Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Perspective on Individualized Treatment Effects Estimation from Time-series Health Data

Ghadeer O. Ghosheh, Moritz Gögl, Tingting Zhu

TL;DR

This paper surveys the estimation of individualized treatment effects (ITE) from time-series health data, contrasting randomized controlled trials with observational EHR data and outlining the challenges posed by time-varying confounding. It categorizes time-series ITE methods into outcome-estimation approaches (Marginal Structural Models, G-formula, and balanced representations) and deconfounding strategies (latent factor models and noisy proxies), detailing their assumptions and architectures. The review emphasizes the prevalence of simulated and semi-synthetic datasets for validation and notes the scarcity of real-world counterfactual benchmarks, while outlining future directions such as irregular sampling, missing data handling, and connections to dynamic reinforcement learning. Overall, the work serves as a comprehensive resource for researchers at the intersection of causal inference, time-series modeling, and personalized medicine, guiding methodological development and practical deployment in real-world health settings.

Abstract

The burden of diseases is rising worldwide, with unequal treatment efficacy for patient populations that are underrepresented in clinical trials. Healthcare, however, is driven by the average population effect of medical treatments and, therefore, operates in a "one-size-fits-all" approach, not necessarily what best fits each patient. These facts suggest a pressing need for methodologies to study individualized treatment effects (ITE) to drive personalized treatment. Despite the increased interest in machine-learning-driven ITE estimation models, the vast majority focus on tabular data with limited review and understanding of methodologies proposed for time-series electronic health records (EHRs). To this end, this work provides an overview of ITE works for time-series data and insights into future research. The work summarizes the latest work in the literature and reviews it in light of theoretical assumptions, types of treatment settings, and computational frameworks. Furthermore, this work discusses challenges and future research directions for ITEs in a time-series setting. We hope this work opens new directions and serves as a resource for understanding one of the exciting yet under-studied research areas.

A Perspective on Individualized Treatment Effects Estimation from Time-series Health Data

TL;DR

This paper surveys the estimation of individualized treatment effects (ITE) from time-series health data, contrasting randomized controlled trials with observational EHR data and outlining the challenges posed by time-varying confounding. It categorizes time-series ITE methods into outcome-estimation approaches (Marginal Structural Models, G-formula, and balanced representations) and deconfounding strategies (latent factor models and noisy proxies), detailing their assumptions and architectures. The review emphasizes the prevalence of simulated and semi-synthetic datasets for validation and notes the scarcity of real-world counterfactual benchmarks, while outlining future directions such as irregular sampling, missing data handling, and connections to dynamic reinforcement learning. Overall, the work serves as a comprehensive resource for researchers at the intersection of causal inference, time-series modeling, and personalized medicine, guiding methodological development and practical deployment in real-world health settings.

Abstract

The burden of diseases is rising worldwide, with unequal treatment efficacy for patient populations that are underrepresented in clinical trials. Healthcare, however, is driven by the average population effect of medical treatments and, therefore, operates in a "one-size-fits-all" approach, not necessarily what best fits each patient. These facts suggest a pressing need for methodologies to study individualized treatment effects (ITE) to drive personalized treatment. Despite the increased interest in machine-learning-driven ITE estimation models, the vast majority focus on tabular data with limited review and understanding of methodologies proposed for time-series electronic health records (EHRs). To this end, this work provides an overview of ITE works for time-series data and insights into future research. The work summarizes the latest work in the literature and reviews it in light of theoretical assumptions, types of treatment settings, and computational frameworks. Furthermore, this work discusses challenges and future research directions for ITEs in a time-series setting. We hope this work opens new directions and serves as a resource for understanding one of the exciting yet under-studied research areas.
Paper Structure (18 sections, 1 figure, 1 table)

This paper contains 18 sections, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Randomized Controlled Trial Data
  3. Observational Data: From Efficacy to Effectiveness
  4. Challenges of Treatment Effects Estimation
  5. From Average to Individualized Treatment Effects Estimation
  6. Estimating Individualized Treatment Effects from Time-series Data
  7. Assumptions
  8. ITE Outcome Estimation Methods in time-series data
  9. Marginal Structural Models (MSM)
  10. G-formula
  11. Balanced Representations
  12. ITE Deconfounding methods for time-series data
  13. Latent Factor Models
  14. Noisy Proxies
  15. Datasets and Evaluation
  16. ...and 3 more sections

Figures (1)

  • Figure 1: Illustration of a causal model underlying a dynamic ITE estimation setting with time-varying treatments and covariates. The arrows indicate causal dependencies between variables. Here, the observed covariates are blood pressure (BP) readings that act as time-varying confounders affecting all subsequent treatment decisions (ACE inhibitor vs. ARB) and the outcome of interest (risk of stroke). We also showcase an example of an unobserved "hidden" confounder, namely potential comorbidity that is not directly represented in the data. All connections from covariates and unobserved variables to the treatment variables are depicted in red, and would need to be accounted for to remove confounding bias effectively.