According to a study published in Nature Computational Science, a novel AI tool developed with federal funding by researchers at Johns Hopkins and Duke universities to anticipate the spread of an infectious disease beats current cutting-edge forecasting approaches.
Image Credit: Corona Borealis Studio/Shutterstock.com
The technique could transform how public health professionals forecast, track, and control infectious disease epidemics such as flu and COVID-19.
COVID-19 elucidated the challenge of predicting disease spread due to the interplay of complex factors that were constantly changing. When conditions were stable the models were fine. However, when new variants emerged or policies changed, we were terrible at predicting the outcomes because we did not have the modeling capabilities to include critical types of information. The new tool fills this gap.
Lauren Gardner, Study Author, Johns Hopkins University
The technology underlying the new tool did not exist during the coronavirus pandemic. For the first time, the team predicts disease transmission using big language modeling, a generative AI popularized by ChatGPT.
Instead of approaching prediction as a math issue, the PandemicLLM model uses reasoning to analyze inputs such as current infection spikes, novel variations, and mask mandates.
The team fed the model streams of information, including data never before used in pandemic prediction tools, and discovered that PandemicLLM could accurately predict disease patterns and hospitalization trends one to three weeks in advance, consistently outperforming other methods, including the top performers on the CDC's CovidHub.
The team fed the model data streams, including data that had never been used in pandemic prediction tools. They discovered that PandemicLLM consistently outperformed other methods, including the best-performing ones on the CDC's CovidHub, in accurately predicting hospitalization trends and disease patterns one to three weeks out.
Gardner added, “A pressing challenge in disease prediction is trying to figure out what drives surges in infections and hospitalizations, and to build these new information streams into the modelling.”
The model is based on four categories of data:
- Spatial data at the state level, such as demographics, health care, and political connections
- Epidemiological time series data such as reported cases, hospitalizations and vaccine rates
- Public health policy data, including stringency and types of government policies
- Genomic surveillance data, including disease variant features and prevalence
After digesting this data, the model can anticipate how the various components will interact to influence the disease's behavior.
To test it, the researchers applied it retroactively to the COVID-19 pandemic, focusing on each state in the United States over a 19-month period. Compared to existing models, the new tool performed especially well throughout the pandemic.
Traditionally, we use the past to predict the future. But that does not give the model sufficient information to understand and predict what is happening. Instead, this framework uses new types of real-time information.
Hao “Frank” Yang, Study Author and Assistant Professor, Civil and Systems Engineering, Johns Hopkins University
With the correct data, the model can be applied to any infectious disease, such as avian flu, monkeypox, and RSV.
The researchers are currently investigating whether LLMs can mimic how people make health-related decisions. They hope that politicians will be able to create safer and more efficient rules.
Gardner added, “We know from COVID-19 that we need better tools so that we can inform more effective policies. There will be another pandemic, and these types of frameworks will be crucial for supporting public health response.”
Johns Hopkins PhD student Hongru Du; Johns Hopkins graduate student Yang Zhao; Jianan Zhao of the University of Montreal; Johns Hopkins PhD student Shaochong Xu, Xihong Lin of Harvard University, and Duke University Prof. Yiran Chen are the study authors.
Journal Reference:
Du, H., et al. (2025) Advancing real-time infectious disease forecasting using large language models. Nature Computational Science. doi.org/10.1038/s43588-025-00798-6.