What if, during the COVID-19 pandemic, we were able to forecast how populations should respond to lockdowns and policy changes? What if we could anticipate outbreaks of dengue, influenza and HIV? What if we were one step ahead of infection across diseases and regions?
With AI, these possibilities are closer than we think. In just 60 years, the United Nations expects the world’s population to peak at approximately 10.3 billion, roughly 2.1 billion more than the current count. With a rapidly growing population, disease modeling is more crucial than ever. Infectious disease experts race to leverage artificial intelligence to predict the next disease outbreak before it reaches pandemic-level potential.
In a scoping review published in Nature Communications, scientists from the University of Florida Emerging Pathogens Institute and the Yale School of Public Health carefully analyzed over 15,000 studies that applied models for various infectious diseases. They found that hybrid models — those combining AI with traditional epidemiological frameworks — are redefining what’s possible in outbreak forecasting.
AI is already transforming disease modeling, but no one has analyzed the pros and cons of AI in epidemiological modeling at this scale. The review examined areas where dense data was underutilized in disease forecasting to find opportunities for modeling with AI.
“We wanted to see how wide a coverage we could get, and it really wasn’t clear because you had very different literatures dealing with AI alone and epidemiology alone,” said EPI member Burton Signer, Ph.D., who served as a consultant and writer for the review. “So, we thought, well, how much has actually been done putting them together?”
Singer sees AI as an instrumental tool in predicting disease outbreaks.
“I spent basically my whole career doing epidemiology … and it didn’t make any sense to me that you shouldn’t try to put (AI and epidemiology) together,” Singer said.
After cleaning up the data and screening the remaining studies, their search narrowed the 15,460 studies to 245 peer-reviewed papers that successfully integrated machine learning techniques into traditional models. These hybrid models are faster, smarter and better equipped to adapt to rapidly changing disease dynamics — a Goldilocks zone of robot and human collaboration.
Mechanistic models, the traditional way to model disease outbreaks, aren’t perfect. Their reliability depends on accurate data. In large quantities, this data is overwhelming, often simplified and sometimes unavailable for modeling. To add another layer to the laboriousness, data from satellites, social media and search queries requires time-consuming efforts to sift through and manually extract useful information.
“That’s where AI really plays a role. You can access things rapidly and with great diversity using the AI technology that you can’t do otherwise,” Singer said.
Application areas of AI in epidemiological modeling
Model parameterization and calibration
These processes involve identifying the best values for epidemiological model parameters. Parameterization assigns specific values to these parameters, whereas calibration fine-tunes those values to ensure the model’s results closely match real-world data.
Disease intervention assessment and optimization
Disease intervention assessment and optimization use epidemiological models to assess the potential impacts of interventions like vaccination campaigns and contact tracing programs, and determine the best strategies considering constraints such as cost and feasibility.
Retrospective epidemic course analysis
Reconstructing epidemic trajectories under factors like transmissibility and human behavior, retrospective epidemic course analysis aims to show how these influence transmission patterns, identify key factors and guide future preparedness.
Transmission interference
Transmission interference leverages observed data like case counts, contact tracing details and genome sequences to identify the underlying disease transmission patterns, which helps in implementing focused control measures.
Infectious disease forecasting
Infectious disease forecasting aims to predict the future trajectory of disease outbreaks using epidemiological models. When combined with traditional data collection methods, future disease outbreak models can be more accurate and account for a broader range of variables.
Outbreak detection
Outbreak detection refers to identifying new disease outbreaks or unusual increases in cases of existing diseases. Early detection enables a rapid public health response, which can help contain disease spread, lessen its impact and prevent healthcare systems from becoming overwhelmed.
Despite the clear positives, researchers are careful to note that this innovation is not about abandoning classical methods. Instead, they believe AI can complement traditional epidemiological practices, bringing speed, precision and scalability to an already rigorous scientific discipline.
The team found that only a small fraction of hybrid models fully leveraged AI’s potential. Most lacked behavioral realism, failing to simulate how individuals actually respond to risk, policy or misinformation. The researchers identified a significant gap in integrated models that failed to include the economic factors decision makers took into account to balance health benefits and costs. Other models were not tested in real-time settings or lacked transparency in assumptions.
“Here’s the crazy thing about (the initial number of studies): that whole literature, it’s focused on trying to get replication, replication of what I might call ideal circumstances,” Singer said.
Singer and the other review authors call for broader collaboration across public health, computer science, behavioral science and data ethics to fully harness these technologies. They also highlight the need for better access to real-time, high-quality data streams.
With growing populations, more infectious disease outbreaks and rising temperatures, complex problems require multidisciplinary, innovative responses. When done carefully and skillfully, the combination of AI and epidemiology can be a powerful solution.
Written by: Sydney Burge