Bayesian inference for health state utilities using pairwise comparison data.

The National Institute for Health and Clinical Excellence (NICE) is responsible
for making recommendations about which treatments are available on the NHS.
An important part of the decision making process is to estimate the cost effectiveness of a treatment, measured in cost per QALY gained. If a treatment costs
more than £30000 per QALY the NHS does not consider it to be cost effective.
QALYs are calculated using life years and QALY weights. which represent the
quality of life of a condition. An example of a QALY weight is a utility. which is
a measure of preference for a health condition. A utility is measured on a scale
between 0 and 1, where 0 is the utility of death and 1 is the utility of perfect
health. This thesis uses discrete choice modelling to estimate utilities for health
states defined using the Asthma quality of life questionnaire. A Bayesian approach is used to estimate the utilities in order to quantify utility. A probit and
legit model are considered for the likelihood where the parameters represent the
decrease in utility associated with increasing levels of the attributes of the asthma
quality of life questionnaire. An MCMC is run using three prior distributions on
the parameters: Gamma(l.lO). Gamma(5.15) and Uniform(O. 1). The model is
also extended to include a multiplicative random effect. Bayes factors are used
as a model comparison in the standard model, Results from both the standard
model and random effects model are also compared with maximum likelihood
estimates.