RANDOMISATION METHODS FOR ADAPTIVE AND MULTI-ARM CLINICAL TRIALS

Abstract

Randomised control trials (RCTs) typically compare one experimental treatment to a control. However, over recent years, with the growing availability of many treatments for evaluation and the increasing complexity of determining which are promising, a growing need has emerged for more efficient trial designs. Accordingly, adaptive designs have gained popularity in clinical research, including multi-arm multi-stage designs and platform trial designs. Such designs aim to improve the clinical trial process by allowing the removal and/or addition of treatment arms during the course of the trial. In almost all clinical trial designs, randomisation remains a fundamental principle to ensure unbiased treatment comparisons. This is no less true of adaptive designs, yet randomisation routines have received less attention for such studies compared to historical fixed sample designs. Therefore, throughout this thesis, the key considerations discussed include the importance of proper randomisation approaches and allocation ratios to achieve clinical trial objectives in adaptive trials. First, I compare different randomisation approaches in the context of multi-arm trials to achieve various trial objectives, such as group size balance, covariate balance, effect precision, low allocation predictability, and high power. Next, two adaptive clinical trial designs are considered: multi-arm multi-stage designs build on multi-arm designs by incorporating interim analyses with stopping rules. In particular, if an experimental treatment shows poor performance, it can be dropped early. In this design, allocation ratios can be fixed throughout the trial, flexible (adjusted based on observed interim data), or pre-specified but variable across the study stages. Finally, attention shifts to randomisation methods in a platform trial design, to which new treatments can be added or dropped over time, and allocation ratios might need to be modified to achieve the study objective(s). Here, the focus is on randomisation approaches with different allocation ratios, aiming to achieve high covariate balance, especially for a newly added arm in the trial.