MAP-EM Polya Trees for Survival Analysis with Complex Censoring

Conference

2026 SSC Annual Meeting in Hamilton

Title

MAP-EM Polya Trees for Survival Analysis with Complex Censoring

Author

Yixing Zhao, Liqun Diao

Abstract

Data from medical studies and clinical trials are often subject to a mixture of censoring mechanisms, but few methods can handle all of them in a single framework. We address this problem by modeling event times using a Polya tree prior. This approach can handle right-censoring, current-status data, and general interval-censoring by directly updating the posterior distribution from censored observations. Our method is built around a unified maximum a posteriori expectation-maximization (MAP-EM) algorithm. The algorithm uses closed-form updates, which makes it stable and efficient across different censoring settings. We establish the asymptotic properties of the estimator and perform extensive simulation studies to evaluate its performance in finite samples. Across a wide range of scenarios, the proposed method achieves superior accuracy compared to benchmark approaches, including the optional Polya tree, fiducial methods, and the nonparametric maximum likelihood estimator (NPMLE).