Valuing EQ-5D-5L using discrete choice experiments with duration: Investigating dimension order

Objectives Discrete Choice Experiments with duration (DCETTO) used to value EQ-5D-5L usually present the health state dimensions in a ‘standard’ order (e.g. Mobility-Self care-Usual Activities-Pain/discomfort-Anxiety/depression then duration). Given respondents may use heuristics to complete the task, dimension order may impact the valuation process and the estimated utility values. Our aim was to assess the importance of dimension order in the use of DCETTO to value EQ-5D-5L. Methods The same 120 choice sets were administered across three arms to an online sample of 1,080 Australians (completing 15 each). Arm 1 presented the ‘standard’ order, Arm 2 randomised the 120 possible EQ-5D-5L orders between respondents, and Arm 3 randomised within respondents. Data were modelled using conditional logit regression. Across the arms we compared the magnitude and consistency of the dimension level coefficients and the value set characteristics, and the impact of dimension order allowing for heterogeneity. Results The consistency and stability of the models across each arm was similar. The level 2/3 usual activities coefficients were disordered (across all arms) and arms 1 and 2 had two other inconsistencies. Mobility and pain/discomfort had the overall largest decrements, and usual activities the smallest. Arm 3 produced the largest value range (1 to -0.98). Cognitive follow up questions did not suggest that one arm was more difficult, and completion strategies reported did not differ. Conclusions The similarity in the models suggests that the order in which dimensions are presented does not have a major impact on the value set characteristics. This suggests that the dimension order may not be as important as other design issues such as the design of the duration attribute, the health state selection procedure or task presentation method. However, it is worth considering whether between subject randomisation should be used (our Arm 2), as this would minimise any potential bias.