Early alarms sound online when illnesses go viral

As the Centers for Disease Control and Prevention waited for physicians and others to send data on influenza cases, it monitored Google Flu Trends, developed to determine the level of illness based on how often people used the company's search engine to search for flu-related topics.

Lynette Brammer, MPH, an epidemiologist with the CDC's influenza division, said Google's flu data, which are supposed to be a real-time measure of flu, followed nearly the same trend as the CDC's figures, normally released one or two weeks after it gets reports on flu cases. Both numbers “went up and went down at the same times,” she said.

The promise of Google Flu and other Internet resources, particularly social media sites such as Twitter or other online chatter, has some health experts saying that physicians can use the Web as an early-warning or just-in-time tracking system for outbreaks of not only the flu but also other diseases.

By monitoring such sites, physicians could get a sense of what to expect in their communities in the near future and enable them to prepare better for spikes in disease. Preparations could include temporarily increasing staffing, starting flu testing for patients with influenza-like illness, and ensuring that there are sufficient supplies on hand, such as face masks and vaccine, and plans in place to manage the influx of patients.

For example, Google Flu allows users to customize reported activity by region, state or large city. While the CDC has access to people with influenza only when they seek medical care, online flu monitoring can identify illness among anyone in the community, including people who choose to stay home rather than see a doctor.

“We know flu occurs and starts in different parts of the country in different years,” said William Schaffner, MD, chair of the Dept. of Preventive Medicine at Vanderbilt University School of Medicine in Tennessee. “Social media might give us a week or two advanced notice about the onset of influenza and where it is occurring.”

How accurate is Web activity?

That's not to say there aren't some kinks to work out. Sometimes, activity can be over- or underreported, compared with CDC data, based on patient awareness or media activity. A Feb. 13 article in the journal Nature noted that although Google Flu activity peaked the week of Dec. 29, 2012, just like the CDC, the site over-estimated the number of Americans with the flu compared with CDC data.

Nature, which in Feb. 19, 2009, published Google's own study purporting Google Flu accuracy in prerelease tests (a study co-written by Brammer), also noted that the site, which launched in 2008, undercounted influenza A(H1N1) cases early in the 2009-10 season, before the public became aware of what was commonly called swine flu. “People's behaviors and their searches and what they talk about [online] might change depending on what is in the news a lot,” Brammer said.

Google, in a statement, vouched for its system, and said it is continually tweaking its algorithms to increase accuracy. Google Flu was developed to be a complementary data source to traditional influenza surveillance methods, said Kelly Mason, a Google spokeswoman.

A key question is “what sort of predictions matter,” said Mark Dredze, PhD, an assistant research professor in the Dept. of Computer Science at Johns Hopkins University in Baltimore. He designed a method of tracking flu cases using Twitter.

“Did Google's [over]estimate change how we responded” to the flu season? Dredze asked. “If the answer's no, then it doesn't matter.”

Going beyond the flu

Still, as technology advances, some researchers think these online services could be used to monitor other infectious diseases and coordinate disaster responses. Google doesn't just track flu; it has expanded into providing information on worldwide searches related to dengue fever.

The potential medical and public health benefits of these online services extend beyond tracking influenza activity to detecting upticks of less common diseases in the community, such as meningitis, which can be challenging to diagnose, said Clark Freifeld, a research software developer at Boston Children's Hospital.

“If a doctor knows there has been a lot of bacterial meningitis in the community, that might help inform his diagnosis when seeing a patient,” said Freifeld, who is co-founder of HealthMap, a real-time surveillance map of worldwide disease outbreaks, including influenza.

Online and mobile services, including Twitter and text messaging, could be critical tools for disaster response, particularly in situations where communication can be a challenge. Dredze points to the 2010 Haiti earthquake, where there were few well-organized response teams on the ground immediately after the disaster.

Haitians sent text messages to family and friends, which were passed on to responders, about possible locations of people buried beneath rubble, according to a report presented in April 2012 at the World Wide Web Conference in Lyon, France. The annual international event provides a forum to discuss the evolution of the Web and the impact of its associated technologies on society and culture.

“There are many instances like this where Twitter is going to introduce capabilities that we didn't think were possible,” Dredze said.

But before that happens, more research is needed to determine, for instance, what diseases can be monitored most accurately by the various online services and whether there are certain periods in the year when Web surveillance tools are more reliable, Dredze said.

“The formal public health structure is at the moment fascinated by social media [and other online services] and what they might be able to do,” Dr. Schaffner said.