TAMER: A Test-Time Adaptive MoE-Driven Framework for EHR Representation Learning
Yinghao Zhu, Xiaochen Zheng, Ahmed Allam, Michael Krauthammer
TL;DR
TAMER introduces a test-time adaptive Mixture-of-Experts framework to address patient heterogeneity and distribution shifts in EHR representation learning. By inserting a test-time adaptation layer between a backbone model and a soft MoE, TAMER dynamically reconfigures representations and expert routing as new patient data arrive, improving mortality and readmission predictions across four real-world datasets. The approach shows consistent in-domain gains, robustness to distribution shifts, and favorable ablations, while maintaining modest computational overhead. Collectively, TAMER offers a practical, plug-in solution for dynamic, personalized clinical predictions in real-world healthcare environments.
Abstract
We propose TAMER, a Test-time Adaptive MoE-driven framework for Electronic Health Record (EHR) Representation learning. TAMER introduces a framework where a Mixture-of-Experts (MoE) architecture is co-designed with Test-Time Adaptation (TTA) to jointly mitigate the intertwined challenges of patient heterogeneity and distribution shifts in EHR modeling. The MoE focuses on latent patient subgroups through domain-aware expert specialization, while TTA enables real-time adaptation to evolving health status distributions when new patient samples are introduced. Extensive experiments across four real-world EHR datasets demonstrate that TAMER consistently improves predictive performance for both mortality and readmission risk tasks when combined with diverse EHR modeling backbones. TAMER offers a promising approach for dynamic and personalized EHR-based predictions in practical clinical settings.