Research Briefs

Predicting flu season

New techniques help anticipate peak periods

Predicting flu season: Photo of people wearing face masks

People wear face masks in Mexico during a 2009 outbreak of the flu. Scientists have created a pilot system to forecast flu outbreaks. (Photo by Henry Merino, Wikimedia Commons.)

December 20, 2013 | Using methods from weather and climate forecasting, researchers have developed a forecasting system for the flu that can incorporate real-time data to predict the week that influenza levels will peak. A research team that included NCAR’s Alicia Karspeck made 12 weekly forecasts for 108 U.S. cities during the 2012-13 flu season, and found the system accurately predicted the outbreak’s peak for 60 percent of the cities weeks in advance.

The researchers, led by Jeffrey Shaman of Columbia University, combined a model of disease transmission with data from Google Flu Trends, which tracks internet searches on flu-related terms, and with data from the Centers for Disease Control, which tracks the number of people testing positive for flu.

Karspeck focused on advanced data assimilation techniques, which combined observations with computer simulations to estimate the current state of flu cases. This enabled the researchers to tune the model with real-time data as the season progressed.

The study was published earlier this month in Nature Communications.

Karspeck’s usual work is with the data assimilation system for the ocean component of the Community Earth System Model, a global climate model that incorporates data about the atmosphere, land surface, ocean and sea ice to make long-term climate predictions. But five years ago, while at a conference, she had a discussion about forecasting with a former graduate school classmate who was creating a model of influenza outbreaks.

“It’s a sort of technology transfer, in the sense that we have a very strong group here at NCAR that works in data assimilation to support forecast initialization,” Karspeck said. “We took the same technology and methodologies and crossed disciplines.”

Data assimilation allows the model to reflect actual influenza conditions in a population, she said, such as the number of susceptible people, the number of people who already have the flu, and those who have some immunity to it. Data assimilation was also used to tune the model in real time, so that it could better reflect population and virus characteristics that are relevant to the evolving outbreak. 

The researchers believe that this leads to more accurate predictions as the season progresses.

The number of people who have the flu usually peaks sometime between December and April. The researchers will test the model again this winter and plan to validate the forecasts after the season ends.

Having advance notice of when the number of people infected will rapidly increase could help reduce the severity of flu outbreaks by motivating more people to be vaccinated in advance, or giving officials the option to close schools or cancel events that could facilitate flu transmission. Influenza is associated with the deaths of anywhere from 3,000 to 49,000 people each year in the United States.

Jeffrey Shaman, Alicia Karspeck, Wan Yang, James Tamerius, Marc Lipsitch, Real-time influenza forecasts during the 2012–2013 season, Nature Communications, 4, December 3, 2013. doi:10.1038/ncomms3837