An investigation of time effect in Bayesian adaptive multi-arm multi-stage design and platform trial

Summarise with VIRAT
Authors: Ziyan Wang
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Platform trials continuously evaluate multiple treatments by allowing new arms to join at different stages. The control data collected before a new arm joins is often used to enhance statistical power. However, this approach relies heavily on the exchangeability assumption, frequently violated by systematic response changes over time (time trends). Such trends can bias treatment effect estimates and inflate type I errors, especially under BRAR. This thesis investigates the impact of time trends in adaptive multi-arm multi-stage (MAMS) and platform trial designs, proposing robust analytical methods. I explore scenarios with both equal and unequal strength of time trends across trial arms. Results show equal-strength trends exacerbate bias in BRAR with early stopping rules, motivating the use of flexible models robust to various time patterns. For unequal-strength time trends, existing methods yield biased estimates. Thus, I extend these methods to handle unequal trends, achieving unbiased estimates, albeit with reduced power. Additionally, I generalise estimands to align explicitly with clinical research objectives, emphasising their importance for valid inference. Among evaluated approaches, flexible mixed-effects models consistently provide unbiased treatment effect estimates and maintain statistical power. Finally, I expand adaptive MAMS designs to fully accommodate platform trial complexities, demonstrating robustness through extensive simulation studies. This thesis extends our knowledge of platform trials by addressing the time trend problem via advanced analytical methodologies for managing time trend challenges in platform trials in practice.

Fields of Study & Subjects

Statistical Methods in Clinical Trials Statistical Methods and Bayesian Inference Optimal Experimental Design Methods

Publication details

Journal / source
ePrints Soton (University of Southampton)
Publication date
9 March 2026
Language
EN
Document type
Thesis