CAST: Time-Varying Treatment Effects with Application to Chemotherapy and Radiotherapy on Head and Neck Squamous Cell Carcinoma
Everest Yang, Ria Vasishtha, Luqman K. Dad, Lisa A. Kachnic, Andrew Hope, Eric Wang, Xiao Wu, Yading Yuan, David J. Brenner, Igor Shuryak
TL;DR
CAST introduces a novel causal machine learning framework that models treatment effects as continuous functions over time, enabling dynamic analysis of treatment efficacy in medical survival data, exemplified on head and neck cancer treatments.
Contribution
The paper presents CAST, a new method combining parametric and non-parametric approaches to estimate continuous, time-varying treatment effects in survival data, addressing limitations of fixed time-point methods.
Findings
CAST successfully models how treatment effects evolve over time.
Application to HNSCC data reveals dynamic patterns of treatment efficacy.
The framework enhances personalized treatment planning by capturing temporal effects.
Abstract
Causal machine learning (CML) enables individualized estimation of treatment effects, offering critical advantages over traditional correlation-based methods. However, existing approaches for medical survival data with censoring such as causal survival forests estimate effects at fixed time points, limiting their ability to capture dynamic changes over time. We introduce Causal Analysis for Survival Trajectories (CAST), a novel framework that models treatment effects as continuous functions of time following treatment. By combining parametric and non-parametric methods, CAST overcomes the limitations of discrete time-point analysis to estimate continuous effect trajectories. Using the RADCURE dataset [1] of 2,651 patients with head and neck squamous cell carcinoma (HNSCC) as a clinically relevant example, CAST models how chemotherapy and radiotherapy effects evolve over time at the…
Peer Reviews
Decision·Submitted to NeurIPS 2025
**Strengths** - The paper focuses on the continuous-time survival heterogeneous treatment effect estimation problem, which is an important but under-explored area. - The robustness of the effect estimation experiments are interesting. **Weaknesses** - The paper seems to be a straightforward extension of the CSF model to continuous-time prediction using quadratic parametric or non-parametric function smoothing methods. - The use of independent CSF models for different time horizons seems flawed
Strength: This paper aims to address an important problem in continuous time causal effect estimation and survival analysis. The proposed CAST framework is evaluated on a real-world dataset, demonstrating its practical utility and relevance to clinical applications. Weakness 1. The method is evaluated on only a single dataset. Incorporating additional datasets would strengthen the generalizability and robustness of the results. 2. A simulation study would be valuable to assess CAST’s performa
Some aspects of this methodology, which is very straightforward, are unique. However, the authors do seem to ignore some of the more recent research in time-varying treatment effects within G-formula and double-robust methodology, more specifically targeted maximum likelihood estimation, that have incorporated ML algorithms for this very purpose. These methodologies have also confronted several mechanisms of bias that are not addressed with CAST (e.g., informative censoring, collider and stratif
Strengths: - The topic is relevant. - The paper is well written. It is easy to follow and nice to read. - The evaluation study on the RADCURE dataset provides valuable insights. Weaknesses: - No technical contribution: The method is very straight forward. The proposed algorithms are not novel but simply employ weighted least squares / smoothing splines. There is no technical contribution. - Limited evaluation: While the evaluation on the RADCURE dataset is extensive, there should be some sort
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAdvanced Causal Inference Techniques · Statistical Methods in Epidemiology · Statistical Methods and Bayesian Inference