Enabling QALY Estimation in Mental Health Interventions: Mapping the Health of the Nation Outcome Scales onto the EQ-5D-5L

Background: Mental health intervention studies often lack preference-based measures required to estimate quality-adjusted life-years (QALYs) for health economic evaluations. In such circumstances, a mapping study is the second-best alternative to estimate QALYs. This study aimed to develop a mapping algorithm to predict EQ-5D-5L value sets from the Health of the Nation Outcome Scales (HoNOS) among adults with severe mental illness (SMI). Methods: Trial data from a community mental health intervention conducted in Germany between 2020 and 2023 were assessed over 24 mo. The data included more than 900 adults aged 18 to 82 y living with SMI. Four econometric approaches were used to map HoNOS to EQ-5D-5L: ordinary least square (OLS), generalised linear regression model (GLM), fractional regression model (FRM), and an adjusted limited dependent variable mixture model (ALDVMM). The German EQ-5D-5L value set was used, while keeping the Norwegian value set in the appendix . Model performance was assessed using the root mean squared error (RMSE), mean absolute error (MAE), and the squared correlation between observed and predicted EQ-5D-5L ( r 2 ). A 10-fold cross-validation was applied to validate our mapping algorithms. Results: The FRM consistently performed best across all evaluation criteria in both the full sample and cross-validation. In cross-validation, when HoNOS items were used as predictors of the German value set, the FRM yielded RMSE = 0.2204, MAE = 0.1622, and r ² = 36.5%. Similar results were observed when HoNOS subscales were used, with slightly higher RMSE (0.2245) and MAE (0.1664) and a lower r ² (34.2%). Calibration plots indicated good model fit, with predictions closely aligning with the reference line representing equality between the observed and predicted values. Scatter plots further supported the superior performance of the FRM. A similar pattern was observed for the Norwegian value set. Conclusions: The preferred mapping algorithm enables the estimation of EQ-5D-5L utilities from HoNOS data, facilitating the calculation of QALYs when only HoNOS information is available. As our sample includes a broad spectrum of mental health patients, further validation within specific diagnostic groups is recommended to improve predictive accuracy.