FYI.

This story is over 5 years old.

Tech

Tracking the Spread of Viruses Is Going Viral

With the germane search terms, researchers are tracking germs.
A novel coronavirus, via Rocky Mountain Laboratories and NIH.

Google’s consistent success, chirpy optimism and viscerally obnoxious glasses make its failures pretty satisfying, so bask for a minute—like many did—in the failure of Google Flu Trends last winter.

Google set up a website that used flu-related searches, location data and other formulae to estimate rates of infections, but the results were almost twice the actual estimated rate of infection as reported by the Centers for Disease Control and Prevention. It was a bad year for the flu—at its peak, 6 percent of the population was coughing and sneezing—but it was nowhere near the 11 percent Google’s flu algorithm predicted.

Advertisement

Go ahead; bask in it. Feels good, right? Okay, that’s enough.

Because really, deep down, we all want doctors and scientists to be able to understand and predict disease outbreaks better. Digital resources, like search terms, Tweets and Facebook statuses are tools that they’re still figuring out how to use.

New research, published in the New England Journal of Medicine, looks at the challenges faced by the new and growing field of digital epidemiology, specifically as it is applied to the H7N9 flu virus that emerged in China last year.

Back 20 or 30 years ago, epidemiology was done via patients feeling sick and going to doctors. If doctors found and diagnosed a notifiable disease, they would report the case to a public health organization. The health organization records the data in a central depository and eventually an overall appraisal of the outbreak would emerge.

“This works well for a lot of things,” Marcel Salathé, author of the study, told me, “but you’re missing a lot of things, number one, and number two it’s rather slow.”

In the past five years or so, as people have turned to the web and their smart phones before they turn to doctors, epidemiology has followed them.  “By the time you go to the doctor, literally the fact that you feel ill is already old news on the web, because you’ve already Googled it or tweeted about it or Facebooked it or what have you,” Salathé said.

For four years, Google was able to correlate rates of infection and pertinent search terms with remarkable accuracy, until last winter, when it may have outsmarted itself.

Advertisement

Many attributed the failure of Google Flu Trends to the media attention that the potent flu induced. People searched for “flu” because they heard that influenza had shut down Boston, even though the searcher could have lived in Hawaii where infection rates were normal.

This is one of digital epidemiology’s key challenges, according to Salanthé: deducing who is posting about a virus because its story has gone viral and figuring out who is actually afflicted.

“At the beginning of an outbreak, it’s newsworthy so everyone talks about it, even though there’s only a few cases,” he said. “Then, as there are more cases, it’s not that newsworthy anymore because everyone has already talked about it, so you have an almost opposite pattern from the beginning. Conversation goes down, as cases go up.”

The solution is adding nuance to the algorithms so the formula can recognize first-person accounts versus when someone is talking about someone else, or just repeating what others have reported. It takes a lot of data, but fortunately there’s no shortage of data on the Internet, it’s just a matter of filtering correctly to get down to the right data.

Old models of epidemiology may have been slow and missed a lot, but according to Salanthé, “we feel confident” in the data it does collect, more than collecting what the hyperbolic hypochondriacs are saying on Facebook.

“That data has a health care person who has validated that. That kind of validation is still our weakness,” he admitted, “but I think it’s a technical problem that we get better and better and better at.”

Salathé also talked about the preventative potential that digital epidemiology. There is now a record of things that were never recorded before—vague sentiments and opinions. Doctors can look at them as quantified data to see how sentiments and opinions that correlate to health—such as attitudes toward vaccinations, which last year his team found strongly correlates to rates of infection—propagate. As it turns out, according to someone on the faculty at Penn State for both biology and computer sciences, calling some “viral” is actually an oversimplification.

Exhibiting behavior that recalls Mark Twain more than Richard Dawkins, “in general negative vaccination sentiments spread better than positive ones,” Salathé said. “At the same time if you were strongly exposed to positive sentiments, that would predict that eventually you would be negative in the future yourself. Almost like something backfired.”

There’s truly nothing the Internet does better than resentment. All the same, when the next pandemic comes around, I'll be Googling frantically, rooting for Google Flu to be as effective as those smug nerd with stupid glasses can make it.