Philip Polgreen, M.D., an associate professor in the Department of Medicine at the University of Iowa Carver College of Medicine, and his colleagues published a study in the May 4, 2011 issue of the scientific journal PLoS ONE showing that Twitter can be used to track influenza activity. Support for the research was provided in part by the Robert Wood Johnson Foundation’s Pioneer Portfolio. Previous work by Polgreen and Forrest Nelson, Ph.D., an economics professor also at the University of Iowa, includes the development of an electronic prediction market that could help public health officials forecast the timing, severity and spread of seasonal influenza and other infectious diseases. In this post, Polgreen answers some questions about the current study’s findings and implications for the future
Q: Why did you look at Twitter and influenza activity?
A: Right now, public health officials report suspected cases of the flu to the Centers for Disease Control and Prevention (CDC), but that process can take several weeks, a lag that gives the flu an opportunity to spread. When reports of the H1N1 virus that causes swine flu started increasing, we wondered if we could tap into the Twitter stream to find evidence of an upswing in cases of the flu in real time. Twitter is a micro-blogging service that allows millions of users to send and read “tweets” on all kinds of information, including, as it turns out, useful information about people suffering from fever and other flu symptoms.
Q: Your team also found that Twitter could be used to track the rapidly evolving public sentiment with respect to H1N1. How did the team come to that conclusion?
A: We collected and stored tweets containing key words such as “H1N1”, “influenza”, or “swine flu” that were sent at the start of the outbreak. At the start of the outbreak in April 2009, we saw a flurry of tweets, including some expressing fear about the virus. But as time went on, public health messages indicating that the H1N1 virus was not as deadly as expected kicked in and we saw a gradual decline of tweets talking about such concerns.
Q: The study also found that public interest in hand-washing seemed tied to public health messages aimed at slowing the spread of the flu. Can you explain why that finding is important?
A: If this method proves accurate, public health officials may one day use it to find out whether people understand key messages aimed at flu control and prevention. If not, they can tailor the messages to increase the knowledge of, say, the importance of hand-washing, a habit that can protect people from the flu and can contain its spread.
Q: You also did a second analysis that used Twitter to track disease activity. Can you explain what your team found?
A: We analyzed tweets that contained the words “fever”, “flu”, “muscle aches” and other symptoms, finding that Twitter data could be used to estimate incidence of the flu in real time. In addition, we found that Tweets from people experiencing flu symptoms tracked closely with the information collected by the CDC (data that comes out two to three weeks after people report feeling sick) in both time and location. If this method of tracking disease is confirmed by additional research, public health officials could use it as an early warning of a potential uptick in flu cases in a specific geographic area.
Q: Why is early warning a critical part of protecting the public?
A: Public health officials need as much information as they can gather in order to combat the flu and other infectious diseases. With early warning that a flu strain is particularly virulent in some part of the country or is spreading rapidly in others, public health officials can ramp up production of a vaccine or push out public health messages urging people to line up for a flu shot, which can turn down the dial on the outbreak.
Q: Does this have the potential to accelerate the progress we’ve made in protecting people from the flu or other infectious diseases?
A: Yes. This is one of the first large-scale efforts to investigate if data from Twitter can be used to predict the flu or to track public interest in a disease like H1N1 influenza. Additional research will need to confirm the accuracy of this method and extend it to other infectious diseases. But if all goes well, public health officials may one day be able to tap into the Twitter stream to get real-time information that could pinpoint an outbreak’s location and show when it is starting to spread to other areas of the country. This method will not replace current disease-tracking methods but could augment existing approaches to surveillance.