|
Abstract:
|
While cost-effectiveness analyses (CEAs) are crucial for health economic decision making, estimation is challenging for several reasons. Cost and effectiveness are correlated and follow complex joint distributions which cannot be captured parametrically. Effectiveness (typically survival time) and cost both tend to be right-censored. Moreover, CEAs are often conducted using observational data, requiring robust confounding control. Finally, current CEA methods do not address treatment effect heterogeneity in a principled way - opting to either present results marginally or for pre-specified subgroups. Motivated by these challenges, we develop a Bayesian nonparametric approach using an Enriched Dirichlet Process (EDP) prior, which yields a flexible posterior model for the joint cost-survival distribution. We identify causal CEA estimands, conduct inference via posterior g-computation, and leverage the induced posterior clustering of the EDP to detect subpopulations with different cost-effectiveness profiles. We outline a sampling scheme for posterior inference, evaluate frequentist properties, and present an application to endometrial cancer.
|
