A new basket trial design based on clustering of homogeneous subpopulations

Abstract

ABSTRACT As new findings in oncology suggest a focus on individualized and targeted therapies, the demand for adequate clinical trial designs rises, whereby the focus is mainly on early development phases (phase I and II). Phase II oncology trials are often planned and analysed by Simon two-stage design, which corresponds to a one-armed trial design with the option to stop early for futility. Whereas a classical phase II study focuses on one tumour type and location, the relatively new basket trial design allows testing the efficacy of a single drug simultaneously in a number of patient subsets, which correspond to different tumour types. Such trials can be analysed in various ways, including separate analyses of all baskets or by pooling across all baskets. The work presented here tries to find an adequate compromise between these two extremes by implying rules for clustering some baskets, which are reasonably homogeneous. By means of Monte-Carlo simulations, we compare the efficiency of our proposed cluster-based basket trial design with a standard approach proposed recently which only allows for complete pooling or separate analyses. The results suggest that our new design offers a considerable advantage in power, sensitivity and specificity as well as in average sample size compared to the standard approach. The proposed clustering design is an attractive option to conduct basket trials in oncology with higher efficiency and better performance.