Abstract

Objectives

Mental health measures are common in population surveys but cannot be directly used for utility-based economic evaluations. This study explores methods to estimate health utility associated with mental health.

Methods

We used data from 12 701 respondents participating in a Belgian population survey. We compared direct methods (linear and inflated beta regression) that could directly generate utility values, with indirect methods that first estimate EuroQol 5-Dimension 5-Level instrument (EQ-5D-5L) dimension responses using non-parametric or ordered logistic regression before generating utilities. Regression models used 12-item General Health Questionnaire (GHQ-12) responses as independent variables, controlling for sociodemographic factors. Model performance was assessed using root mean-squared error and mean absolute errors. We also calculated error statistics stratified by utility quantiles.

Results

For overall model performance, root mean-squared error values were identical for the linear and ordered logistic models (0.142), higher for the beta model (0.148), and highest for the nonparametric approach (0.155). Ordered logistic models performed the best in mean absolute errors, followed by linear, beta, and nonparametric approaches. Stratified analyses revealed heterogeneity in prediction accuracy across the utility quantiles. Models showed poorer performance in the lowest utility quantiles. Indirect mapping is conceptually more robust because it aligns with the dimensional structure of EQ-5D-5L and minimizes variations associated with the use of different value sets.

Conclusion

This study provides up-to-date algorithms for mapping mental health data to health utility. The mapping models can be applied using GHQ-12 responses and basic sociodemographic variables to generate utility values for use in economic evaluations in which utility data are not available.