Statistical methodology for umbrella and basket clinical trials

Abstract

Umbrella and basket trials evaluate multiple experimental treatments, multiple diseases, or both under a single trial infrastructure and protocol. There is a need to develop efficient statistical methodology for the additional statistical complexities in the design and analysis of umbrella and basket trials, to allow evaluation of targeted treatments in less time and with fewer patients. This thesis addresses this issue by developing and investigating efficient statistical methodology for problems relating to the design and analysis of basket and umbrella trials. I begin with a review in Chapter 2 to identify methods used in the design and analy sis of ongoing and completed umbrella trials, thereby characterising the methodological landscape and gaps in the literature. Thereafter, I focus on developing novel statistical methodology; i) I propose in Chapter 3 different flexible approaches for treatment alloca tion in umbrella trials in the presence of multiple biomarkers; (ii) I propose in Chapter 4 a new two-stage adaptive umbrella design that enables borrowing of information across the control arm, and accommodates adaptation of allocation ratios following the interim analysis; and finally (iii) propose a new Bayesian modelling strategy for randomised basket trials based on a distributional discrepancy in Chapter 5. Throughout, simulations are used to compare the statistical methods and show the validity of the approaches developed. The proposed statistical methodologies show promise of bringing the much-needed effi ciency to the design and analysis of umbrella and basket trials. In particular, this thesis demonstrates that pre-specification of an efficient treatment allocation strategy, and bor rowing information across treatment arms with a commensurate effect in an adaptive design can greatly improve the statistical performance of umbrella trials. Further, I show that a novel methodology, Treatment response borrowing, is particularly suited for the analysis of small randomised basket trials