By the time public health alerts are issued and the media picks up the story, people have already been talking about an illness for weeks on social media. A new Web application, MappyHealth, uses Twitter to monitor and produce automatic reports on what people are talking about most. It's latest tool in the disease surveillance arsenal.
Twitter can provide indications of a potential disease outbreak faster than traditional disease surveillance activities. By the time public health alerts are issued and the media picks up the story, people have already been talking about an illness for weeks on social media, the Department of Health and Human services says.
A new Web application, MappyHealth , uses Twitter to monitor and produce automatic reports on what illnesses people are talking about the most. It's latest weapon in the disease surveillance arsenal.
Local public health operations recognized the potential social media could have in monitoring illness in communities , “but lacked the time and resources to make it a useful data source,” said Diana Kushner, who managed the project for HHS’s Office of the Assistant Secretary for Preparedness and Response (ASPR).
ASPR held a contest for developers to build an app that could deliver a list of the top-five trending illnesses from a specified region in a 24-hour time period. Of 33 submissions, MappyHealth came out on top.
“Having real-time information available in the public domain through social media like Twitter could be revolutionary for health officials watching out for the first clues to new, emerging infectious diseases in our communities and for modernizing our public health system,” said Dr. Nicole Lurie, a rear admiral in the U.S. Public Health Service, in a press release .
Similar Web-based apps can examine a disease outbreak after the fact, but ASPR wanted something that could follow trends in real time. Both the 2009 H1N1 pandemic and the Haiti cholera outbreak showed that social media trends can indicate disease outbreaks earlier than conventional surveillance methods, like emergency room reporting.
Early identification can help minimize the spread of disease so the hope is that local and regional health officials use MappyHealth to cross reference Twitter data with conventional surveillance methods.
MappyHealth won the challenge because of its relative accuracy when filtering by keyword (based on a set taxonomy of terms to search that was based on common terms the public uses to refer to different illnesses) and its ability to filter tweets by location. ASPR announced the official launch of the app last Thursday.
MappyHealth today shows that mentions of the common cold on Twitter are up 69 percent in the last 24 hours. Mentions of influenza are up 25 percent and there’s been a spike in talk about STD’s around Washington, D.C.
Last year, I talked to Jennifer Olsen, Fusion Cell Branch Chief at ASPR as the agency was taking its first look at the value of monitoring twitter.
Back then, using Twitter to monitor public health was in its beginning stages. There wasn’t a set taxonomy of words people use to refer to their sicknesses and HHS was still trying to find the best combination of words that would capture ways people were talking about illnesses. ASPR was using mentions of the flu as a baseline.
“We have a list of terms that we’re pulling for Twitter and trying to evolve those,” Olsen said in an interview last October.
Olsen said one of the biggest challenges was being able determine false spikes in mention of illnesses like flu on Twitter. False spikes occur when there is a large amount of mentions that don’t necessarily coincide with increased prevalence of an illness.
“Both Kim Kardashian and Miley Cyrus had the flu the same week in April and it really threw off everything as far as our numbers go,” she said.
People are likely to tweet about the flu when they get the vaccine or when talking about a news article. These mentions contribute to false spikes.
She also noted that people with gastrointestinal illnesses or illnesses related to food poisoning may be more likely to tweet about the food or the restaurant where they ate than the illness.
“This is an area that will continue to evolve for as long as Twitter is used as a data source,” Kushner said. “Some ways to counteract the ‘celebrity effect’ and other false spikes is to use qualifier terms linked to the taxonomy, to eliminate retweets from your numbers, and other smart filtering techniques.” All of the competitors in the challenge did a good job of “filtering through the noise to get to the heart of the data,” she said of the contest.
The greatest successes of Twitter monitoring came during the H1N1 pandemic. ASPR used Twitter to identify school closures by verifying user reports with official sources. “At a federal level, there is no way to see these school closures,” Olsen said in last year’s interview. “What we have seen is that the stream may not always be useful for getting a spike in flu, but it would be useful in getting specific questions related to flu.”
ASPR is looking for a way to combine streams of data from electronic health records, social media, and news to find consistency across all three during a disaster.
“That would give us some understanding of where we can have more confidence in each of those streams,” she said.
photo by shawncampbell/flickr