The effect of health state selection on the valuation of EQ-5D

Many aspects of valuation

s are poorly understood, and the size and composition of health states used in valuation studies vary widely. Our aims were to investigate the impact of the selection of a subset of EQ-5D health states in generating a set of visual analog scale (VAS)-based values. Our aims were to investigate the optimal number of health states, the sample size required per health state, and which combinations of health states are required to establish a EQ-5D VAS value set.

Data were drawn from a United Kingdom (UK) general population postal survey in which all health states defined by EQ-5D were evaluated using VAS. We used a simulation approach to address each question, and the performance of each estimation model was assessed by the average value of the mean absolute errors.

Despite the constraint of the small sample size in the study dataset, the results suggest that the optimal number of states for a valuation study is 32, 100 observations per health state were sufficient, and multiple subsets of health states were feasible.

Selecting health states not included in the "Measurement and Valuation of Health" dataset in valuation studies is quite specifically encouraged. However, it remains prudent to ensure that the selection of health states covers as wide a range of severity as possible. Setting the number of observations per health state to 150 might allow for correction of any errors in data collection or processing. The extent to which the results of this study based on VAS valuation data can be safely generalized to the design of time tradeoff valuation studies remains an open question.