CAUTION, RI [Brown University] – By adding behavioral components to an infectious disease model, Brown University researchers developed a new modeling approach that captures the ups and downs in new COVID-19 cases that have occurred over the past 16 months.
The approach published in the journal Scientific reports, could be useful for predicting the future trends of the current pandemic and the course of future developments.
“We know that human behavior is important in the spread of infection,” said Vikas Srivastava, Brown assistant professor of engineering and lead researcher. “We wanted to see if we could quantify these behavioral aspects, incorporate them into a model, and see if that model was able to capture infection rates that we’ve seen in the US and elsewhere.”
A frequently used approach to modeling the transmission of infectious diseases is the so-called SIR model. The approach divides a population into different categories: vulnerable, infected, and recovered. The model moves people from one category to another based on two parameters. The communicability of the disease, along with the frequency with which people come into contact with one another, predicts how quickly people will go from susceptible to infection. The recovery rate moves people from infected to recovered. (“Recovered” in these models generally means “no longer contagious,” which includes those who died from the infection.)
The standard SIR model produces a curve with a single peak – the one that infectious disease experts asked to flatten through social distancing, masks, and other measures that reduce virus transmission. But in the past 16 months, actual case numbers in individual states, in the US as a whole, and in other nations have not turned up a single curve. Instead, they generated multiple waves of infection that posed significant challenges to the infectious disease modeling community, says Srivastava.
As the pandemic spread, Srivastava taught a class that included a section on infectious disease modeling at Brown’s School of Engineering. He and his students were surprised to see the discrepancy between model predictions and actual case numbers.
“We saw cases go up and down and create multiple peaks, but the models weren’t able to capture that,” said Srivastava. “It made us think about the behavior and reaction of the population in order to explain and predict what would happen.”
Srivastava worked with two Brown students, Zachary LaJoie and Thomas Usherwood, on developing the new modeling strategy. They modified a standard SIR model to include the effects of vaccination and also added two behavioral parameters to the model. The first “level of caution” estimates people’s tendency towards safe behavior – social distancing, wearing masks, and other safety measures – as reported cases increase. The parameter also captures government measures in response to increasing case numbers such as closures and quarantines that increase safe behavior. A second parameter, the “feeling of security”, models people’s confidence in returning to pre-pandemic activities when more people are vaccinated.
The team then used an optimization algorithm to calibrate the values for the new parameters based on the case numbers reported in the US and cities.
“For example, when we looked at New York City, we saw our levels of caution rise around the same time that government measures went into effect in late March,” LaJoie said. “Then as the cases went down later, we saw the caution decrease and there was another surge in cases going on vacation.”
If the model is properly adapted to the data, it will provide insight into how the pandemic might develop in the future. For example, the team was able to measure how different rates of vaccine intake could affect the number of cases. Rising vaccination rates could lead to a decrease in cases, but it could also reduce precautionary measures and make unvaccinated people feel more secure. This could put upward pressure on case numbers even if vaccines drag them down. In fact, the model predicts scenarios in which infection rates will briefly rise when vaccines are introduced, before they eventually decline again.
In the US, for example, the model captures a brief period of increasing infections in mid-April before rates began to decline again. Larger increases in places like India are similar to the more extreme increases after vaccination that the model predicts. At current vaccination rates, the model predicts that cases in the United States will be close to zero by August 2021.
Findings like this, the researchers say, could be useful where vaccination programs are just getting started.
“When we developed the model, we focused on the US, but it would definitely be useful for making predictions in other places like India, Europe or South America, where the case numbers are still pretty high,” said LaJoie.
The modeling approach could also be applied to future outbreaks or pandemics.
“There’s really nothing in this model that restricts it to just COVID-19 as a disease,” Usherwood said. “We think this is applicable to any situation where people’s behavior is important, which is basically any infectious disease.”