New method improves survival effect estimates with changing treatments
Doubly Robust Learning of Heterogeneous Survival Treatment Effects under Time-Varying Treatments
Machine Learning
Summary
Figuring out how treatments affect patient survival is tricky when treatments and patient conditions change over time. The authors propose a new method that can handle these time-varying treatments and changes in patient health information while also dealing with patients leaving the study early. Their method uses sophisticated math to remain accurate even if some parts of the analysis are slightly off. Experiments show this approach works well for estimating who will benefit most from different treatment sequences.
What this means in practice
- •For hospital data teams: Help tailor sequential treatment plans by accurately estimating survival benefits in patients with evolving medical conditions.
- •For pharmaceutical analysts: Improve evaluation of drug effectiveness by modeling how treatments and dropouts vary over time in clinical trial data.
Authors
Tomàs Garriga, Munib Mesinovic, Luke De Bretton-Gordon, Gerard Sanz, Axel Brando, Tingting Zhu
Abstract
Estimating heterogeneous survival treatment effects (HSTEs) from time-to-event data under time-varying treatments is important in fields such as medicine (e.g., to decide which sequence of treatments is better in terms of survival probability). However, this task is challenging because treatment decisions, patient characteristics, and censoring probability often evolve together, requiring sequential adjustment for both treatments and censoring. Existing survival learners mostly consider a single baseline treatment assignment, while longitudinal treatment-effect learners typically assume uncensored outcomes. In this paper, we introduce a novel doubly robust learner for heterogeneous survival treatment effects under time-varying treatments and history-dependent censoring. Our framework separates the intervention window from the survival horizon, with treatment prescribed for a finite period, after which care follows its natural course. We further establish the favorable theoretical properties of Neyman-orthogonality, rate double robustness and quasi-oracle efficiency. The framework accommodates flexible machine learning models for nuisance estimation and treatment-effect regression. We evaluate our learners through several numerical experiments, demonstrating their effectiveness.