New AI Tool Accurately Predicts Spread of Infectious Disease

Now, a newly developed AI tool is redefining outbreak prediction by significantly outperforming existing state-of-the-art forecasting methods. By combining generative AI with real-world health data, the model enables more accurate predictions of disease trends and hospitalizations, offering a powerful new tool for pandemic preparedness and response.

The AI tool, named PandemicLLM, was developed by a research team at Johns Hopkins University (Baltimore, MD, USA) by applying large language modeling for the first time, most famously seen in tools like ChatGPT, to meet the challenge of disease forecasting. Unlike traditional mathematical models, PandemicLLM "reasons" through information, using context to interpret and synthesize vast and varied data inputs. This innovation marks a major departure from the purely statistical approaches that dominated during the COVID-19 pandemic.

PandemicLLM incorporates four main categories of data: state-level spatial data (including demographics, healthcare infrastructure, and political affiliations), epidemiological time series (such as reported cases, hospitalizations, and vaccine uptake), public health policy data (such as mask mandates and lockdowns), and genomic surveillance information (like variant characteristics and prevalence). With this diverse input, the model constructs a holistic view of the current outbreak environment and forecasts likely outcomes over the next one to three weeks.

To test its performance, researchers retroactively applied PandemicLLM to 19 months of U.S. COVID-19 data at the state level. The model consistently outperformed all existing tools, including the top-ranking models in the CDC’s CovidHub, particularly excelling during volatile phases of the pandemic. Its success was attributed to its ability to integrate new, previously untapped data streams into its forecasts.

The adaptability of PandemicLLM means it can be tailored to predict the course of various infectious diseases beyond COVID-19, including bird flu, RSV, and monkeypox, provided the relevant data is available. The research team is now exploring how large language models might also simulate human decision-making around health behaviors, a development that could further refine future public health strategies.

“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 study author Lauren Gardner of Johns Hopkins, a modeling expert who created the COVID-19 dashboard that was relied upon by people worldwide during the pandemic. “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.” The results of the study were published on June 6 in Nature Computational Science.