How best to predict where coronavirus strikes? ODU forecasters have spent the past year trying.
Since the start of the pandemic, Virginia officials and residents have wanted to know the same things: what will happen next? How bad will COVID-19 cases get in my city? When will this all be over?
Researchers at Old Dominion University’s Virginia Modeling, Analysis and Simulation Center quickly sought to create a model that could answer some of those questions, putting a tool online within weeks of the virus’ arrival that predicts cases in each city or county.
The forecasting has drawn renewed attention from state officials, who are using the center’s predictive muscle to aid the vaccine rollout around Virginia. And the ODU researchers recently published an academic article about their work.
The model online looks at the number of cases that have emerged in each city or county in recent days to predict how the next week will look. In Norfolk, for example, the tool predicts there will be 230 more cases over the next week.
The tool also estimates the number of individuals exposed each day who eventually tested positive.
About a month ago, the state asked the ODU team to help with modeling related to its vaccination efforts, said Maria Reppas, a spokesperson for the Virginia Department of Health. They developed a 21-day forecasting model for vaccine shipments and administration based on the state’s historic data, she said in an email.
The state then asked the center to compare preregistration and census population data to recommend how many doses to give out per day at its new community vaccination clinics “that will best fit the community being targeted.” The forecasters also help with analysis on how to best invite people to clinics and community outreach efforts.
Lead project scientist Christopher Lynch and research assistant professor Ross Gore published a study last week in the Journal of Medical Internet Research, in which they compared their forecasting model with six other prediction methods.
The study took Lynch and Gore deeper into the complex world of forecasting.
To assess which prediction method works best, the pair homed in on the stretch between March 7 and April 22, 2020. Because it was the start of the pandemic, it saw a rapid growth in cases. They ran through the data using all seven methods, looking at how closely predictions matched the outcomes.
One challenge: there’s a sort of chicken-and-egg problem.
If you predict there will be an increase, and then people change their behavior to stave off such an increase, was the model effective?
“We know that in the long term we’re fighting what those forecasts are showing because we don’t want it to be as bad as any of the initial predictions,” Lynch said.
In the end, a different prediction model fared better than the one used for ODU’s online tool.
But the researchers have not changed the method behind it because, for their purposes, the other strategy wouldn’t make enough of a difference.
They have added sections, however, including the forecasts for vaccines administered and vaccine shipments.
That could help direct someone looking for a vaccine to perhaps assess where the most doses will soon be available, Lynch said.
Despite a year’s worth of research, the forecasters know that predicting the future, especially in times of crisis, is still just a best guess.
“There are no crystal balls out there,” Gore said.
Katherine Hafner, 757-222-5208, katherine.hafner@pilotonline.com
