All Times EDT
Keywords: Bayesian three-outcome decision-making; Proof-of-concept study; Exchangeability-nonexchangeability model; Basket trials
With the recent accelerated approvals of four histology-agnostic novel agents (pembrolizumab in 2017, larotrectinib in 2018, entrectinib in 2019, and dabrafenib plus trametinib in 2022), we have seen an increasing trend of conducting basket trials that evaluate a new investigational therapy on different histologies. Various statistical models have been proposed that aim to effectively leverage information across cohorts to boost statistical efficiency (relative to basket-wise analyses). However, there is a surprising lack of research to establish a quantitative decision-making framework for multiple interims and final analyses in this basket trial context. Considering the limited sample size in early oncology trials, the robustness of a dichotomized go/no-go decision may be suboptimal. In many cases where an “inconclusive” decision is more appropriate based on results solely from the primary endpoint, evidence from other clinical measures would contribute to the final decision-making. For this reason, the three-outcome decision making (3ODM) framework with an additional “consider” zone has gained much popularity in early proof-of-concept studies. We propose to incorporate the 3ODM framework to the basket trial setting for robustness and flexibility. Our comprehensive simulations study is also the first to compare between the modeling of the log odds ratio (on top of the benchmark response rates, RRs) and the modeling of the RRs in most current basket trial designs, considering the potential drastic differences for the reference and target RRs across baskets. While our proposed enhancement could be readily extended to different basket trial methods, we take the exchangeability-nonexchangeability (EXNEX) model as an example to illustrate the EXNEX+3ODM framework (with and without an interim analysis) and evaluate its operating characteristics via simulations.