A powerful new AI tool designed by researchers at Johns Hopkins and Duke universities could transform how public health officials predict and manage infectious disease outbreaks like flu and COVID-19. This cutting-edge system, called PandemicLLM, has outperformed current top forecasting methods, offering fresh hope for handling future pandemics more effectively.
The tool uses large language modeling the same generative AI technology behind ChatGPT — to “reason” through complex, changing data. It moves beyond traditional mathematical forecasting by considering factors like new virus variants, mask mandates, and recent spikes in infections.
“COVID-19 taught us that old models broke down when things changed like when new variants arrived or public policies shifted,” said Lauren Gardner from Johns Hopkins University, who also created the widely used COVID-19 global dashboard. “Our new tool can handle those sudden shifts because it includes all kinds of important real-world information.”
Smarter Predictions, Better Results
PandemicLLM digests four main types of data to predict outbreaks:
•State-level spatial data – covering demographics, healthcare capacity, and even political leanings.
•Epidemiological time series – including daily COVID-19 cases, hospital admissions, and vaccination rates.
•Public health policy records – such as lockdown rules, travel restrictions, and mask guidelines.
•Genomic surveillance data – revealing details of virus mutations and their spread.
By combining this rich mix of data, the AI can predict disease patterns and hospitalization trends up to three weeks in advance — far better than other leading models, including those used by the U.S. CDC’s CovidHub.
Hao “Frank” Yang, another researcher on the project, said the breakthrough came because the AI doesn’t just look to the past to predict the future. “Traditional models assume what happened yesterday will happen tomorrow,” he explained. “But this new tool processes real-time information to understand what’s truly going on.”
Tested and Proven
To test PandemicLLM, the team applied it to data from all 50 U.S. states over a 19-month period during the COVID-19 pandemic. The tool proved especially accurate during periods of change such as when new variants appeared or policies shiftedwhen other models typically struggled.
Importantly, the system is flexible. With the right data, it can be adapted for any infectious disease like bird flu, monkeypox, or RSV making it a valuable tool beyond COVID-19.
Shaping the Future of Public Health
The researchers now hope to take this technology further by modeling how people make health decisions during outbreaks — such as when to wear masks or get vaccinated. This could help public officials craft better policies that save lives.
“We know another pandemic will come,” Gardner warned. “This kind of AI framework will be critical to respond quickly, smartly, and safely when that happens.”
The study highlights the growing role AI could play in protecting public health not just by predicting disease spread, but by shaping smarter strategies to keep populations safe.