Utility Score Mapping From DASS-21 to EQ-5D-5L, SF-6Dv2, and CORE-6D

Objective: To develop algorithms mapping the Depression Anxiety Stress Scale-21 (DASS-21) onto the EuroQol 5-Dimension 5-level (EQ-5D-5L), the Short-Form 6-Dimension version 2 (SF-6Dv2), and the Clinical Outcomes in Routine Evaluation 6-Dimension (CORE-6D). Methods: An online survey was conducted to collect responses to DASS-21, EQ-5D-5L, SF-6Dv2, and CORE-6D from the general adult population in Quebec, Canada (N=6,414). Ordinary least squares (OLS) and beta-mixture models for direct mapping and the ordered probit (OPROBIT) for response mapping were used. Using a cross-validation, predictive performance was assessed by mean absolute error (MAE), root mean square error (RMSE), and absolute error (AE). Furthermore, Bland-Altman plots were used to assess the distribution of observed versus predicted utilities. We also simulated utilities using the estimated models to predict unobserved individual-level utilities and to compare the resulting distributions with the observed data, thereby supporting robust and realistic applications in economic evaluations. Results: The beta-mix models were the best-performing algorithms for EQ-5D-5L (MAE=0.0950, RMSE=0.1521, AE>0.10=28.22%) and CORE-6D (MAE=0.1000, RMSE=0.1445, AE>0.10=32.71%), while the response mapping was the best model for SF-6Dv2 (MAE=0.1240, RMSE=0.1681, AE>0.10=47.40%). Models tended to underestimate utilities in mild to moderately severe health states across all 3 instruments, while they overestimated utilities in more severe health states for the EQ-5D-5L and SF-6Dv2. The simulated distributions from the beta-mixture models closely resembled the observed data, although differences remained. Conclusion: Mapping models were developed to estimate EQ-5D-5L, SF-6Dv2, and CORE-6D from DASS-21. Linear regression could not reproduce the observed utility distribution, supporting the use of more flexible specifications.