Assurance Methods for Adaptive Clinical Trials with a Delayed Treatment Effect

Modern oncology clinical trials increasingly face challenges posed by delayed treatment effects (DTEs), where therapeutic benefits emerge only after an initial delay period. Conventional methods that assume proportional hazards often underestimate these effects, leading to underpowered or inefficient studies. This thesis develops a Bayesian framework that integrates assurance methods, expert elicitation, and adaptive design principles to improve the planning and evaluation of such trials.

The first part establishes assurance as a Bayesian alternative to traditional power calculations, incorporating parameter uncertainty through prior distributions. Structured expert elicitation is used to construct these priors, ensuring that clinical knowledge is captured transparently and quantitatively. The approach is then extended to survival models representing DTEs, allowing uncertainty in both delay duration and post-delay treatment effects to propagate through assurance calculations.

Building on this foundation, the thesis introduces adaptive design strategies, particularly group sequential and predictive approaches, that use elicited priors to inform interim decision rules. These methods enhance trial efficiency and ethical conduct while maintaining statistical validity. All methods are implemented in a freely available R package, {DTEAssurance}, together with two interactive Shiny applications that enable practitioners to design and evaluate complex trials in real time.

Finally, anonymised oncology datasets from the Vivli platform are analysed to characterise empirical patterns of DTEs in modern immunotherapy trials. Together, these developments provide a reproducible and practical framework for integrating expert knowledge and Bayesian reasoning into adaptive survival trial design, advancing the methodological foundations of clinical trial planning and bridging innovation with real-world application.