CMStatistics 2021: Start Registration
View Submission - CMStatistics
B1166
Title: Bayesian causal forest with AFT model: Estimating heterogeneous treatment effects on a survival outcome Authors:  Rongqian Sun - The Chinese University of Hong Kong (China) [presenting]
Xinyuan Song - Chinese University of Hong Kong (Hong Kong)
Abstract: Estimating heterogeneous treatment effects has drawn increasing attention in medical studies, considering that patients with divergent features can undergo a different progression of disease even with identical treatment. We consider a joint framework of Bayesian causal forest (BCF) and accelerated failure time (AFT) model to directly capture the possibly heterogeneous treatment effect through two separate Bayesian additive regression trees (BART). The nonparametric BCF structure controls the regularization imposed on treatment effect and flexibly reflects the complex relationship between pre-treatment covariates, treatment indicator, and survival time while requiring no prespecified functional forms. The AFT model is used to derive the conditional average and sample average treatment effect on the scale of log survival time under the potential outcomes framework. Bayesian backfitting Markov chain Monte Carlo algorithm with blocked Gibbs sampler is conducted for estimation of the causal effects. Simulation studies show the satisfactory performance of the proposed method, especially under a small sample size. The proposed model is then applied to a clinical trial that compares two therapies for HIV-infected patients to demonstrate its usage in detecting and visualizing heterogeneous treatment effects.