Advanced AI Identifies Hidden Bird Flu Exposure Risks in Maryland Emergency Rooms

Researchers from the University of Maryland School of Medicine have introduced a powerful new application of artificial intelligence (AI) to enhance infectious disease surveillance. Utilizing a generative large language model (LLM), the team analyzed electronic medical records from 13,494 patient visits across Maryland hospitals’ emergency departments in 2024. The focus was on patients presenting symptoms such as cough, fever, congestion, or conjunctivitis—signs associated with early H5N1 avian influenza (bird flu) infections.

The AI system sifted through emergency notes to identify patients with reports of high-risk exposure to poultry, livestock, or farm work—details often documented incidentally rather than specifically suspected for bird flu. Out of these notes, the AI flagged 76 cases mentioning relevant exposure, such as working as butchers or on farms. After brief human review, 14 patients were confirmed to have had recent contact with animals known to carry H5N1, including poultry, wild birds, and livestock. Although these patients were not specifically tested for the virus, the AI's ability to find these rare high-risk exposures demonstrates its potential to uncover overlooked cases among routine respiratory illness patients.

According to lead researcher Katherine E. Goodman, Ph.D., JD, this approach could significantly fill gaps in public health monitoring by rapidly identifying high-risk individuals who might otherwise go unnoticed. The study highlights the importance of early detection, especially as H5N1 continues to circulate among animals in the U.S., with over 1,075 dairy herds affected across 17 states and more than 175 million poultry and wild birds testing positive during recent outbreaks. Human infections remain rare but are likely underreported due to limited testing, raising concerns about possible undetected spread and the emergence of new strains capable of human-to-human transmission.

The AI review, performed in just 26 minutes and costing approximately three cents per patient note, showcases the scalability and efficiency of this technology. Co-author Dr. Anthony Harris emphasizes that this method could help establish a national network for early infectious disease detection. The AI model demonstrated a 90% positive predictive value and 98% negative predictive value when tested on historical data from 2022-2023, though it was cautious in its exposure assessments, often flagging low-risk cases like dog exposures that require human review.

Looking ahead, researchers aim to refine the AI for real-time surveillance within electronic health records, enabling healthcare providers to respond swiftly to emerging threats during respiratory virus seasons. As Dr. Mark T. Gladwin notes, this innovative use of big data and AI positions Maryland at the forefront of a transformative approach to disease monitoring, potentially preventing larger outbreaks of avian influenza through early detection and targeted intervention.