Assessing the Direct Impact of Death on Discrete Choice Experiment Utilities.

Abstract

The dead state can affect the value sets derived from discrete choice experiments (DCEs). Our aim was to empirically assess the direct impact of the immediate death state on health utilities using discrete choice experiment with time (DCETTO). A sample of the general population in Quebec, Canada, completed two approaches: DCETTO followed by a best-worst scaling with time (BWSTTO) (hereafter referred to as DCEBWS), versus DCETTO followed by the dominated option and the immediate death state (hereafter referred to as DCEDOD), both designed with the SF-6Dv2. In DCEBWS, all participants first completed 10 DCETTO choices (i.e., option A vs B), followed by 3 BWSTTO. In DCEDOD, the same participants first completed the same 10 DCETTO choices, followed by a repeated choice between the dominated option (i.e., A or B) and the immediate death state. A conditional logit model was used to estimate value sets. The performance of models was assessed using goodness of fit using Bayesian information criterion, parameters' logical consistency, and levels' significance. The direct impact of the death state on DCE latent utilities was evaluated by examining the magnitude of coefficients, assessing the agreement among the value sets estimated by DCETTO with DCEBWS and with DCEDOD using Bland-Altman plots, the proportion of worst-than-dead(WTD) health states, and analyzing the range of estimated values. From 398 participants, a total of 348 participants were included for final analysis. The number of parameters with illogical consistency and non-significant coefficients was lower in DCEBWS. The observed consistency in the relative importance of dimensions across all approaches suggests a stable and reliable ranking. The utility range for DCEDOD (-0.921 to 1) was narrower than for DCETTO (-1.578 to 1) and DCEBWS (-1.150 to 1). The DCEDOD estimated a lower percentage of WTD health states (20.01 %) compared to DCETTO (47.19 %) and DCEBWS (33.73 %). The agreement between DCETTO and DCEBWS was slightlystronger than between DCETTO and DCEDOD, and the mean utility values were higher in DCEDOD than in DCEBWS. The inclusion of the immediate death state directly within DCE increased utility values. This increase was higher when the immediate death was included in a sequence within a DCETTO (i.e., DCEDOD) than when it was included in a continuum of DCETTO (i.e., DCEBWS). The use of DCEDOD was potentially better suited to incorporate the dead state into a DCE.