Federated Inverse Probability Treatment Weighting for Individual Treatment Effect Estimation

2025-03-06

AI Summary

In multi-center healthcare settings, privacy constraints prevent data sharing, existing individual treatment effect (ITE) estimation methods are incompatible with federated learning, and confounding bias impairs estimation accuracy. To address these challenges, we propose FED-IPTW—the first ITE estimation framework designed specifically for federated environments. FED-IPTW innovatively co-optimizes global and local covariate–treatment disentanglement via distributed propensity score estimation, inverse probability weighting (IPTW), and joint covariate–treatment decorrelation modeling, effectively mitigating IPTW weight miscalibration and compounded local bias under federation. Evaluated on synthetic data and real-world eICU data, FED-IPTW achieves statistically significant improvements over state-of-the-art methods in key metrics—including Precision in Estimation of Heterogeneous Effect (PEHE) and Average Treatment Effect (ATE) estimation error—demonstrating robust causal inference capability. The framework enables trustworthy, privacy-preserving personalized decision support for mechanical ventilation in ICU settings.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Intelligent Robots: State Estimation
• Reasoning under Uncertainty: Causality

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Systems and Infrastructure for Web, Mobile and WoT: Federated Web and WoT systems, including distributed, federated and edge-based data processing
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Individual treatment effect (ITE) estimation is to evaluate the causal effects of treatment strategies on some important outcomes, which is a crucial problem in healthcare. Most existing ITE estimation methods are designed for centralized settings. However, in real-world clinical scenarios, the raw data are usually not shareable among hospitals due to the potential privacy and security risks, which makes the methods not applicable. In this work, we study the ITE estimation task in a federated setting, which allows us to harness the decentralized data from multiple hospitals. Due to the unavoidable confounding bias in the collected data, a model directly learned from it would be inaccurate. One well-known solution is Inverse Probability Treatment Weighting (IPTW), which uses the conditional probability of treatment given the covariates to re-weight each training example. Applying IPTW in a federated setting, however, is non-trivial. We found that even with a well-estimated conditional probability, the local model training step using each hospital's data alone would still suffer from confounding bias. To address this, we propose FED-IPTW, a novel algorithm to extend IPTW into a federated setting that enforces both global (over all the data) and local (within each hospital) decorrelation between covariates and treatments. We validated our approach on the task of comparing the treatment effects of mechanical ventilation on improving survival probability for patients with breadth difficulties in the intensive care unit (ICU). We conducted experiments on both synthetic and real-world eICU datasets and the results show that FED-IPTW outperform state-of-the-art methods on all the metrics on factual prediction and ITE estimation tasks, paving the way for personalized treatment strategy design in mechanical ventilation usage.

Problem

• Estimating individual treatment effects in healthcare
• Addressing data privacy in federated clinical settings
• Reducing confounding bias in decentralized data analysis

Innovation

• Federated learning for decentralized hospital data
• Inverse Probability Treatment Weighting adaptation
• Global and local decorrelation of covariates