New model uses social media patterns to predict disease outbreaks

Vaccination rates are falling in many communities due to widespread misinformation and previously eliminated or controlled illnesses like measles are surging across the United States and Canada. 

Researchers at the University of Waterloo have developed a new approach that could help public health officials predict where outbreaks might occur. By analyzing social media posts, the method identifies early signs of increasing vaccine skepticism - a warning signal that could emerge before any disease begins to spread. 

"In nature, we have contagious systems like diseases," said Dr. Chris Bauch, professor of Applied Mathematics at Waterloo. "We decided to look at social dynamics like an ecological system and studied how misinformation can spread contagiously from user to user through a social media network."

The team trained a machine learning model on the mathematical concept of a tipping point - the moment when a system suddenly shifts into a new state. "It doesn't matter if you're looking at a person's body having an epileptic seizure, or an ecological system like a lake getting overrun by algae, or the loss of herd immunity within a population," Bauch said. "Mathematically, there's a common underlying mechanism." 

To test their model, the researchers analyzed tens of thousands of public posts on X (formerly Twitter) from California just before a major measles outbreak in 2014. Traditional methods - such as simply counting skeptical tweets - provided very little warning before the outbreak. 

"The usual methods of predicting an outbreak by doing a statistical analysis of skeptical tweets don't provide much lead time before an outbreak," Bauch said. "By using the mathematical theory of tipping points, we were able to get a much bigger lead time and detect patterns in the data much more effectively." They verified the accuracy of the "tipping point" method by comparing posting patterns in California to those in comparable areas around the same time, where no outbreaks occurred.

This research reflects Waterloo's commitment to strengthening evidence-based decision-making and public trust in science - a core goal of the University's Societal Futures network and its new TRuST initiative, which brings philosophers, computer scientists, communicators and ethicists together to understand why trust in science falters and how to rebuild it. 

While initially tested on X, the model can be easily adapted for TikTok or Instagram; however, it would require more computing resources to analyze images and videos compared to X's predominantly text-based format.

 "Ultimately, we would like to turn this into a tool for public health officials to monitor which populations are at the highest risk for a tipping point," said Bauch. "Applied mathematics can be a powerful quantitative tool aiding in predicting, monitoring, and addressing threats to public health."