The Centers for Disease Control and Prevention (CDC) recently announced the winner of the “Predict the Influenza Season Challenge”: Dr. Jeffrey Shaman of Columbia University’s Mailman School of Public Health and his team submitted an algorithm to predict peak flu season using Google Flu Trends and CDC’s Influenza-Like Illness (ILI) data.
The challenge was unique in that it asked participants to use digital data to forecast the start, the peak week, and the intensity of the U.S. flu season—an approach that is in its infancy, according to Matthew Biggerstaff, an epidemiologist in the Influenza Division at CDC, who is part of the team that facilitated the challenge.
“Sort of like how weather forecasting was fifty years ago, flu forecasting is a young science,” he said. “There has been a lot of work done over the past couple of years on influenza forecasting, but never a centralized effort with standardized milestones like in the challenge.”
The challenge was intensive, requiring teams to submit forecasts biweekly to the CDC team from December to March. Sixteen teams began the competition and eleven submitted final solutions at the closing. Teams used various sources for flu data, including public keyword search information from Wikipedia and Google as well as data from social networking sites like Twitter.
Dr. Shaman’s team stood out among the others because it incorporated easily understood measures of how likely the forecasts were to be accurate.
“The winning prediction developed a method that can give you a percentage accuracy for flu prediction like the weather, [but instead of a] 40% chance of rain, [it’s a] 40% chance that flu will peak in a specific week,” Biggerstaff said.
In the long term, Biggerstaff envisions something similar to a “hurricane season prediction” for the flu; something he says is fairly reliable for preparedness purposes. But even within the next few years, the work done for the challenge is valuable. Biggerstaff noted that state and local governments could potentially use Dr. Shaman’s forecasting algorithm to make decisions about staffing, communication packages, and situational awareness. With an accurate season-peak prediction, health officials could increase their flu shot messaging and healthcare organizations could make sure they have enough healthcare staff to handle a potential increase in patients in advance.
While the results are not perfect, the challenge was a great launching point for the field, and Biggerstaff said the CDC plans on continuing to work with all the teams to make flu forecasts more reliable. In fact, he’s already looking towards next steps:
“We hope to write a peer-reviewed article about this and continue to work with the teams to keep this group of researchers motivated and working on this problem.”_For more information on how this challenge can inform your challenge design, see the HHS Idea Lab blog. If you would like more information on challenges and prize competitions, become part of the Federal Challenge and Prize Community listserv. If you are interested in entering a challenge like this one, see the list of government challenges at Challenge.gov._ This article is part of this month’s editorial theme on Social Media. Check out more articles related to this theme.Edit