Google Tool Predicts Influenza Activity
'Flu Trends' May Heighten Physician Awareness
By David Mitchell
12/3/2008
Family physicians may wish to add a new resource to their influenza armamentarium: a Web-based tool from online search giant Google. Dubbed "Google Flu Trends," the tool can provide estimates of seasonal influenza activity in geographic areas faster than CDC surveillance reports by examining data from search-engine queries about flu-related topics.
Sources at the CDC, which collaborated with Google in the development of Flu Trends, said the advance notice provided by the online resource should help physicians and public health officials.
"You can look up your local area and see what kind of flu activity might be predicted based upon search patterns of people in that area," said Tony Fiore, M.D., M.P.H., medical officer in the CDC's Influenza Division. "It's kind of an interesting thing. It's not real surveillance data in a sense, but based on looking through the past five years of data, it's pretty predictive of actual surveillance data."
"You can look up your local area and see what kind of flu activity might be predicted based upon search patterns of people in that area," said Tony Fiore, M.D., M.P.H., medical officer in the CDC's Influenza Division. "It's kind of an interesting thing. It's not real surveillance data in a sense, but based on looking through the past five years of data, it's pretty predictive of actual surveillance data."
Value of Early Detection
Lyn Finelli, Dr.P.H., head of the Influenza Division's influenza surveillance team, said Google Flu Trends is an average of a week to 10 days faster than the agency in identifying influenza activity.
Finelli said that early detection could allow more people to get vaccinated before flu activity spikes in a given area. It takes about two weeks for an influenza vaccination to take effect, she said, whereas it can take six to eight weeks for the flu to "sweep through a community, depending on its size."
For example, Finelli said that the 2003-04 influenza season was severe and caught people unprepared because it started early in October.
"Most people get vaccinated in October, November and even some in December," she said. "A heads up would have allowed more people to get vaccinated."
Finelli said that early detection could allow more people to get vaccinated before flu activity spikes in a given area. It takes about two weeks for an influenza vaccination to take effect, she said, whereas it can take six to eight weeks for the flu to "sweep through a community, depending on its size."
For example, Finelli said that the 2003-04 influenza season was severe and caught people unprepared because it started early in October.
"Most people get vaccinated in October, November and even some in December," she said. "A heads up would have allowed more people to get vaccinated."
Useful Despite Limited Specificity
Although not every person searching for flu-related terms on the Internet is sick, that doesn't affect the accuracy of the tool, according to Google spokeswoman Katy Bacon.
"Because the data is based on patterns in hundreds of millions of searches, these individual cases would not throw the model off," she said.
Google Flu Trends could serve a function similar to that performed by the U.S. Influenza Sentinel Provider Surveillance Network, or USISPSN, said Jonathan Temte, M.D., Ph.D., an associate professor in the department of family medicine at the University of Wisconsin, Madison, and a member of the CDC's Advisory Committee on Immunization Practices.
"It is limited -- as is the USISPSN -- by a lack of specificity," Temte said. "Hence, it will serve as a marker of influenza-like illness. That said, I like it."
"Because the data is based on patterns in hundreds of millions of searches, these individual cases would not throw the model off," she said.
Google Flu Trends could serve a function similar to that performed by the U.S. Influenza Sentinel Provider Surveillance Network, or USISPSN, said Jonathan Temte, M.D., Ph.D., an associate professor in the department of family medicine at the University of Wisconsin, Madison, and a member of the CDC's Advisory Committee on Immunization Practices.
"It is limited -- as is the USISPSN -- by a lack of specificity," Temte said. "Hence, it will serve as a marker of influenza-like illness. That said, I like it."
CDC Expects Some Oseltamivir Resistance
The CDC is expecting drug resistance to influenza A (H1N1) in the 2008-09 influenza season "at least similar" to that seen last season, but the agency is not altering its antiviral treatment recommendations, said Tony Fiore, M.D., M.P.H., medical officer in the CDC's Influenza Division
The CDC reported that 11.9 percent of 1,026 H1N1 viruses tested last season were resistant to oseltamivir, which is marketed as Tamiflu. That was an increase from less than 1 percent of 588 viruses tested during the 2006-07 influenza season.
Most U.S. influenza cases seen during the 2007-08 season were caused by influenza A (H3N2) and B viruses, according to CDC reports.
Overall, 2.1 percent of influenza A and B viruses in the United States were resistant to oseltamivir last season. Worldwide, 16 percent of 7,535 viruses tested were resistant to the drug.
"Some of the Southern Hemisphere countries are seeing much higher levels of resistance in H1N1," Fiore said. "South Africa actually had all of their H1N1 resistant, and Australia was in the neighborhood of 30 (percent) to 40 percent. The recommendations have not changed at this point. Oseltamivir is still one of two recommended drugs for use."
Zanamivir, sold as Relenza, is the other drug recommended for antiviral therapy.
As it did during the 2007-08 influenza season, the agency is discouraging use of adamantanes. Nearly 100 percent of H3N2 viruses tested were resistant to these drugs last season.
To stay up-to-date on antiviral resistance during the current influenza season, Fiore said physicians should pay close attention to CDC surveillance reports.
"We'll be carefully monitoring antiviral resistance patterns over the course of the season," he said. "If it's a big H1N1 season, it's possible we may need to change recommendations in mid-course, but there are no changes now."
The CDC reported that 11.9 percent of 1,026 H1N1 viruses tested last season were resistant to oseltamivir, which is marketed as Tamiflu. That was an increase from less than 1 percent of 588 viruses tested during the 2006-07 influenza season.
Most U.S. influenza cases seen during the 2007-08 season were caused by influenza A (H3N2) and B viruses, according to CDC reports.
Overall, 2.1 percent of influenza A and B viruses in the United States were resistant to oseltamivir last season. Worldwide, 16 percent of 7,535 viruses tested were resistant to the drug.
"Some of the Southern Hemisphere countries are seeing much higher levels of resistance in H1N1," Fiore said. "South Africa actually had all of their H1N1 resistant, and Australia was in the neighborhood of 30 (percent) to 40 percent. The recommendations have not changed at this point. Oseltamivir is still one of two recommended drugs for use."
Zanamivir, sold as Relenza, is the other drug recommended for antiviral therapy.
As it did during the 2007-08 influenza season, the agency is discouraging use of adamantanes. Nearly 100 percent of H3N2 viruses tested were resistant to these drugs last season.
To stay up-to-date on antiviral resistance during the current influenza season, Fiore said physicians should pay close attention to CDC surveillance reports.
"We'll be carefully monitoring antiviral resistance patterns over the course of the season," he said. "If it's a big H1N1 season, it's possible we may need to change recommendations in mid-course, but there are no changes now."
Fiore said physicians might make use of Flu Trends by increasing the amount of testing they do for influenza when the tool indicates a high level of flu activity in their area. He said they also could inform patients who were not vaccinated or who did not respond to vaccination that they might be candidates for antiviral treatment.
Family physician Pedro Ballester, M.D., of Warren, Ohio, agreed that the online tool might help doctors with diagnostics because they would be more aware of flu activity in their area. However, he added, some family physicians are playing a smaller role in influenza vaccination than they once did because more patients are going to retail clinics, health departments and senior centers for influenza vaccinations.
Ballester said the tool also might benefit pharmacies, which would be better prepared to have antiviral medications on hand, and hospitals, which might have capacity issues in the case of a seasonal influenza epidemic.
Family physician John Bucek, M.D., of Somerville, N.J., also acknowledged that Flu Trends could heighten physician awareness and offer a simple alternative to communicating with local health departments. He said that an extra week's lead time could "help everyone who has to educate patients about influenza."
However, Bucek said the tool could have a negative aspect.
"People who market and sell products already use this kind of data mining from Internet behavior," he said. "If companies that make flu drugs or symptom meds have the data, they can direct their marketing more precisely to boost their sales. If they know when patients will be the most symptomatic or the most afraid, they can tailor the commercial message better to trigger purchases based on fear more than evidence of efficacy."
Ballester said the tool also might benefit pharmacies, which would be better prepared to have antiviral medications on hand, and hospitals, which might have capacity issues in the case of a seasonal influenza epidemic.
Family physician John Bucek, M.D., of Somerville, N.J., also acknowledged that Flu Trends could heighten physician awareness and offer a simple alternative to communicating with local health departments. He said that an extra week's lead time could "help everyone who has to educate patients about influenza."
However, Bucek said the tool could have a negative aspect.
"People who market and sell products already use this kind of data mining from Internet behavior," he said. "If companies that make flu drugs or symptom meds have the data, they can direct their marketing more precisely to boost their sales. If they know when patients will be the most symptomatic or the most afraid, they can tailor the commercial message better to trigger purchases based on fear more than evidence of efficacy."
