To advance the science of forecasting infectious diseases, the Centers for Disease Control and Prevention has been hosting a competition in which researchers use social media, as well as data from CDC’s routine flu surveillance systems, to predict the timing, peak, and intensity of the flu season.
For this flu season’s competition, University of Massachusetts Amherst biostatistician Nicholas Reich joined forces with Carnegie Mellon University, Columbia University, and a group at Los Alamos National Laboratory in a group they named the FluSight Network.
“What makes the flu hard to predict is that there are not a lot of early indications about how the season is going to go,” Reich said. “We’re still trying to figure out what those early indicators are of a bad season.”
A current assessment of the flu, and an ability to look to the future, can be invaluable for the CDC, Reich said.
“Being able to anticipate those events gives them a head start on putting into place the risk-communication efforts to encourage people to get the vaccine, perhaps earlier than usual,” he said.
Each of the member groups in the FluSight Network has competed individually before this year. Reich estimated that the team has a combined 10 years of forecasting experience.
The group’s strategy is to submit a single model to the CDC that combines all of their models together, Reich said. To create these models, the network studies climate data as well as recent digital trends from Twitter, Google, and Wikipedia.
He said researchers primarily focus on those specific sites to look for trends in searches on flu-related terms.
“We saw this as this great opportunity to work together and have the best pieces of our best models all attribute to one forecast,” he said. “It’s sort of like a symphony. Any one of these models is playing their own instrument, but they work better together and they sound better together.”
About 20 teams are participating in the CDC’s challenge this year. The nationwide competition began in early November and is expected to end some time in April.
The teams measure flu trends weekly by attempting to predict the percentage of doctor’s office visits linked to flu-like illness. Each week, they send their updated forecasts to the CDC before they’re then posted online at predict.phiresearchlab.org.
Reich said it has been an abnormally early flu season but points out that it’s a little too early to predict its severity in New England.
But, he added, his group’s model has predicted a 50 percent chance that the season will peak before mid-January.
“Based on what we are seeing in New England so far this year, there is a small signal suggesting this season will peak earlier than usual, although we still have a fair amount of uncertainty about exactly when the peak of flu season will be,” he said.Sophia Eppolito can be reached at email@example.com. Follow her on Twitter @SophiaEppolito.