To explore the acceptability, consistency, and accuracy of eliciting health state utility values using discrete choice experiment (DCE) and DCE with life duration dimension (DCE ) as compared with conventional time trade-off (TTO) by using the SF-6Dv2.
During face-to-face interviews, a representative sample of the general population in Tianjin, China, completed 8 TTO tasks and 10 DCE/DCE tasks, with the order of TTO and DCE/DCE being randomized. The fixed-effect model and conditional logit models were used for TTO and DCEs data estimation, respectively. Acceptability was assessed by self-reported difficulties in understanding/answering. Consistency was observed by the monotonicity of model coefficients. Accuracy was evaluated by investigating differences between observed and predicted TTO values using intraclass correlation coefficient, mean absolute difference, and root mean square difference.
A total of 503 respondents (53.7% male; range, 18-86 years) were included, with comparable characteristics between respondents who completed DCE (N = 252) and DCE (N = 251). No significant difference was observed in self-reported difficulties among 3 approaches. The monotonicity of coefficients could not be achieved for 2 DCE approaches, even when combining the inconsistent levels. The health state utility values generated by DCE were generally higher than those generated by TTO, whereas DCE was lower than TTO. The TTO had a better prediction accuracy than the DCEs.
Two DCE approaches are feasible for eliciting health state utility values; however, they are not considered to be easier to understand/answer than TTO. There are systematic differences in the health state utility values generated by 3 approaches. The issue of non-monotonicity from 2 DCE approaches remains a concern.