Why Bayesian Nonparametric Models Should Be a First-Line Approach for Health State Valuations

Health state valuation is fundamental to the calculation of quality-adjusted life years and, consequently, to cost-effectiveness analysis. Parametric models have traditionally been used to estimate health state values; however, their reliance on prespecified functional forms may limit their ability to capture the complex characteristics of preference-based utility data, including nonlinearity, heterogeneity, skewness, and sparsity. In the present article, we argue that Bayesian nonparametric (BNP) methods should be considered a preferred or first-line approach for modeling health state valuations, particularly when data are limited or the underlying functional form is uncertain. Drawing on more than two decades of methodological and applied research, we discuss their flexibility in representing health state preferences, predictive performance, probabilistic quantification of uncertainty, and ability to borrow strength from existing valuation studies. These features provide a strong rationale for wider adoption and systematic evaluation of BNP methods, particularly in challenging valuation settings.