The consideration of power of a clinical trial to detect the treatment effect is integral in the planning and designing stage of any trial. Recent research however argues that the power of the subsequent evidence synthesis model including the new trial should receive attention in the design stage of any new trial besides the power of the new trial in isolation. This thesis begins with a survey aiming to assess the extent of the use of previous evidence in the design stage of a new trial. The survey concludes that there is a lack of such use of previous evidence, particularly in determining the sample size of the new trial. The findings of the survey set the background and the context of the methodology developed in this thesis. The thesis aims to develop a simulation based methodological framework for designing new trials based on the power of the subsequent evidence synthesis model including the new trial, using several evidence synthesis models. The purpose of this methodology is to offer guidance to researchers in computing the sample size of a new trial on the basis that the new trial will eventually be a part of an evidence synthesis model. The variety of evidence synthesis methods includes standard meta-analysis methods, indirect comparison methods, mixed treatment comparison methods and meta-regression methods. This approach treats the pooled effect size of the initial evidence synthesis model to be the mean of the predictive distribution of the new trial. The predictive distribution of the new trial provides the distribution of an effect size of a new trial that is deemed sufficiently similar to the existing trials to be eligible for the synthesis. Under each evidence synthesis method, we develop various models to design new trials by varying the variance of the predictive distribution. The power of the new trial is shown to increase with increased variance in the predictive distribution. In contrast, the power of the updated evidence synthesis is shown to decrease with the increased variance of the predictive distribution. The new trials designed using fixed effects principles yield the lowest power. However, the updated evidence synthesis model including the new trial designed using the fixed effects principles shows the highest power. This is a common phenomenon noticed in all evidence synthesis methods explored. The framework also includes a component that develops a methodology to design new trials using Bayesian meta-analysis principles. The reasons for the differences found in power results of the Bayesian and frequentist approaches are investigated. The variance of the predictive distribution of a new trial clearly influences both the power of the new trial and the updated evidence synthesis. Trialists are advised to adapt the fixed effects method in designing new trials, whenever the assumption of a common true effect is possible. The Bayesian meta-analysis method developed here does not produce sufficient power in the updated meta-analysis and the methodology requires further refinements.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:617662 |
| Date | January 2014 |
| Creators | Mendis, Agampodi Shanthi Jeewaka Ruwan |
| Contributors | Sutton, Alexander; Jones, David |
| Publisher | University of Leicester |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://hdl.handle.net/2381/29056 |
Page generated in 0.0017 seconds