COVID data transformed disease projection models—researchers explain what’s next

Scientists sometimes compare predicting the course of epidemics to forecasting the weather. But there’s a major difference—the impact of human behavior—says Alessandro Vespignani, director of Northeastern University’s Network Science Institute.

Consider what happens during a downpour, he says. “If we all open an umbrella, it will rain anyway.”

“In epidemics, if we all open the umbrella in the sense that we behave differently, the epidemic will spread differently,” Vespignani says. “If we are more risk averse, we might avoid places. We might wash our hands more and so on and so forth.”

That makes modeling the interplay between human behavior and infectious disease transmission one of the remaining key challenges in epidemiology, according to a paper Vespignani and colleagues published in the Proceedings of the National Academy of Sciences.

“It’s very difficult to integrate behavior in the models,” especially since existing behavioral models often lack real-world data calibration, says Vespignani, Northeastern’s Sternberg Family Distinguished Professor.

But now, thanks to what they learned during COVID-19, researchers say they have found a solution.

The pandemic released a global flood of data in terms of traceable illness and death, accompanied by electronic data such as geolocation from mobile phones that indicated changing patterns in daily commutes, Vespignani says.

Being allowed access to such large data sets led the researchers to some surprising discoveries about the best ways to incorporate behavioral changes into models of disease progression, Vespignani says.

“We are really moving the frontier of epidemic and outbreak analytics and forecasting to the next level,” he says.

“All the data accumulated in the past few years and the knowledge is creating an understanding that hopefully will put us in a different place the next time we have to manage an infectious disease threat.”

The study looked at three different behavioral models—one data-driven and two mechanistic—across nine geographic areas during the first wave of COVID to evaluate how well they were able to capture the interplay between disease transmission and behavior.

The mechanistic model, which describes the mechanism of behavioral changes, often has superior or equivalent performance to the data-driven model, which employs mobility data to capture behavioral changes, in coming up with both a short-term forecast and retrospective analysis, Vespignani says.

“In a sense that was a bit of a surprise,” given scientists’ traditional preference for data modeling, he says.

A major advantage of mechanistic models is how they took into consideration that individuals exposed to the news of the pandemic started to change their behavior even before mandates were established, Vespignani says.

And risk aversion grew as COVID spread and more people were infected.

“There is a spontaneous component to what people do that has to be integrated in which we think about the trajectory of the disease,” Vespignani says.

“That opens new scenarios in the way we are going to forecast and analyze infectious diseases in the future when we can finally (put) this behavioral component to work.

“In many cases in the past, we had to work with very limited data sets, generally about the flu. We didn’t have such large-scale data,” he says.

“Now with COVID-19 we have data from across the world at all geographical resolutions, so we can really test the models.”

For the study, researchers incorporated data from departments of health and government in Bogota, Chicago, Jakarta, London, Madrid, New York and Rio de Janeiro, as well as Santiago, Chile, and the Gauteng province in South Africa.

“We have data about deaths. We have data about infections. We have data about hospitalizations,” Vespignani says.

In addition to the health data, the researchers also had unprecedented access to tech company analytics on mobility and consumer behavior, Vespignani says. “During COVID there was an all-hands-on-deck effort and so we finally got data that was not available before,” he says.

In the future, researchers can use the models to incorporate behavior changes into projections not only of pandemics but also of seasonal respiratory illnesses, Vespignani says.

It will help health and government officials develop the best approaches to communicating risk and developing risk-reduction strategies, he says.

“As soon as (disease) incidence grows, and you or your friends start to get sick, you will be more careful. You will start to behave differently,” Vespignani says. “Finally, through equations, through specific mechanisms, we can integrate (the behavioral changes) into the description of the progression of the disease through the population.”

This story is republished courtesy of Northeastern Global News news.northeastern.edu.

Scientists sometimes compare predicting the course of epidemics to forecasting the weather. But there’s a major difference—the impact of human behavior—says Alessandro Vespignani, director of Northeastern University’s Network Science Institute.

Consider what happens during a downpour, he says. “If we all open an umbrella, it will rain anyway.”

“In epidemics, if we all open the umbrella in the sense that we behave differently, the epidemic will spread differently,” Vespignani says. “If we are more risk averse, we might avoid places. We might wash our hands more and so on and so forth.”

That makes modeling the interplay between human behavior and infectious disease transmission one of the remaining key challenges in epidemiology, according to a paper Vespignani and colleagues published in the Proceedings of the National Academy of Sciences.

“It’s very difficult to integrate behavior in the models,” especially since existing behavioral models often lack real-world data calibration, says Vespignani, Northeastern’s Sternberg Family Distinguished Professor.

But now, thanks to what they learned during COVID-19, researchers say they have found a solution.

The pandemic released a global flood of data in terms of traceable illness and death, accompanied by electronic data such as geolocation from mobile phones that indicated changing patterns in daily commutes, Vespignani says.

Being allowed access to such large data sets led the researchers to some surprising discoveries about the best ways to incorporate behavioral changes into models of disease progression, Vespignani says.

“We are really moving the frontier of epidemic and outbreak analytics and forecasting to the next level,” he says.

“All the data accumulated in the past few years and the knowledge is creating an understanding that hopefully will put us in a different place the next time we have to manage an infectious disease threat.”

The study looked at three different behavioral models—one data-driven and two mechanistic—across nine geographic areas during the first wave of COVID to evaluate how well they were able to capture the interplay between disease transmission and behavior.

The mechanistic model, which describes the mechanism of behavioral changes, often has superior or equivalent performance to the data-driven model, which employs mobility data to capture behavioral changes, in coming up with both a short-term forecast and retrospective analysis, Vespignani says.

“In a sense that was a bit of a surprise,” given scientists’ traditional preference for data modeling, he says.

A major advantage of mechanistic models is how they took into consideration that individuals exposed to the news of the pandemic started to change their behavior even before mandates were established, Vespignani says.

And risk aversion grew as COVID spread and more people were infected.

“There is a spontaneous component to what people do that has to be integrated in which we think about the trajectory of the disease,” Vespignani says.

“That opens new scenarios in the way we are going to forecast and analyze infectious diseases in the future when we can finally (put) this behavioral component to work.

“In many cases in the past, we had to work with very limited data sets, generally about the flu. We didn’t have such large-scale data,” he says.

“Now with COVID-19 we have data from across the world at all geographical resolutions, so we can really test the models.”

For the study, researchers incorporated data from departments of health and government in Bogota, Chicago, Jakarta, London, Madrid, New York and Rio de Janeiro, as well as Santiago, Chile, and the Gauteng province in South Africa.

“We have data about deaths. We have data about infections. We have data about hospitalizations,” Vespignani says.

In addition to the health data, the researchers also had unprecedented access to tech company analytics on mobility and consumer behavior, Vespignani says. “During COVID there was an all-hands-on-deck effort and so we finally got data that was not available before,” he says.

In the future, researchers can use the models to incorporate behavior changes into projections not only of pandemics but also of seasonal respiratory illnesses, Vespignani says.

It will help health and government officials develop the best approaches to communicating risk and developing risk-reduction strategies, he says.

“As soon as (disease) incidence grows, and you or your friends start to get sick, you will be more careful. You will start to behave differently,” Vespignani says. “Finally, through equations, through specific mechanisms, we can integrate (the behavioral changes) into the description of the progression of the disease through the population.”

This story is republished courtesy of Northeastern Global News news.northeastern.edu.