Supra-states and censoring in health utility modelling: an examination of their importance in the EQ-5D tariff

The methodology used when estimating utilities assumes that the best health state in a health-related quality of life (HrQoL) measure is equivalent to perfect health, with both valued at one. In this paper we explore the impact of relaxing this assumption by treating measured utilities as censored data. We compare Random Effects Tobit (Tobit RE), and Censored Least Absolute Deviation (CLAD) models of the UK EQ-5D TTO data with the recommended Generalised Least Squares Random Effects (GLS RE) model. Both the Tobit RE and CLAD models indicate that value of perfect health is greater than one (1.106, 1.036). The Tobit RE model values three of the mildest EQ-5D states (11112 , 11211, 21111) as greater than one; capturing the large scale non-trading in measured values for the states. The Tobit RE model utility estimates for poor health states are close to the GLE RE estimates. Possible health policy and research responses to the findings are discussed.