Abstract

Background: Typically, models that were used for health state valuation data have been parametric. Recently, many researchers have explored the use of non-parametric Bayesian methods in this field.

Objectives: In the present paper we report on the results from using a nonparametric model to predict a Bayesian SF-6D health state valuation algorithm along with estimating the effect of the individual characteristics on health state valuations.

Methods: A sample of 249 states defined by the SF-6D have been valued by a representative sample of 577 members of the Lebanese general population, using standard gamble. Results from applying the nonparametric model were reported and compared to the original model estimated using a conventional parametric random effects model. The covariates’ effect on health state valuations was also reported.

Results: The nonparametric Bayesian model was found to perform better than the parametric model 1) at predicting health state values within the full estimation data and in an out-of-sample validation in terms of mean predictions, root mean squared error and the patterns of standardized residuals, and 2) at allowing for the covariates’ effect to vary by health state. The findings also suggest an important age effect with sex, having some effect, but the remaining covariates having no discernible effect.

Conclusion: The nonparametric Bayesian model is a powerful technique for analyzing health state valuation data and is argued to be theoretically more flexible and produces better utility predictions from the SF-6D than previously used classical parametric model. In addition, the Bayesian model is more appropriate to account the covariates’ effect. Further research is encouraged.