A Review of Deep Learning for Individualized Treatment Effect Estimation in Healthcare Time Series

The proliferation of healthcare time series data presents a transformative opportunity for personalized healthcare. However, utilizing these observational data for clinical decision support requires moving beyond risk prediction to causal inference. In particular, estimating time-varying individual treatment effects is crucial for answering "what-if" questions in dynamic settings, where patient states and treatment efficacy evolve continuously over time. Traditional statistical and machine learning methods face well-documented difficulties in capturing the long-range temporal dependencies, high dimensionality, and non-linear treatment-response dynamics characteristic of such data. Deep learning has emerged as a powerful paradigm for addressing these challenges, motivating a rapidly growing body of work on time-varying causal inference from healthcare time series data. This review aims to provide a systematic overview of deep learning for time-varying ITE estimation in healthcare, connecting theoretical foundations with methodological practice. We first establish the theoretical foundations, then present a taxonomy of deep learning methodologies, followed by an analysis of current clinical applications and model evaluation practices. Finally, we discuss key unresolved challenges and outline a roadmap for future research on time-varying ITE estimation.