Medical decisions such as benefit-risk assessments of treatments should be based on the best
clinical evidence but also require subjective value judgements regarding the impact of disease and
treatment outcomes. This thesis argues for a Bayesian implementation of Multi-Criteria Decision
Analysis (MCDA) for such problems. It seeks to establish whether suitable Bayesian models can be
constructed given the variety of data formats and the interdependencies between the many
variables involved.
A modelling framework is developed for joint multivariate Bayesian inference of treatment effects
and preference values based on data from clinical trials and stated preference studies. This method
allows the sampling uncertainty of the parameters to be reflected in the analysis, overcoming a
recognised shortcoming of MCDA. Markov Chain Monte Carlo simulation is used to derive the
posterior distributions. The models are illustrated using a case study involving treatments for
relapsing remitting multiple sclerosis.
The clinical evidence synthesis has several advantages over existing multivariate evidence synthesis
models, including a comprehensive flexible allowance for correlations, compatibility with any
number of treatments and outcomes, and the ability to estimate unreported treatment-outcome
combinations.
The preference models can analyse data from a variety of elicitation methods such as discrete
choice, Analytic Hierarchy Process and swing weighting. In the case of swing weighting no Bayesian
analysis has previously been presented, and the results suggest a possible flaw in the standard
deterministic analysis that may bias the preference estimates when judgements are subject to
random variability. A novel meta-analysis model for preference elicitation studies is also presented.
The framework has the unique ability to analyse data from multiple methods jointly to yield a
common set of preference parameters.
These results demonstrate the flexibility of the Bayesian approach, and the depth of insight it can
provide into the impact of uncertainty and heterogeneity in multi-criteria medical decisions.