Merging Adolescent and Adult Preference Data Into a Potential Latent Scale EQ-5D-Y-3L Value Set: A Latent Class Approach to Analyze Contribution of Each Group to the Preference Weights

OBJECTIVES: Stakeholders of different jurisdictions are advising combining adolescent and adult preferences into EQ-5D-Y-3L value sets, to acknowledge children informing their societal values. This study aimed to quantify the specific contribution of adolescent and adult respondent preferences, to a single, combined value set.

METHODS: Discrete choice experiment (DCE) data were utilized from the US valuation of the EQ-5D-Y-3L, which included an online sample of approximately 1,500 adults and 700 adolescents (11-17). Each respondent completed 15 DCE tasks. DCE responses were analyzed by fitting a range of latent class models (2-7 classes). Within the best-fitting model, the contribution of each class to estimate value set coefficients was determined by the ‘scale-adjusted class share’ (SACS): a metric combining the class’s proportion of respondents (class share) adjusted for the within-class scale, such that more indifferent preferences (lower scale) weighted less. The contribution of adolescent vs adult respondents was assessed by the proportion of respondent within class in each group corrected by the scale attributable to each group.

RESULTS: The analytic DCE dataset consisted of 714 adolescents and 1,529 adults. The latent class model with 6 classes was selected as the best fitting model. Weighting class-share by within-class scales resulted in SACS of: Class 4 (54.5%), Class 6 (22.0%), Class 3 (9.7%), Class 2 (7.4%), Class 1 (3.6%), and Class 5 (2.9%). Adults contributed more highly to SACS of all classes except Class 1, accounting for 78.7% of the total value set contribution compared to 21.3% for adolescents. Assuming equal sample size the contribution will be 35% and 65% for adolescent and adults respectively. To reach a 50% contribution per group the sample size for adolescent would need to be 4,1 time the one for adults.

CONCLUSIONS: Adolescents tended to contribute less informative to a potential value set. This should be considered when designing EQ-5D-Y valuation studies.