Position: Public Health Systems Should Embrace a Multi-Layered Epidemic Early-Warning with LLM Agents and Local Knowledge Enhancement

OU DENG, Qun Jin

TL;DR: Multi-Layered Epidemic Early-Warning with LLM Agents and Local Knowledge Enhancement

Abstract: We posit that public health systems worldwide should adopt a multi-layered epidemic early-warning mechanism, coupling large language model (LLM) agents with locally enriched knowledge bases. Specifically, we propose a three-tier framework comprising (i) distributed multi-agent data ingestion, (ii) centralized vector-based analytics and Reinforcement Learning (RL)-driven optimization, and (iii) regionally maintained expert repositories for final validation. By synchronizing real-time social media data, clinical records, and domain insights, our approach aims to accelerate detection, refine risk assessment, and expedite intervention for novel infectious threats. In particular, we highlight benefits for multi-modal data fusion, cross-lingual coverage, and privacy preservation. We further address critiques regarding model reliability, data governance, and resource allocation, outlining how federated learning protocols and human oversight mitigate these challenges. Ultimately, we reaffirm that integrating LLM-centric workflows with local expertise and iterative refinement offers a scalable path to strengthening epidemic surveillance, providing an adaptive, context-aware shield against emerging outbreaks.

Primary Area: Research Priorities, Methodology, and Evaluation

Keywords: Epidemic Early-Warning, Social Network, Large Language Model, Agents, Multi-Modal, Knowledge Base

Submission Number: 201