Tails from the Peak District: adjusted censored mixture models of EQ-5D health state utility values

Health state utility data generated using the EQ-5D instrument are typically right bounded at one with a substantial gap to the next set of observations, left bounded by some negative value, and are multi modal. These features present challenges to the estimation of the e¤ect of clinical and socioeconomic characteristics on health utilities. We present an adjusted censored model and then use this in a flexible, mixture modelling framework to address these issues. We demonstrate superior performance of this model compared to linear regression and Tobit censored regression using a dataset from repeated observations of patients with rheumatoid arthritis. We �nd that three latent classes are appropriate in estimating EQ-5D from function, pain and sociodemographic factors. Analysis of utility data should apply methods that recognise the distributional features of the data.