Bayesian Models with Spatial Correlation Improve the Precision of EQ-5D-5L Value Sets

Area of Science:

• Health Economics
• Psychometrics
• Statistical Modeling

Background:

• Health utilities derived from EQ-5D-5L value sets are crucial for economic evaluations.
• Existing models may lack precision in predicting health state utilities.

Purpose of the Study:

• To assess if modeling spatial correlation improves the precision of EQ-5D-5L value sets.
• To compare Bayesian models with spatial correlation against existing linear and CALE models.

Main Methods:

• Utilized data from 7 EQ-5D-5L valuation studies.
• Compared predictive precision using root mean squared error (RMSE) for out-of-sample predictions.
• Evaluated models on omitting individual and blocks of health states.

Main Results:

• Bayesian models with spatial correlation consistently improved precision over the linear model across 7 countries when omitting single states.
• RMSE reductions were observed in Canada, China, Germany, Indonesia, Japan, and Korea.
• Performance varied when omitting blocks of states, with CALE models outperforming in 4 countries.

Conclusions:

• Bayesian models with spatial correlation and CALE models show promise for enhancing EQ-5D-5L value set precision.
• Study design, including the number of health states captured, can further improve precision.
• Recommends considering Bayesian and CALE models for future value set development.