The Digital Health Perspectives

In recent years, the confluence of AI and public health has paved the way for unprecedented advancements in epidemic intelligence and health threat surveillance. This report encapsulates the pioneering work conducted under the auspices of two landmark projects of the JRC: DigLife (Digital Innovation in Life and Health Sciences) and ETOHA (Emerging Health Threats and the One Health Approach). These initiatives, in exchange with other JRC units, DG SANTE, ECDC, WHO and DG HERA through recurrent touchpoints, embark on a journey that began in early 2024 to explore the transformative applications of AI in epidemic intelligence. In particular, the report provides detailed insights of how generative AI, particularly Large Language Models (LLMs), is being leveraged to extract and analyse epidemiological information. At the forefront of this exploration is the innovative use of AI to harness unstructured epidemiological data from the Epidemic Intelligence from Open Sources (EIOS) system, which includes platforms such as ProMED and WHO Disease Outbreak News. This report delves into the methodologies developed to extract and structure critical epidemiological information, thereby enhancing the surveillance and response capabilities for health threats. The analysis confirms that generative AI holds immense potential to fundamentally reshape public health surveillance, enabling more timely, accurate, and proactive responses to disease outbreaks. The report highlights core applications, including the use of LLMs for Epidemic Information Extraction, which has culminated in the creation of an epidemiology knowledge graph. A key methodological advancement is the use of an ensemble approach with multiple LLMs, which enhances reliability and mitigates the limitations of individual models. Through comparative analyses including In-context Learning and Fine-tuning approaches, we introduce some fine-tuned LLMs optimised for granular and rapid epidemic intelligence extraction from EIOS data. Furthermore, this report elucidates the integration of Retrieval Augmented Generation (RAG) techniques over epidemiological reports, exemplified by WHO Disease Outbreak News, to generate AI-driven health threat and disaster storylines. By leveraging AI to analyse, surveil, and respond to disease outbreaks, this report underscores the potential for digital health perspectives to innovate public health surveillance and response strategies. In addition, it provides several functional AI-based prototypes that underline the above-mentioned use cases, and can be further enhanced into more operational use. Despite this high potential, the report also provides an analysis of significant challenges. These include technical limitations of LLMs, such as the black-box problem, a propensity for hallucinations, and slow knowledge adaptation. Moreover, substantial ethical and societal challenges are identified, including data equity concerns that could reinforce health disparities, acute data privacy and surveillance risks, and the amplification of medical misinformation. The report concludes that the future of generative AI in public health is a collaborative one, where technology serves as a powerful accelerator but does not displace ultimate human responsibility. A trustworthy future will require a “problem-first, technology-second” approach, prioritising ethical design, human oversight, and global collaboration to ensure the benefits of this technology are accessible for all.

CONSOLI Sergio;  BERTOLINI Lorenzo;  BIAZZO Indaco;  CERESA Mario;  COMTE Valentin;  MARKOV Peter;  ORFEI Lia;  RONCO Michele;  SCHUH Lea;  STEFANOVITCH Nicolas;  STILIANAKIS Nikolaos; 

2026-07-14

Publications Office of the European Union

JRC142475

978-92-68-41024-0 (online),   

1831-9424 (online),   

EUR 40768,    OP KJ-01-26-276-EN-N (online),   

https://publications.jrc.ec.europa.eu/repository/handle/JRC142475,   

10.2760/6953152 (online),