How Sickweather predicted an early flu season 6 weeks ahead of the CDC [VIDEO]

“This is the earliest regular flu season we’ve had in nearly a decade, since the 2003-2004 flu season,” said Thomas Frieden, director of the Centers for Disease Control and Prevention, to the Daily Mail.
A health emergency was declared in Boston on Wednesday after 700 confirmed cases of the flu were reported. Of the 123 million doses of flu vaccine the CDC has prepared for this year’s flu season, 112 million have already been administered to people. Forty-one states are now reporting cases of the flu, which has afflicted 10 times as many people as it did in 2012.
And Sickweather, the small startup based in Baltimore County, saw this coming as early as Oct. 18 when it tweeted: “Oh, hello #Flu, you’re a little early this year.”

Sickweather, which tracks the spread of a variety of symptoms and illnesses — including the flu — across the U.S. by mining information from people’s Facebook and Twitter updates, first noticed a spike in its own data in early October.
“[There was an] approximately 77 percent increase from August to early October [in flu updates], which included reports of users saying that they had been diagnosed by a doctor with the flu,” said Sickweather CEO Graham Dodge in an e-mail.
The next steps Sickweather took prepared it to send that tweet about flu season arriving early:

• Sickweather compared its flu-related data to flu data compiled by Google Flu Trends and CDC flu data from 2011.
• “[We] saw that our data showed an overall 30 percent increase in activity compared to the previous year [2011],” said Dodge.

Dodge said the team “had no idea” it would take the CDC six more weeks to make a similar observation about how early the 2012-2013 flu season was to arrive. The CDC made that announcement Dec. 3, writing on its website, “Significant increases in flu activity in the United States have occurred in the last two weeks, indicating that an early flu season is upon us.”
Watch an animation showing the spread of flu across the U.S., as represented by Sickweather data:
[youtube https://www.youtube.com/watch?v=UZgOr2KLzt0]
Of course, this isn’t a shortcoming of the CDC so much as it is a bold step forward in public health informatics, a term used in an academic paper titled “You Are What You Tweet: Analyzing Twitter for Public Health” and co-authored by Johns Hopkins University computer science professor Mark Dredze and Michael Paul, a second-year Ph.D. candidate in computer science at Hopkins. Not coincidentally, Dredze and Paul are two of Sickweather’s three advisers.
One finding in Dredze and Paul’s paper is that a damn good correlation exists when it comes to comparing the accuracy of flu-reporting via Twitter feed and flu-reporting as confirmed by the CDC. (See graph below, or page 4 of the paper.)

 
Typically a delay of two weeks takes place in between seeing flu-related data in real-time and the CDC processing and reviewing flu reports from hospitals and doctors, said Dodge.
“The ‘two-week gap’ is just something us and others looking at social media have noticed when looking at CDC’s historical data compared to social media reports,” he said.
So, by tracking flu data in real-time via social media, Sickweather can issue a judgment about an early flu season weeks in advance.
Granted, it’s an imperfect science compared with waiting for hospitals and doctors to send confirmed flu reports. For example, on Twitter, people might be of the opinion they have the flu without knowing for certain, and will tweet so regardless.
Nonetheless, social data this year presented a convincing warning about an early, widespread flu season.
And if that means social media has a use beyond just mindless banter about Scotch and Earl Grey tea, well: we’ll drink (a glass of orange juice) to that.