Is CDC's Google search-term tracking just the beginning?

Following search trends to predict influenza outbreaks could be the first step in using the Internet to react quickly to the spread of many illnesses.

The announcement on Tuesday that the Centers for Disease Control and Prevention had teamed with Google to track possible flu outbreaks based on terms users enter in its search engine is just the beginning for how the agency could use the data to track other illnesses, said a top CDC executive.

Comment on this article in The Forum."I think it could be applied widely," Joe Bresee, chief of epidemiology and prevention in CDC's influenza division, told Nextgov in an interview. "Using this technology on other outbreaks such as food-borne diseases is a great idea. CDC, Google and others are working on these issues. Flu is the first of the line, but lots of other diseases and collaborations are right behind."

Google Flu Trends was developed by Google.org, the philanthropic arm of the search giant. The application tracks the popularity of search terms such as influenza, flu, thermometer, muscle aches and other terms Internet users use to search for information on treating symptoms to determine where cases of influenza might be on the rise.

Google collected data on search terms from the past five years and compared it to surveillance data CDC had accumulated on flu from hospitals, labs and doctors offices. Google identified a correlation between the number of searches for a group of keywords related to the flu in a given area and the number of cases of infection in that area. By monitoring the search terms, Google can identify spikes in cases as much as two weeks before CDC can, because the agency relies on data compiled from thousands of sources, which takes time to find relationships.

Bresee said the flu was a perfect test case for the system because of the large amounts of data available. By comparing CDC's data to Google's search data, researchers found a strong correlation on the national level. They then used data from more than 30 states to validate the model for smaller geographic areas.

"It's exciting because it looks like not only do their curves resemble our curves, but they tend to precede our curves a little bit, which raises the possibility we can use this data to get messages out more quickly," he said.

Having information on the location of outbreaks in advance can be crucial in reducing the impact of a flu outbreak and preventing deaths, said Dr. Phillip M. Polgreen, assistant professor of medicine and epidemiology at the University of Iowa and author of a similar study based on Yahoo's search data.

"It's enough time to make a difference," he said. "There's a vaccine that must be given one to two weeks in advance. We can increase vaccination rates among high-risk people like health care workers and [people] over 65. Similarly, clinicians can make decisions about anti-virals, which must be given in the first 48 hours, that they wouldn't otherwise [be able to do]. In the case of the flu, this sort of information can translate into preventative actions."

Polgreen called the Google program promising, but cautioned that the system would have to be tested thoroughly before relying on it, and even then it shouldn't replace CDC's traditional collection of data from hospitals and other health care groups.

Polgreen and Bresee said the common nature of the flu made it easier to track the data and find a correlation between certain search terms and incidents of the disease. Both were optimistic the same approach could be used for other diseases, but they admitted it would be more difficult for less common symptoms.

"It remains to be seen if this can be applied to other diseases," said Polgreen. "I think the specificity and sensitivity of search-term surveillance will need to be defined and looked at. What will it be able to pick up? How specific? Symptoms can represent a wide range of diseases."

Polgreen added that sickness itself is not the only reason that can drive an individual to search for symptoms. News reports on a disease or articles about a celebrity falling ill can increase searches, which would provide false positives.

One way to get a more accurate picture would be to track visitors to federal Web sites, Polgreen said. Institutions such as CDC and the National Institutes of Health are primary sources of medical information on the Web for most Americans. He said tracking the number of visitors interested in particular topics or diseases on these sites could provide more information about a possible outbreak.

Polgreen also said government should allow other researchers to access the same information to conduct their own studies. "One way to [accelerate the process] is to open up access to this information, the search trends and disease outbreak information," he said. "Allow it to be done by several groups."

Bresee predicted search-term tracking will need more scrutiny before it becomes stable. "It's a new technology and it's exciting, but we don't know whether people's search patterns and behaviors are going to change over time," he said. "The collection of terms now predicting influenza might not be as good in five years. It bears watching."