Methods for overcoming barriers to using adaptive designs in Low and Middle-Income Countries

Abstract

Research indicates that populations in Low and Middle-Income Countries (LMICs) could benefit from clinical trials due to their high disease burden. Yet, they are underrepresented in clinical research. This is because the advancement of clinical research in LMICs is hindered by several obstacles, among which insufficient funding has been identified as a key barrier. One notable innovation, that holds promise for increasing the number of trials conducted in LMICs, is the utilisation of adaptive designs as they can increase the value derived from limited resources. Considering that adaptive designs are not always beneficial, the methods presented in this thesis determine instances where adaptations could provide greater utility within the context of individually (IRTs) and cluster randomised trials (CRTs) with continuous outcomes. The thesis begins by proposing a seamless MAMS framework where a cheaper (intermediate) endpoint is used at an interim analysis. With adaptive designs heavily reliant on the quality and quantity of available information at interim analyses for decision-making, this study proposes an optimal timing for conducting the interim analysis to avert wrongly dropping a relevant arm at the interim analysis. To offer a contextual foundation for the subsequent chapters on CRT methodologies, the thesis next presents the results of a comprehensive review undertaken to explore various strategies for specifying the intra-cluster correlation coefficient (ICC) and other essential sample size parameters. The review highlighted that many HTA trials neither reported the uncertainty around the assumed ICC nor justified their selected values, which could affect the trial’s validity. Based on the review, I then evaluated how uncertainty in the ICC impacts whether a parallel-group or stepped-wedge CRT design is more efficient in terms of the required sample size. Here, the uncertainty was captured by placing independent priors on key parameters and averaging over the possible range of values. The results indicated that in many cases, when there is uncertainty regarding the ICC, a steppedwedge CRT design tends to be more efficient compared to a parallel-group CRT design. A limitation of the above approach is that its utility can be highly dependent on the choice of prior. Thus, I next introduce an adaptive design where pre-trial knowledge about the ICC is captured by placing a prior on it, which is then updated at an interim analysis using the study data to form a posterior that allows reestimation of the sample size. It was clearly demonstrated in the results that when there is low-quality evidence available to guide the choice of prior, this approach to sample size reestimation provides greater utility than previously proposed frequentist methods. Results show the proposed methodologies are both robust and cost-effective. Therefore, I conclude by discussing how the adoption of these methods could enhance LMICs’ capacities to conduct high-quality research.