The internet technology and search engine giant Google launched a new tool this week called Flu Trends that the company says accurately estimates the level of flu in each state of the USA in nearly real time by keeping track of certain Google search queries.

Flu Trends is free to all and can be viewed on the internet. Google said traditional flu tracking systems take between 1 and 2 weeks to collect and publish data whereas Google can count searches and release the results in near real time.

During the 2007-2008 flu season Google engineers shared their initial results with epidemiologists at the US Centers for Disease Control and Prevention and together they established that the Google data strongly and consistently matched real CDC flu data for each of the nine surveillance regions of the United States. They also found it was possible to estimate and publish the results two weeks faster than the CDC reports.

The Google and CDC research that went into developing Flu Trends has been accepted in principle for publication in the journal Nature and in the meantime can be read about in early manuscript form via a PDF download that was made available on 11 November.

The head of influenza surveillance at the CDC, Dr Lyn Finelli told Reuters in a telephone interview that Flu Trends:

"Takes Google search terms of influenza-like illness and influenza and it emulates a signal that tells us how much influenza activity there was."

Google and the CDC hope Flu Trends will help serve as an early warning system for flu outbreaks. Flu affects tens of millions and kills hundreds of thousands of people worldwide every year and early detection helps save lives.

Every week, millions of people around the world look for health information online. In America alone, about 90 million people a year look for information on specific diseases on the internet. During the flu season there are more searches for information about flu, as there are more searches about sunburn in the summer and allergies in the allergy season.

When Google and the CDC compared their data, they found there was a close relationship between the number of people searching for flu-related information and the number of people that report having flu symptoms, for instance when they go to their doctor or are admitted into hospital.

Although not every person who looks for flu-related information actually become sick, it is possible to determine the kind of pattern that links the two on a state by state basis. For example, when Google compared search terms entered by users with data from the CDC surveillance system, they found that certain search terms were used more frequently during the height of the flu season than at other times. By counting how often they saw these queries, Google said they could estimate how much flu was circulating in each state of the US.

Finelli said that when they validated the Google model last year it tended to predict the CDC surveillance data.

"They were able to tell us on a day-to-day basis the relative direction of flu activity for a given area. They were about a week ahead of us," she said, according to the Reuters report.

The Google data could be used as an early warning signal for flu activity, said Finelli, giving the CDC time to alert hospitals, clinics and doctors so they can stock up on tests and drugs, not just for flu but co-infections which often sicken the patient more.

Two weeks early warning also gives people greater chance to get vaccinated in time before the outbreak intensifies in their area.

The CDC validated the Google model by aggregating historical logs of online searches entered between 2003 and 2008. Then they worked out the pattern of weekly counts for 50 million of the most common search terms entered by users in the US. They deliberately left out billions of other infrequently used queries.

Using the IP address of each query they knew roughly where that user was, to the nearest major US city, so they were able to produce estimates for each type of query for each week in each city and state. This was then compared to the CDC data for those regions and dates.

Google said no information about the identity of users is retained.

Flu Trends can be viewed free of charge at google/flutrends/.

"Detecting influenza epidemics using search engine query data.
Jeremy Ginsberg ,Matthew H. Mohebbi, Rajan S. Patel, Lynnette Brammer, Mark S. Smolinski and Larry Brilliant.
Pre-publication manuscript, made available 11 November 2008 in Draft form.

Click here for prepublication manuscript (PDF download).

Sources: Google, Reuters.

: Catharine Paddock, PhD