From extreme weather to another wave of COVID-19, forecasts give decision makers valuable time to prepare. However, when it comes to COVID, long-term predictions are challenging because they involve human behavior.
While it can sometimes seem like there is no logic to human behavior, new research is working to improve COVID forecasts by incorporating this behavior into predictive models.
Allied Health researcher Ran Xu of the UConn College of Agriculture, Health and Natural Resources, along with his collaborators Hazhir Rahmandad of the Massachusetts Institute of Technology and Navid Ghaffarzadegan of Virginia Tech, published a paper PLOS Computational Biology In it, they describe how they applied relatively simple but nuanced variables to improve modeling skills, with the result that their approach outperformed a majority of the models currently used to make decisions recommended by the federal centers for the Disease Control and Prevention (CDC) were taken.
Xu explains that he and his collaborators are methodologists and were interested in studying what parameters affect the predictive accuracy of the COVID prediction models. First, they turned to the CDC Prediction Hub, which serves as a repository of models from across the United States.
“Currently, there are over 70 different models, mostly from universities and some from companies, which are updated weekly,” says Xu. “Each week, these models give forecasts for cases and the number of deaths over the next few weeks. The CDC uses this information to make their decisions, such as where to strategically focus their efforts or whether to advise people to social distance. “
The human factor
The data was a culmination of over 490,000 point forecasts for weekly deaths at 57 US locations over the course of a year. The researchers analyzed the duration of the prediction and how relatively accurate the predictions were over a 14-week period. Analyzing further, Xu says they noticed something interesting as they categorized the models based on their methods:
“With purely data-driven models, such as machine learning and curve-fitting models, we found that they are better at predicting in the short-term, while theory-driven models are better at predicting longer-term.”
This may seem strange at first, but Xu explains that the difference is due to human behavior.
“It’s weird and not weird per se,” says Xu. “Of course, if you don’t have theory and the models just work with a lot of data and machine learning, they will do a good job in the short term. But what really matters in the medium to long term is you need a theory that explains why people do the things they do.”
Including the behavioral component in the model was relatively easy, says Xu.
“As we looked at all of these 60-70 models, we felt an important behavioral mechanism was missing. This mechanism is when people see more deaths or perceive COVID infections as dangerous, then voluntarily limit their mobility or engage in social distancing. However, once the mortality rate drops, people return to their normal activity. Looking through the models, few of them model this endogenous feedback loop.”
insights for the future
The researchers argue that this feedback loop, while largely overlooked by other models, offers the greatest utility for mid- to long-term predictions.
According to Rahmandad, the research suggests that the key to modeling for long-term predictions is not necessarily building more complicated models, but strategically incorporating the right elements.
“To create a predictive model that is successful in the long term, we can start small with a simple, mechanistic model,” says Rahmandad. “We can then include important mechanistic features – in particular the endogenous representation of human behavior in interaction with the evolving pandemic.”
Examining the models on the CDC hub, some have behavioral components, Xu says, but few consider how they change over time or as the disease progresses.
“I think incorporating behaviors into infectious disease modeling and developing relevant behavioral theories is still an area that needs more research, and we don’t currently have comprehensive theories that explain how people behave during pandemic/infectious disease outbreaks behave,” says Xu. “This requires the collaboration of several disciplines such as social scientists, epidemiologists and methodologists.”
After including the feedback loop, the researchers found that the model works very well in predicting the course of COVID, and Xu stresses that this shows the importance of including behavioral dynamics in infectious disease modeling.
“The purpose of developing this model is not to provide real-time predictions, but it can provide insights for future predictive models. This simple model performs even better, especially over the longer term.”