Use of Bayesian mthods for the design, analysis and synthesis of clinical trials

Burke, Danielle Lisa (2015). Use of Bayesian mthods for the design, analysis and synthesis of clinical trials. University of Birmingham. Ph.D.

Preview

Burke15PhD.pdf
PDF - Accepted Version
Download (2MB)

Abstract

This thesis explores Bayesian methods for the statistical design, analysis and synthesis of clinical trials, and compares these with a frequentist approach in several settings to determine key differences, advantages and limitations.

A review of randomised trials indicates that Bayesian methods are rarely applied, but useful for making probability statements and incorporating prior evidence, especially in trials with small sample sizes. These advantages are illustrated in a trial in congenital lower urinary tract obstruction, which has few events but elicited prior distributions about treatment effect.

Bayesian methods are then developed for meta-analysis of phase II trials and multiple outcomes, with an emphasis on informing phase Ill trial decisions. A Bayesian random-effects logistic regression is advocated for meta-analysis of a binary outcome to account for all parameter uncertainty, and to derive prediction intervals for the treatment effect in a new phase III trial. Bayesian multivariate meta-analysis methods arc then encouraged to make joint inferences across multiple outcomes and incorporate prior distributions for missing correlations. However, a simulation study identifies that external evidence or clinical guidance is needed to ensure appropriate prior distributions for between-study variances and correlations to avoid misleading results.

Researchers should thus consider Bayesian methods for clinical trials, but recognise potential difficulties when adopting the approach in practice.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):