Background: Conventionally, models used for health state valuation data have been parametric. Recently, a number of researchers have investigated the use of non-parametric Bayesian methods in this area.
Objectives: In this paper, we present a non-parametric Bayesian model to estimate a preference-based index for a five-dimensional health state classification, namely EQ-5D.
Methods: A sample of 2997 members of the UK general population valued 43 health states selected from a total of 243 health states defined by the EQ-5D using time trade-off technique. Findings from non-parametric modelling are reported in this paper and compared to previously used parametric estimations. The impact of respondent characteristics on health state valuations is also reported.
Results: The non-parametric models were found to be better at predicting scores in populations with different distributions of characteristics than observed in the survey sample. Additionally, non-parametric models were found to be better at allowing for the impact of respondent characteristics to vary by health state. The results show an important age effect with sex having some effect.
Conclusion: The non-parametric Bayesian models provide more realistic and better utility estimates from the EQ-5D than previously used parametric models have done. Furthermore, the model is more flexible in estimating the impact of covariates.
Keywords: Covariates; EQ-5D; Non-parametric Bayesian methods; Preference-based health measure; Time trade-off.