Basket Weaving: Can Bayes Help Us Develop a Better Path for Evaluating Efficacy in Basket Trials?

PURPOSE: As drug development efforts in oncology focus on targeting increasingly rare driver mutations, new challenges for evaluating treatment efficacy are introduced. The rarity of some driver mutations presents difficulty in recruiting sufficient numbers of clinical trial patients without widening eligibility. Basket trials are increasingly being used to evaluate treatment responses for patients with a variety of different tumour histologies that share a common targetable mutation. Evaluation of outcomes for these “histology-independent therapies” in basket trial settings presents a challenge as outcomes might be heterogeneous across histologies. As these studies are usually single-arm trials, evaluating the comparative effectiveness of the treatment under study requires indirect comparisons. An overview of the challenges of evaluating both non-comparative and comparative efficacy in basket trials will be provided. Discussants will outline how Bayesian hierarchical modelling approaches have been applied to address these challenges and how they may play a role in HTA decision-making.

DESCRIPTION: Bart Heeg will introduce the topic and speakers (10 minutes). Samantha Wilkinson will provide some background on the rationale and growing use of basket trials in oncology, and outline challenges faced by pharmaceutical companies (15 minutes). Sofia Dias will illustrate some of the challenges with evaluating efficacy in basket trials, discussing (i) the pros and cons of analysis approaches which ignore cross-histology heterogeneity by pooling together the data for multiple histologies versus no-pooling approaches, and (ii) the difficulty of performing unbiased indirect treatment comparisons (15 minutes). Emma Mackay will discuss how Bayesian hierarchical modelling approaches can improve upon complete pooling and no-pooling approaches, and can be used to mitigate confounding in indirect treatment comparisons while preserving limited precision/power due to small sample sizes (15 minutes). We will utilize a live smartphone-based polling feature during the workshop to foster active participation and make the learning process more engaging.