Analyzing online searches and social media activity has often been suggested as a way to track and maybe even predict the spread of diseases. And it’s a great idea — if it’s done right, it could offer public health workers real-time surveillance and a jumpstart at containing dangerous outbreaks. But there’s a hitch. How can we attempt to decipher between online activity triggered by the possibility of actual disease symptoms and online activity triggered by simple curiosity?

That was the question Sherry Towers and her colleagues set out to answer. At the very least, they wanted to gain some deeper insights into what kind of variables were related to spikes in social media behavior. With an answer to a question like that, researchers who want to use social media activity to help monitor disease trends could better discern which activities were more likely to reflect actual illness and which activities were just people expressing concern online. And the handful of Ebola cases diagnosed in the U.S. last year presented the perfect opportunity.

Towers, a research professor at the Simon A. Levin Mathematical, Computational and Modeling Sciences Center at Arizona State University, knew the likelihood was extremely low that any of the U.S.-based Google searches and tweets on Ebola would be related to actual symptoms of the virus. In turn, it was the ideal scenario to study what other types of variables — besides actual symptoms — led to spikes in online activity.

“Many studies have shown that the spread of ideas in the population might be infectious,” Towers told me. “When Ebola was first identified in the U.S., it was very noticeable how the media exploded and lots was happening on social media. It struck me that people had attempted to use social media in the past to track the spread of disease, but the problem is how do you disentangle that from people who are just expressing concerns or interest? How do you get rid of the background?”

To conduct the study, which was published this month in the journal PLOS ONE, Towers and her colleagues examined daily Ebola-related Internet searches and Twitter data in the U.S. over a period of six weeks. Quite fittingly, they then used a mathematical contagion model to determine if news coverage was a significant factor in online variations of Ebola-related activity. TV news coverage data was based on the daily number of Ebola-related news videos appearing on two major networks. And according to the study, the researchers did indeed find “evidence of a contagion.”

That contagion was the news, which inspired tens of thousands of Ebola-related tweets and Internet searches with each news clip. In fact, the study found that trends in Ebola news reporting explained nearly all the variations in online activity. In other words, between 65 percent and 76 percent of the variance in Ebola-related online activity fit within the parameters of the news media contagion model.

“We were actually surprised at how much the variation followed temporal patterns in the news media,” Towers said. “It showed that what’s causing people to tweet was not so much independent thought or independently talking with one another. It was driven by the media.”

Towers and her study co-authors from Arizona State University, Purdue University and Oregon State University write:

The vast majority of published digital epidemiology results show a positive correlation between digital data and the temporal evolution of epidemics or outbreaks; but what typically are not seen are the analyses that show no significant correlation, likely due to the “file drawer” effect where uninteresting or null results simply are not published. Indeed, several of the authors can attest to the fact that use of Twitter data to predict outbreaks is fraught with difficulties (unpublished data), and accounting for potential sources of bias is extraordinarily difficult. However, in this new age of readily accessible, and rapidly evolving, temporal and geospatial information in social media, digital epidemiology has a hopeful future as a tool to detect newly emerging infectious diseases and track the spread of established diseases.

Towers told me that contagion models of disease are being used more and more in the social sciences to track the spread of ideas and to help determine if people are “infecting” each other with the urge to tweet or search for a particular topic online or whether people are being “infected” by an outside source. If you can use mathematical models to better understand the dynamics of people’s online behavior, you can increase the likelihood of being able to use the online behavior as likely indicators of disease spread, she said.

“We used the model to explain how Ebola interest proliferated in the social media world,” Towers said. “And it turns out the vector is the news media.”

However, Towers did note that the model explored in her study would be most applicable to emerging diseases with which people are unfamiliar, as opposed to diseases such as the flu, which don’t always make a big news splash.

To read a full copy of the Ebola media study, visit PLOS ONE.

Kim Krisberg is a freelance public health writer living in Austin, Texas, and has been writing about public health for more than a decade.