outcomes in the subgroups and the prevalences of the subgroups. Knowledge of these might be very scarce in the planning phase of such a trial resulting in a considerable risk of choosing an inappropriate sample size. To mitigate these risks an internal pilot study design is proposed and its properties including type I error rate, power and sample size distribution are explored in Monte Carlo simulations. Adaptive enrichment designs allow to restrict recruitment to certain subgroups following interim analyses. Approaches to hypothesis testing in such designs are reviewed and their properties are compared in a simulation study. Finally, an adaptive enrichment design including an internal pilot study is presented and its characteristics are discussed.