Title: Estimating Medication Prescription Policy Tree Using Causal Machine Learning

Abstract: Metastatic castrate-resistant prostate cancer is an advanced stage of prostate cancer where the body has stopped responding to hormone treatment, making it complex to treat. Androgen-receptor targeting agents (ARTAs) are preferred therapies over others for their less toxic profile and improved survival. No clinical trials have been conducted comparing the two ARTAs. Also, no existing study has used patient characteristics and multiple comorbid conditions to identify heterogeneity and subpopulations that would benefit from either medication. We leverage the recent advances in machine learning for causal inference. We use a non-parametric causal survival forest to identify heterogeneous sub-populations and estimate an optimal policy learning tree. Our research has implications as the learned policy tree would aid as a decision tool for physicians.