How Google can predict your flu-related hospital visit

Four years ago, Google launched Google Flu Trends, which estimates the prevalence of flu outbreaks based on how many people are searching for things like “fever” or “flu medication.” The search metrics have tracked incredibly closely with Center for Disease Control data (as you can see above). And that, Adam Cole reports, has public health researchers thinking they may have a powerful new tool on their hands:

The principle is pretty simple. On the Internet, people who type “fever” or “flu medication” into a search bar are more likely to end up in a hospital a few days later. Data collected from millions of these searches can be shared as they happen, many days before official reports.

But this speed comes with questions of accuracy. Can these signals be trusted? “Skepticism is healthy,” Polgreen says. “Because there’s a novelty effect, there may be excessive expectations both among researchers and the public.”

With this in mind, a team of researchers at Johns Hopkins wanted to see if Google’s estimates would prove accurate and useful in the everyday operation of a hospital. They compared Google searches originating in Baltimore to the number of patients who showed up with flu-like symptoms at a local emergency room.

“It seems like a stretch, but what we found — amazingly — is that there’s a really high correlation between these searches in the community and what we’re seeing in hospitals,” says Richard Rothman, the study’s co-author.

The model is not, however, fool-proof: During the 2009 panic over possible avian bird flu cases in the United States, flu-rated Google searches spiked without any corresponding increase in actual cases.