New research suggests it may be possible to forecast flu
outbreaks in much the same way meteorologists predict
weather, a potential boon for public health officials and
Using real-time U.S. data gathered by Google, along with a
computer model showing how flu spreads, the researchers
offered a system that could generate local forecasts of the
severity and length of a particular flu outbreak.
This kind of forecasting could improve preparation and
management of annual flu outbreaks in the United States, said
Irene Eckstrand of the National Institutes of Health.
Influenza kills 250,000 to 500,000 people each year around
the globe; the U.S. annual flu death toll is 35,000.
If the forecasts are reasonably accurate, they could help
public health officials target vaccines and anti-viral drugs
to areas of greatest need, said study co-author Jeffrey
Shaman of Columbia University's Mailman School of Public
"If you have a six-week forecast with good confidence that
you're going to have an outbreak in New York City and
nothing's going on in L.A., you'd send the vaccines there (to
New York) because there's enough time to distribute them ...
before there's an actual outbreak," Shaman said.
He suggested that flu forecasts might be distributed through
TV weather programming. Individuals then could decide whether
to get the flu vaccine, keep their distance from people who
sneeze or cough and closely monitor symptoms.
This pilot study, published in the journal Proceedings of the
National Academy of Sciences, looked only at the New York
City area, using data from 2003 through 2008.
If all goes well, the system could offer rudimentary
forecasts as soon as next year's flu season.
It might be possible to issue a few flu forecasts this
season, though those would be in "test-case form," Shaman
"We have to try it for other regions, other cities," said
Shaman. "We have to look and see how it worked during the
pandemic years ... we have to see the differences in
performance depending on the aggressiveness of the strain of
The computer program the scientists used is a standard
epidemiological model showing how influenza moves through a
population, from those who are susceptible to flu, to those
who have it, to those who have recovered, said study
co-author Alicia Karspeck of the National Center for
Atmospheric Research in Boulder, Colorado.
The problem with this model is that it's nearly impossible to
pinpoint who is susceptible and difficult to track
recoveries, though it is possible to figure out the
trajectory of an outbreak, Karspeck said.
To conduct their research, the authors said, they needed
real-time data, and they found it in an online tool called
Google Flu Trends, which uses search terms people put into
the Web-based search engine to figure out where influenza is
The tool, launched in 2008, then notifies the U.S. Centers
for Disease Control and Prevention in real time.
In a process known as retrospective forecasting, the
scientists tested their findings against what happened in the
New York area from 2003 through 2008. Because they knew what
had happened in these years, they could check their work.
Using the computer program and the flu trends data, they
generated retrospective weekly flu forecasts, which predicted
the peak of the outbreak more than seven weeks before it