Reporting Uncertainty Around Health State Values: A Standard Method and Worked Example.

Abstract

Objectives: Papers reporting value sets typically only report the standard errors (SEs) around each estimated coefficient in value set models. This is important information but does not help those building cost effectiveness models, who need to know the uncertainty around the values of health states in order to conduct sensitivity analyses. This paper's aim is to demonstrate how SEs around HRQoL values can be calculated, using the example of the UK EQ-5D-3L value set.

Methods: We show how information from a model's variance/covariance matrix can be used to estimate SEs for every health state value, whether it is part of the modelling data set or not. Data from the Measurement and Valuation of Health study were used to replicate the original UK value set and the variance/covariance matrix, and to produce standard errors around the values for all 243 EQ-5D-3L states.

Results: The range of the SEs is small compared to the range of the health state values, but are conditional on a correct model specification and may be sensitive to alternative specifications.

Conclusions: Reporting these SEs should become routine practice in reporting value sets, to ensure users are provided with information of parameter uncertainty. These SEs only capture one specific aspect of the sources of uncertainty around HRQoL values, but represent a first step toward a more complete account of uncertainty in the preference weights used to estimate QALYs.