CAST: Causal Modeling of Time-Varying Treatment Effects on Head and Neck Cancer

Everest Yang, Ria Vasishtha, Luqman K. Dad, Lisa A. Kachnic, Andrew Hope, Eric Wang, Xiao Wu, Yading Yuan, David J Brenner, Igor Shuryak

Keywords: Causal inference, Survival analysis, Time-varying effects, Computational oncology, Machine learning for healthcare

TL;DR: CAST is a novel framework that models treatment effects as continuous functions of time in survival analysis, enabling clinicians to identify optimal treatment windows and deliver personalized care across medical contexts.

Abstract: Causal machine learning (CML) is quickly gaining recognition in medical research because it offers better strategies to estimate treatment effects from complex real-world data, helping guide treatment optimization. Causal survival forests (CSF) are a powerful CML method for estimating heterogeneous treatment effects in survival outcomes, which are essential for informed healthcare decision-making. However, CSF only estimates effects at a fixed horizon, rather than at multiple time points. We introduce Causal Analysis for Survival Trajectories (CAST), a novel extension of CSF that models treatment effects as continuous parametric and non-parametric effect trajectories over time. Applied to the RADCURE dataset [0] of 2,651 head and neck cancer patients, CAST reveals how the effects of chemotherapy and radiotherapy evolve over time at the population and individual levels. By capturing the temporal dynamics of treatment response, CAST can help clinicians determine when and for which patient subgroups treatment benefits are maximized.

Submission Number: 4