All Times EDT
Keywords: basket trial, hierarchical Bayesian model, EXNEX, multisource exchangeability model, heterogeneity, information borrowing
Subgroup-specific analysis (SS) is the conventional approach in basket trials that assess the efficacy of a new agent across multiple histological subtypes in one trial. The notable power gain is expected if one assumes homogeneity of response rates in each subtype and borrows information across subtypes by using a hierarchical Bayesian model (HBM). However, the power gain is seriously lost (Freidlin and Korn, 2013) when "subgroup-wise" (type-I) error rate (SWER) needs to be controlled. This is because one must consider a situation where responses in each subtype were heterogeneous. To better meet a balance between potential homogeneity and heterogeneity, there have been proposed several alternative methods: EXNEX model (Neuenschwander et al., 2016) and multisource exchangeability model (MEM) (Hobbs and Landin, 2018). In this study, via numerical simulation with a control of SWER, we will report a power gain with HBM and EXNEX compared to MEM and SS. We will also report the performance of clustering-based method which assesses a response rate in an identified potential responsive subset. All simulation studies are motivated by an actual basket trial in oncology.