Skip to content
Health — Medical-conditions — Health-and-wellness — Engineering — Research — Technology — Science — Public health

AI revolutionizes predictive analysis of infectious diseases

Predictive Instrument for Infectious Disease Threat Employs Advanced Language Modeling

,   and  Administrator
3 min read
Innovative instrument employs extensive language modeling to forecast potential health risks...
Innovative instrument employs extensive language modeling to forecast potential health risks associated with contagious diseases.

AI revolutionizes predictive analysis of infectious diseases

Revamped Article:

A cutting-edge AI tool developed with federal support by researchers at Johns Hopkins and Duke universities, known as PandemicLLM, is setting new standards in predicting infectious disease spread, outperforming present state-of-the-art forecasting methods.

This revolutionary tool could reshape the way public health officials predict, monitor, and manage outbreaks of infectious diseases, such as flu and COVID-19. According to Lauren Gardner, a Johns Hopkins modeling expert and creator of the globally-used COVID-19 dashboard during the pandemic, "COVID-19 highlighted the challenge of predicting disease spread due to the interplay of intricate, ever-changing factors. When circumstances remained stable, the models worked fine. However, when new variants emerged or policies shifted, we were poor at predicting outcomes because we didn't have the modeling capabilities to incorporate critical information. PandemicLLM bridges this gap."

As published today in Nature Computational Science, this groundbreaking work was previously impossible during the coronavirus pandemic due to the lack of technology underpinning the new tool. For the first time, the team utilizes large language modeling, the same technology behind ChatGPT, to predict disease spread.

Unlike conventional approaches that view forecasting as merely a mathematical problem, PandemicLLM employs reasoning, taking into account factors such as recent infection spikes, new variants, and mask mandates. The team fed the model an array of information, including data never previously used in pandemic prediction tools, discovering that PandemicLLM could accurately predict disease patterns and hospitalization trends one to three weeks ahead, consistently outperforming other methods, including those at the top of the CDC's CovidHub.

"A crucial challenge in disease prediction is determining the influences driving disease surges," Gardner said. "The new modeling framework is integral for incorporating these newly developed information streams."

PandemicLLM draws on four types of data:

  1. State-level spatial data, featuring demographics, healthcare system details, and political leanings.
  2. Epidemiological time series data, like reported cases, hospitalizations, and vaccine rates.
  3. Public health policy data, encompassing stringency and kinds of government policies.
  4. Genomic surveillance data, providing insights into the characteristics of disease variants and their prevalence.

Once processed, the model uses its understanding of these elements to project how they will interact to shape the behavior of the disease.

For testing, the team applied the model retroactively to the COVID-19 pandemic, examining each U.S. state over 19 months. In comparison to other models, PandemicLLM was particularly effective when the outbreak was in flux.

"Historically, we predict the future based on the past," said author Hao "Frank" Yang, a Johns Hopkins assistant professor of Civil and Systems Engineering specializing in dependable AI. "However, this doesn't provide sufficient information for the model to comprehend and predict what is happening. Instead, this framework utilizes new real-time information types."

With the appropriate data, PandemicLLM can be tailored to predict the spread of any infectious disease, including bird flu, monkeypox, and RSV. The team is now investigating LLMs' potential to emulate individual health decision-making processes, which could help officials design safer and more effective policies.

"We learned from COVID-19 that we need better tools so that we can inform more effective policies," Gardner added. "There will be another pandemic, and these types of frameworks will be vital for supporting public health response."

Other authors involved were 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 Professor Yiran Chen.

This research was supported by the National Science Foundation (NSF 2229996), the Centers for Disease Control and Prevention (CDC RFA-FT-23-0069, CDC Center for Forecasting and Outbreak Analytics 6 NU38FT000012-01), the Merck KGaA Future Insight Prize, the NSF (2112562), and the Army Research Office (W911NF-23-2-0224).

Posted in Science+Technology

Tagged nsf, artificial intelligence

  1. The groundbreaking AI tool, PandemicLLM, developed by researchers at Johns Hopkins and Duke universities, employs artificial intelligence and technology to predict infectious disease spread more accurately than current methods, aiding public health officials in managing outbreaks like flu and COVID-19.
  2. According to Lauren Gardner, a Johns Hopkins modeling expert and creator of the COVID-19 dashboard, PandemicLLM bridges the gap in predicting disease spread caused by the interplay of complex, ever-changing factors, outperforming other methods during periods of instability such as new variants emerging or policies shifting.
  3. Previously impossible during the coronavirus pandemic, this research utilizes artificial intelligence and large language modeling, similar to the technology behind ChatGPT, to predict disease spread by taking into account factors such as infection spikes, new variants, and mask mandates.
  4. The team behind PandemicLLM draws on four types of data: state-level spatial data, epidemiological time series data, public health policy data, and genomic surveillance data, allowing the model to project how these factors will interact to shape the behavior of the disease and accurately predict disease patterns and hospitalization trends one to three weeks ahead.
  5. Moving forward, PandemicLLM can be tailored to predict the spread of various infectious diseases, including bird flu, monkeypox, and RSV, and the team is investigating the potential of AI and artificial intelligence to emulate individual health decision-making processes, helping officials design safer and more effective policies for health-and-wellness and medical-conditions within the broader field of science and technology.

Read also:

    Related

    Latest