Eliciting preferences to the EQ-5D-5L health states: discrete choice experiment or multiprofile case of best-worst scaling?

Abstract

Choice-based methods have been used widely in assessing healthcare programs. This study compared the binary discrete choice experiment (DCE) and the multiprofile case of best-worst scaling (BWS) in eliciting preferences for the EQ-5D-5L. Forty-eight EQ-5D-5L health states were selected using a Bayesian efficient design and grouped into 24 pairs for the DCE tasks and 8 sets for the BWS tasks (each set has three health states). A total of 100 participants completed 12 pairs and 8 sets in a random order. A probit regression model and ranked order logistic regression model were used to estimate the latent utilities from the DCE and BWS, respectively. Both tasks were well understood by the majority of participants. The DCE tasks were relatively easier and took a shorter time to complete. The intraclass correlation coefficient (ICC) of the DCE was higher than that of the BWS. The variances associated with the latent utilities estimated from the DCE were larger than those from the BWS. The DCE is more feasible and reliable than the BWS in valuing the EQ-5D-5L. Future studies could focus on comparing the consistency and accuracy of these techniques in predicting the health utilities of the EQ-5D-5L.