Maia Majumder was on Twitter earlier this month when she saw a map that terrified her.

The map recorded the number of local newspapers in each county across the United States. Large swaths were shaded light pink, denoting a county that had no local newspaper at all. As a record of the decline of the American newspaper industry, it was disconcerting.

But Majumder, a scientist who specializes in mathematical modeling, saw something different in the splotches of light pink: a disaster for infectious disease surveillance.

Epidemiologists rely on all kinds of data to detect the spread of disease, including reports from local and state agencies and social media. But local newspapers are critical to identifying outbreaks and forecasting their trajectories.

On the map, Majumder saw every county without a local newspaper as a community where health officials and disease researchers could be flying blind.

“We rely very heavily on local news. And I think what this will probably mean is that there are going to be pockets of the U.S. where we’re just not going to have a particularly good signal anymore,” said Majumder, a Ph.D. candidate at the Massachusetts Institute of Technology.

Majumder is a computational epidemiology research fellow at HealthMap, a 12-year-old disease detection project run by researchers from Boston Children’s Hospital. The website uses nontraditional data sources—reports from local news outlets and social media platforms among them— to track global infectious disease activity in real time.

The information HealthMap gathers is used not only by public health authorities, but by researchers around the world who are studying things like the changing geographic distribution of diseases—think Zika—and the impact of climate change on disease patterns, said John Brownstein, one of the co-founders of the site.

“Local media is the bedrock of internet surveillance—the kind of work that we do in terms of scouring the web looking for early signs of something taking place in a community,” explained Brownstein, chief innovation officer at Boston Children’s and a pioneer in the field of using sources other than public health data to do this type of work.

He pointed to the 2009 H1N1 influenza pandemic as an example of a case in which local reporting helped to bring an emerging disease threat to global attention.

The Centers for Disease Control and Prevention was the first agency to detect that two young children in California had been infected with a flu strain circulated in pigs. As officials were trying to figure out if the cases were blips or part of something bigger, however, disease detectives began crawling through recent media reports to determine whether unusual numbers of people with flu-like illness in Mexico had also been sickened by the new virus. It turned out they had been. The new virus was spreading.

“It makes sense that if we see a reduction in local reporting, you’re not going to have that early signaling of something in a community,” Brownstein said.

Without local reporting, it’s also harder to follow an outbreak and assess its progress. Majumder used the example of the massive 2016-2017 mumps outbreakin northwestern Arkansas to illustrate the point.

She and colleagues were trying to figure out why the outbreak was so large—nearly 3,000 cases all told. Had the virus simply found its way into an area with lots of unvaccinated children? Or was something else driving transmission?

That’s the kind of information that is hugely helpful to public health authorities struggling to control an outbreak. But Majumder found she couldn’t easily get her hands on key data; the Arkansas Department of Public Health issued regular updates, but the agency did not archive previous updates on its website. Getting access to official data can be a time-consuming process.

“There is a lot of red tape to get data sharing privileges, even around something as simple as cumulative case counts over time,” she said.

The Northwest Arkansas Democrat-Gazette, however, was covering the outbreak closely, and its coverage provided important context for Majumder’s research. Some of the paper’s stories noted that the vaccine refusal rate in that region was higher than anywhere else in the state; others made clear disease was spreading in a local community of people from the Marshall Islands who had been vaccinated.

Both those factors proved key in helping public health officials better understand what was driving the outbreak, Majumder said.

The World Health Organization urges countries to augment their official disease surveillance efforts—which draw information from networks of doctor’s offices, hospitals, and public health laboratories—with what’s known as “event-based surveillance,” said Dr. Larry Madoff, editor of ProMED, the Program for Monitoring Emerging Diseases. The internet-based outbreak reporting system—which reports on human, animal, and plant disease outbreaks—operates under the auspices of the International Society for Infectious Diseases.

Event-based surveillance is more informal and relies on systems that pick up on media reports, rumors on social media, and the like. That’s the way news of the 2003 SARS outbreak emerged. Internet chatter about a disease that was sickening and killing people in China made its way to the WHO before Chinese authorities eventually disclosed the existence of an outbreak that had been raging for several months.

“It is well-known that event-based surveillance depends on healthy, local journalism,” said Madoff, who is also director of the Massachusetts Department of Health’s division of epidemiology and immunization. “So it would be a reasonable assumption that the loss of local sources would increase the time required to discover an outbreak.”

Like HealthMap, ProMED casts its net globally. In some countries, there are not many local media sources, or if there are, those news sources aren’t available on the internet. In those places, ProMED has to rely on word-of-mouth—which may become the case in parts of the U.S. as local news outlets are shuttered, Madoff said.

He’s not sure if ProMED is starting to miss things because of the shrinking number of local news sources. “There’s a huge volume of information and we get what we get. So it’s hard to say what we’re missing, necessarily. But I can easily imagine that we might,” he said.

Newspapers, of course, aren’t the only source of local news. But a town that can’t support a newspaper may not have radio and television stations either. Radio and TV stations may not archive their reports online, or those reports may not be archived in print form, meaning their archives are harder to search, Majumder said.

Social media cannot fill the gap created by the declining local news coverage, she and others involved in this type of research argued. “With Twitter … you are picking up a signal, but that signal might not be precise,” said Alessandro Vespignani, a professor at Northeastern University whose research focuses on modeling of epidemics. And social media reports can be simply wrong, he said—either by accident or design. News coverage “anchors” the signals picked up on social media, Vespignani said.

“When you combine data from multiple data sources … including data collected from news alerts, we can gain a better handle on the situational awareness in a given community or country,” agreed Mauricio Santillana, an assistant professor at Harvard Medical School whose research focuses on use of novel data sources to track and forecast disease outbreaks.

Majumder said she and colleagues would like to look at whether the growth of “news deserts”—communities without local news sources—has already had an impact on the amount of data HealthMap has been harvesting from under-covered parts of the United States.

One of the reasons the news desert map scared her, she said, was because she realized many of the places without local papers are also places where voters have been complaining of being left behind.

“What that means is they lose access to news which is very, very vital for knowing what’s going on in your town,” Majumder said. “But also from the public health surveillance point of view, we’re losing access to knowing what they need.”

Republished with permission from STAT. This article originally appeared on March 20, 2018