![[2602.12833] TRACE: Temporal Reasoning via Agentic Context Evolution for Streaming Electronic Health Records (EHRs)](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.12833] TRACE: Temporal Reasoning via Agentic Context Evolution for Streaming Electronic Health Records (EHRs)
Summary
TRACE introduces a novel framework for temporal reasoning in electronic health records, enhancing prediction accuracy and clinical safety through structured memory management.
Why It Matters
The TRACE framework addresses significant challenges in applying large language models to clinical data by providing a robust method for temporal reasoning. This is crucial for improving patient care and decision-making in healthcare settings, particularly as reliance on AI in medicine grows.
Key Takeaways
- TRACE enhances temporal reasoning in EHRs using a dual-memory architecture.
- The framework maintains context without the need for fine-tuning or retrieval-based methods.
- Improvements in next-event prediction accuracy and clinical safety are demonstrated over existing methods.
- TRACE features agentic components for structured memory management and safety audits.
- The approach is validated using longitudinal clinical event streams from MIMIC-IV.
Computer Science > Machine Learning arXiv:2602.12833 (cs) [Submitted on 13 Feb 2026] Title:TRACE: Temporal Reasoning via Agentic Context Evolution for Streaming Electronic Health Records (EHRs) Authors:Zhan Qu, Michael Färber View a PDF of the paper titled TRACE: Temporal Reasoning via Agentic Context Evolution for Streaming Electronic Health Records (EHRs), by Zhan Qu and Michael F\"arber View PDF HTML (experimental) Abstract:Large Language Models (LLMs) encode extensive medical knowledge but struggle to apply it reliably to longitudinal patient trajectories, where evolving clinical states, irregular timing, and heterogeneous events degrade performance over time. Existing adaptation strategies rely on fine-tuning or retrieval-based augmentation, which introduce computational overhead, privacy constraints, or instability under long contexts. We introduce TRACE (Temporal Reasoning via Agentic Context Evolution), a framework that enables temporal clinical reasoning with frozen LLMs by explicitly structuring and maintaining context rather than extending context windows or updating parameters. TRACE operates over a dual-memory architecture consisting of a static Global Protocol encoding institutional clinical rules and a dynamic Individual Protocol tracking patient-specific state. Four agentic components, Router, Reasoner, Auditor, and Steward, coordinate over this structured memory to support temporal inference and state evolution. The framework maintains bounded inference co...
