A tool that Virginia Tech scientists created to predict when and how hard the flu will strike a particular area will become part of AccuWeather’s forecast early next year.
The Biocomplexity Institute of Virginia Tech announced Wednesday that it will partner with AccuWeather to implement the tool, which predicts where the flu will hit during the upcoming month and how many people it might sicken in a particular town.
The forecast will be updated weekly by using confirmed cases reported to the Centers for Disease Control, combined with data that model how people live, travel and commute; posts to social media made by sick people; and Google searches of flu symptoms and treatments.
Bryan Lewis, a research associate professor, and Zane Reynolds, an advanced software innovation architect, said their forecasts are about 60 percent to 70 percent accurate and are getting better with each passing year.
They began forecasting flu a few years ago as part of a CDC challenge. The Tech team uses the institute’s disease surveillance platform, EpiCaster.
The tool was first used in 2014 when the Department of Defense turned to Tech for help in tracking the course of the Ebola outbreak in Africa.
EpiCaster also is used for other disease surveillance, mostly of mosquito-borne viruses.
“It’s been our intent to share the forecast and share the technology and methods we are employing,” Lewis said. “You often don’t experience that in the academic setting, where it is publish or perish. This research institute is a hybrid model that we like to deliver things to the public and serve the public as well as contribute to the scientific community, and then also maybe serve some commercial interests, which also benefit the public.”
Tech will share its forecasts with AccuWeather and potentially receive revenue from website impressions and advertising built around the forecasts. Lewis and Reynolds said they weren’t familiar with the exact arrangements, and AccuWeather declined to disclose the terms.
Talks began in early 2017 after AccuWeather founder Joel Myers met Tech professor Madhav Marathe at an innovation and disruptive technology conference.
“We had been looking to improve our flu-forecasting solution around the same time we met Madhav and the team,” said Scott Sameroff, AccuWeather’s director of strategic business development.
Tech’s tool stood out because it offers county-level forecasts for several weeks in advance, he said.
“We also liked how their model is dynamic so that it incorporates factors such as school closings and vaccine efficacy rates throughout the flu season, which affects risk levels,” he said.
Instead of relying solely on statistics, the institute looks at what causes diseases to spread, similar to methods used by weather forecasters as they look at what causes thunderstorms in order to predict when and where they will happen.
Accuracy improves as more data become available and better methods for sorting and scrubbing the information are developed.
“With weather forecasting, you know if it’s going to rain today, and if so an inch versus a quarter of an inch. In the flu-forecasting world, we’ve gotten it down to where we are better at guessing if there’s going to be an inch or a quarter of an inch, but not down to same point of accuracy,” Lewis said.
Reynolds and Lewis — both of whom have gotten their flu shots — said they are hopeful the information will be used to help keep people from becoming ill.
Though the tool is not yet available through AccuWeather, it will be similar to the EpiCaster on the institute’s website at http://epicaster.vbi.vt.edu/epicaster/epicaster.html, which is showing that the flu is off to an early start in Virginia.