Researchers at Johns Hopkins University have developed a new artificial intelligence (AI) tool that can predict the spread of infectious diseases with more accuracy than any of the methods currently in use by the Centers for Disease Control and Prevention (CDC). In a study published in Nature Computational Science, the AI model was successful at retroactively predicting the spread of COVID-19 across all U.S. states.
“We know from COVID-19 that we need better tools so that we can inform more effective policies,” said Lauren Gardner, PhD, professor at the Johns Hopkins Whiting School of Engineering and director of the university’s Center for Systems Science and Engineering. “There will be another pandemic, and these types of frameworks will be crucial for supporting public health response.”
Gardner, a specialist in modeling infectious disease risk, is also the creator behind the interactive dashboard that was used by public health authorities and researchers, as well as the general public, to track COVID-19 in real time during the pandemic. The new model developed by her team relies on large language modeling, the same type of generative AI used in tools like ChatGPT, to predict disease patterns and hospitalization trends up to three weeks in advance.
“COVID-19 elucidated the challenge of predicting disease spread due to the interplay of complex factors that were constantly changing,” said Gardner. “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 didn’t have the modeling capabilities to include critical types of information. The new tool fills this gap.”
The new AI tool could be applied to track and manage outbreaks of a wide range of infectious diseases, including bird flu, monkeypox and the respiratory syncytial virus (RSV). The model combines demographics, epidemiological, public health policy and genomic surveillance data to predict how all these factors will affect the spread and behavior of the disease.
“Traditionally we use the past to predict the future,” added Hao “Frank” Yang, PhD, assistant professor of Civil and Systems Engineering at Johns Hopkins University. “But that doesn’t give the model sufficient information to understand and predict what’s happening. Instead, this framework uses new types of real-time information.”
To test the accuracy of the new model, the research team used it to retroactively predict the spread of COVID-19 across every U.S. state over the course of 19 months. The model outperformed all other predictive methods used by the CDC such as the CovidHub weekly forecast of COVID-19-related hospital admissions. “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 modeling,” said Gardner.
As a next step, the research team will be exploring the use of large language models to better understand and predict how individuals make health-related decisions in the hopes of helping public officials design safer and more effective policies.
