Flexible methods improve causal analysis for censored survival data
Flexible Doubly Robust Proximal Causal Inference for Censored Outcomes
Machine Learning
Summary
Understanding the effect of a treatment on survival time is hard when some data is missing or confusing. The authors developed a new flexible approach called Proximal Survival Bridge learning that uses special tricks to handle missing information and data where patients’ survival times are cut off early (right-censoring). Their method improves accuracy and stability compared to older rigid models by using advanced techniques like kernel estimators and neural networks. They showed how these new tools perform well on both simulated and real-world medical data.
What this means in practice
- •For medical data scientists: Analyze treatment effects in clinical survival studies with incomplete or censored patient data to reduce bias from hidden confounding.
- •For biostatistics teams: Improve modeling of patient survival curves using flexible machine learning tools that better handle complex confounding and censoring patterns.
Authors
Dhanajit Brahma, Ruolin Meng, Lawrence Carin, Ricardo Henao
Abstract
Estimating causal effects on right-censored time-to-event outcomes is challenging under unmeasured confounding. Proximal causal inference addresses unmeasured confounding through proxy variables, but existing proximal survival methods rely largely on restrictive parametric bridge models. We introduce Proximal Survival Bridge learning (PSB), a flexible doubly robust proximal framework for censored outcomes. Building on existing proximal identification of treatment-specific survival curves, the framework yields outcome-bridge (OB), treatment-bridge (TB), and doubly robust (DR) estimators, with two flexible bridge learners: PSB-K, a kernel estimator with a low-rank time decomposition, and PSB-N, a neural estimator trained by a projected maximum-moment-restriction loss. Unlike existing flexible proximal learners for scalar outcomes, PSB directly handles censoring and survival-curve targets. On a standard SEM benchmark, a proximal semi-synthetic ACTG study, and the real-world Framingham study, PSB-K performs strongest on SEM and ACTG, whereas PSB-N is the most stable on Framingham. Additional results with modern causal survival baselines, weakened or invalid proxies, and heavier censoring assess robustness across challenging settings.