Abstract
We develop a Bayesian tree ensemble model to estimate heterogeneous treatmenteffects in censored survival data with high-dimensional covariates. Instead ofimposing sparsity through the tree structure, we place a horseshoe priordirectly on the step heights to achieve adaptive global-local shrinkage. Thisstrategy allows flexible regularisation and reduces noise. We develop areversible jump Gibbs sampler to accommodate the non-conjugate horseshoe priorwithin the tree ensemble framework. We show through extensive simulations thatthe method accurately estimates treatment effects in high-dimensional covariatespaces, at various sparsity levels, and under non-linear treatment effectfunctions. We further illustrate the practical utility of the proposed approachby a re-analysis of pancreatic ductal adenocarcinoma (PDAC) survival data fromThe Cancer Genome Atlas.