Photo by SAEED KHAN/AFP/Getty Images
That “viral” metaphor for social media just got a little more bona fide. According to a recent slate of independent studies, Twitter can accurately track the spread of a virus or disease — and do it much faster than traditional surveillance methods.
From Iowa to Brazil, researchers are discovering there is a distinct association between complaints, worries and random rants on the social media site and the spread of medical issues as wide-ranging as the flu, dengue fever and pollen-induced allergies.
“By looking at the Twitter stream, we were able to track the public concern in real time about vaccination issues, travel issues and responses to public health,” said Dr. Philip Polgreen, an associate professor at the University of Iowa Carver College of Medicine who used Twitter to track the progression of the H1N1 influenza outbreak in 2009.
“But that not only helps us track the progression of something like the flu, it can provide a way of determining the effectiveness of communication about public health and what messages should be reinforced.”
Researchers from the University of Iowa discovered the association for influenza by following Twitter keywords commonly linked with H1N1, such as “swine flu” and “influenza.” The team began collecting the messages in April 2009, shortly after the first wave of H1N1 struck the United States. Not only did they find that tweets from people experiencing flu-like symptoms tracked closely with the information collected by the Centers for Disease Control, they also discovered they were highly accurate in terms of both time and location. The CDC results were much slower — arriving two to three weeks after the patients began feeling sick.
“From a clinical standpoint, data is one of the cornerstones of public health and it’s important to have that data in a timely fashion,” Polgreen said. “If we can get information faster, we can perhaps intervene sooner.”
Previous studies have shown that search engines like Google and Yahoo can be effective at correlating outbreaks with key search phrases like “symptoms of the flu,” but the Twitter method excites researchers more because it provides more context.
Twitter users often don’t shy away from complaining about their exact ailments and when they developed. Their profiles often list their location, and increasingly, other users can pinpoint their whereabouts even more precisely thanks to GPS devices. That kind of rich data can help health professionals find the epicenter of the outbreak, understand how it’s being passed from person to person and estimate how quickly it will spread to other parts of the country.
The researchers themselves underestimated the potential depth of the treasure trove the social network could provide when they began the study.
“It’s easy to dismiss Twitter,” said Alberto Segre, an author of the report and a computer scientist at the University of Iowa. “It’s like, ‘Who’s going to tweet about the flu?’ I personally don’t see what’s in it for the person who’s tweeting. I was as surprised as the next guy that there actually was good information. It can be very noisy, but there’s still a signal.”
In a separate study, Mark Dredze and Michael J. Paul, computer scientists at the Center for Language and Speech Processing at Johns Hopkins University, came to a strikingly similar conclusion after analyzing 1.6 million tweets related to 15 different health conditions.
Using a new algorithm, they trained their computer to sift through more than 2 billion public tweets that were posted between May 2009 and October 2010. They separated ones complaining of legitimate ailments – such as “I just want to be able to drink water – #stupidstomach #flu” – from the ones proclaiming, “I’ve got Justin Bieber fever!”
Their study found that the national flu rate calculated through Twitter has a 96 percent correlation with the rate as reported by the CDC. The tweets also revealed allergy patterns, cancer rates, self-medication behavior and over-the-counter drug misuse.
But the data have their limits.
“It’s not accurate enough to replace traditional methods,” Paul said. “We won’t be able to use this to determine the exact percentage of people who have the flu, but we can use it to see the flu rate is going up suddenly and we should investigate this. There’s a lot of potential to learn so much about people that they don’t necessarily share with their doctors.”
And as Twitter spreads throughout the globe, that potential will likely grow. In fact, software created recently by a computer scientist at Brazil’s Federal University of Minas Gerais has been used to identify a high correlation between official dengue fever statistics and the time and place Brazilians send a tweet saying they’ve contracted the virus.
Polgreen of the University of Iowa finds the Brazil report among the most interesting – perhaps because it could be a game-changer for worldwide disease control — especially in developing countries.
“Social media is emerging as an important source of information in Middle East politics, in entertainment, in science,” Polgreen said. “Why not for health?”