When the first U.S. patients fell victim to the new H1N1 flu, they may not have immediately thought to call their doctor or run to the nearest emergency room. Instead, they probably sat down in front of their computers and Googled “flu symptoms” or “fever” or “chills” or perhaps even “treatment for flu.”
That behavior is exactly what the folks at Google are counting on. Since last fall, the search-engine giant has been nurturing a spin-off service called Google Flu Trends, which aims to identify outbreaks by tracking searches for flu-related terms and provide health officials with early warnings of potential epidemics. The reasoning is that if people are searching for information on the flu, they’re probably sick themselves or know someone who is and a geographic cluster of like-minded Googlers could represent a burgeoning outbreak or, worse, the roots of a new pandemic.
Still, to the extent that data-tracking systems like Google Flu Trends could operate as early-warning networks for infectious diseases, their benefit is that they rely not on hospital data but real-time information from people who are in the process of getting sick. “What we are seeing are trends of what people are thinking about at home, perhaps before they might go to see a doctor,” says Jeremy Ginsberg, lead engineer of Google Flu Trends.
Some public-health experts say this kind of user-fueled data-tracking may start to help government health officials’ efforts to recognize outbreaks. Real-time warnings would allow authorities to stay well ahead of potential pandemics, prepare local populations with appropriate prevention and treatment, and reduce overall illness and deaths. The Google Flu Trends service, which was launched in the U.S. in collaboration with the Centers for Disease Control and Prevention , is now working with Mexican officials to track search trends in that country. The goal is to help authorities discern whether and where the disease is spreading, getting worse or starting to subside.
The World Health Organization uses a similar on-the-ground surveillance strategy a text-mining system of media reports, called the Global Public Health Intelligence Network , designed to suss out mentions of unusual illness. And the CDC has a network of physicians who routinely sample and test patients to see what bugs they have and might be circulating in the community. But these systems have many moving parts, relying on state, local and even community health-care workers to both recognize and report anything out of the ordinary. Once a community doctor sees what he thinks might be an unusual series of flu cases, for example, he would have to alert his local or state health departments, which would then investigate further by testing samples from the sick patients a process that could take up to two weeks.
The folks at Google think that’s too long. Google Flu Trends claims it can pick up signs of health troubles up to two weeks ahead of official health reports, giving communities precious time to protect themselves and hopefully contain the spread of an infectious disease like influenza. Another surveillance company, Veratect, based in Kirkland, Wash., says it picked up the first signs of H1N1 in La Gloria, in Veracruz state, Mexico, as early as April 6, when it received reports of a “strange” respiratory illness there some 18 days before the U.S. Department of Health and Human Services, and the rest of the country, was alerted to the existence of the new virus.
“How useful this [surveillance] is, no one knows for sure,” says Dr. Douglas Owens, professor of medicine at Stanford University. “None of these types of methods has been used under these circumstances directly. But they are certainly worth evaluating.”
Veratect’s tracking method includes analyzing Internet data , maintaining on-the-ground contact with health officials, text-mining news reports and government resources for keywords related to infectious-disease outbreaks and using satellite images of weather patterns to detect and predict the progress of global events like disease and civil unrest. Veratect sources first picked up reports of human respiratory disease at a pig farm in Mexico on April 6; additional reports of a similar illness surfaced on April 16, which is when the company got concerned enough to e-mail officials at the CDC. The CDC was actually already connected to the Veratect news feed when the company again contacted the health agency’s operations staff on April 20, as H1N1 cases began to appear, to reinforce concern over what looked like an emerging epidemic.
“Our system detects something that changes the fabric of everyday life, such as strains in the medical infrastructure or changes in the behavioral patterns of people if they aren’t going to work or attending school,” says Robert Hart, CEO of Veratect. Trained analysts then investigate suspicious reports to determine if additional investigations are needed or alarm bells should be raised. Veratect follows more than 200 diseases around the world in this way, and argues that its system of computer detection and human interpretation can give health officials enough of an early warning of potential disease outbreaks to ready the appropriate medical care.
Compared with Veratect, Google Flu Trends’ metric is somewhat simpler. From 50 million potential search topics, Google engineers narrow down a relevant grouping of flu-related search terms, which they track each fall at the start of the annual flu season. When analyzed side by side with CDC records of confirmed flu cases for the past five flu seasons, Google Flu Trends was 97% to 98% accurate in tracking the disease. And because Google’s analysis is in real time, its estimates of cases come about a week or two before those of the CDC. “Each flu season for the past few years, we see a sharp uptick in flu-related queries at the beginning of flu season,” says Google Flu Trends’ Ginsberg. “One to two weeks later, the official reports come out and show a similar uptick. But because we are able to release estimates in near real time, we can get data out faster while still being accurate.”
“Can you use people going on the Web finding information on flu to predict where flu is taking place” asks Dr. Richard Besser, acting head of the CDC. “Looking back, it is helpful. The question is, Looking forward, can you see that We are open, and continually looking at various approaches for early detection. The sooner we detect a problem, the sooner we can detect and implement protective measures.”
But the predictive power of Google’s system is relatively imprecise, since it depends solely on a large number of people getting sick and hitting their computers. That’s why the H1N1 cases did not pop up as anything unusual in late March and early April. Even today, with more than 400 cases of H1N1 now confirmed in 38 U.S. states, the caseload is too small to register on Google’s radar. It would take thousands, not hundreds, of likely infected people searching for help to distinguish a growing trend from the noise of queries in Google’s database.
Where Google Flu Trends may prove more useful, however, is in the tracking of an epidemic once it is under way. If the current H1N1 outbreak were to worsen and start to spread more quickly, then Google’s system may be able to keep pace with it and alert health officials immediately as the problem grows. “If the disease starts spreading in a particular area, for example, and affects thousands of people, then we hope that our system would detect that within 24 hours,” says Ginsberg. The idea would be to catch the rise in cases before too many people get sick. And that’s what the company is hoping it can do in Mexico: give health officials there a sense of where the cases are, and if the outbreak is expanding.
Dr. Don Weiss, director of surveillance for the Bureau of Communicable Disease in the New York City Department of Health, notes that while systems like Google Flu Trends may be useful, health officials need to remember that the service tracks searches, not confirmed cases of illness or even symptoms that are severe enough to bring a person to the emergency room. Earlier this flu season, for example, when reports of avian influenza overseas hit the news in the U.S., there was a spike in bird flu queries online in New York City. “The system only tells you what people are interested in reading or learning about,” he says.
Still, if public-health officials are to keep ahead of the millions of bacteria and viruses that can potentially make us sick, they’re going to need all the help they can get. And taking their search outside hospitals and doctors’ offices may be an important first step.
See pictures of the swine flu in Mexico.