Seasonal flu epidemics account for as many as half a million deaths worldwide each year. And the rapid spread of new strains can cause many more (the 2009 H1N1 flu pandemic alone killed more than 16,000 people, according to the World Health Organization). Quickly detecting a regional rise in flu-like symptoms such as coughs, sore throats or high fevers can help public health officials take steps to dampen the impact. However, it can take days—even weeks—for trends spotted in clinics to be reported more broadly.

Before visiting a clinic, many flu sufferers visit Web sites for information about symptoms and remedies—a tendency that Google engineers took advantage of to create a real-time flu tracker called ‘Google Flu Trends.’

"We're constantly looking for trends in our search results, and ways to use them to do something useful," says Jamie Yood, communications lead for Google Flu Trends.

By comparing the popularity of the 50 million most common Google search queries in the U.S. with flu-like illness rates measured by the U.S. Centers for Disease Control and Prevention's (CDC) national surveillance program, the Flu Trends team narrowed down the pool to 45 search terms (relating to symptoms, complications and remedies) that correlated with the agency's data on the prevalence of flu symptoms. "It shows the trends—whether flu-like illness rates are going up or down in certain regions," says Yood.

"Oscar nominations" and "March madness" queries also peaked during flu season, but didn't make the cut, Yood jokes.

The CDC's national surveillance program is based on weekly reports from 3,000 health clinics that count the number of patients with a fever and a cough or a sore throat. But it takes up to two weeks for these numbers to be compiled into meaningful and publicly available information about flu trends, Yood says. Google Flu Trends is updated daily, and according to data from the 2007–2008 flu season, it can bridge the CDC's two-week lag, potentially buying officials critical extra time to devise a public health response and curtail the virus's spread.

Google collaborated with the CDC to validate the flu tracker. "They were really excited about the idea of having another source of information that could help with early detection," Yood says of the agency. Together, Google and the CDC published their search query–based flu-tracking model in Nature in February 2009.

But Google Flu Trends doesn't replace the CDC's national surveillance program, Yood stresses. Although it matches CDC-detected trends to within 95 percent, it's less accurate at estimating actual rates of laboratory-confirmed flu, according to a study presented May 17 at the American Thoracic Society 2010 International Conference in New Orleans.

Focusing on flu outbreaks in the U.S. between 2003 and 2008, Justin Ortiz from the Division of Pulmonary and Critical Care Medicine at the University of Washington and colleagues from CDC and PATH, a Seattle-based nonprofit health organization, found that Google Flu Trends deviated greatest from CDC surveillance figures for laboratory-confirmed flu rates during the 2003–2004 flu season, which saw a high number of flu-related deaths in children and, as a result, was a hot topic in the media. During periods of intense media interest or unexpected flu activity, Google Flu Trends might be less accurate in estimating flu rates because of the heightened public interest and increased search activity, the researchers concluded.

"Still, Google Flu Trends provides an excellent public health service, because it's fast and cheap and requires very little infrastructure," Ortiz says.

Google Flu Trends was launched in the U.S. in November 2008. Here, it tracks flu-like illness at the state level, but work is under way to narrow monitoring down to individual cities. The effort has since expanded to track flu-like illness in 20 countries, with results in 38 languages.