Eliciting expert opinion for cost-effectiveness analysis: a flexible family of prior distributions

Abstract

The Bayesian approach to statistics has been growing rapidly in popularity as an alternative to the classical approach in the economic evaluation of health technologies, due to the significant benefits it affords. One of the most important advantages of Bayesian methods is their incorporation of prior information. Thus, use is made of a greater amount of information, and so stronger results are obtained than with frequentist methods. However, since Stevens and O'Hagan (2002) showed that the elicitation of a prior distribution on the parameters of interest plays a crucial role in a Bayesian cost-effectiveness analysis, relatively few papers have addressed this issue. In a cost-effectiveness analysis, the parameters of interest are the mean efficacy and mean cost of each treatment. The most common prior structure for these two parameters is the bivariate normal structure. In this paper, we study the use of a more general (and flexible) family of prior distributions for the parameters. In particular, we assume that the conditional densities of the parameters are all normal. The model is validated using data of a real clinical trial. The posterior distributions have been simulated using Markov Chain Monte Carlo techniques.