Disease detection platform maps Zika fast -- with surprising accuracy

It used to be that to track a spreading disease, the toolbox included Google search and knowing where to look, said Sumiko Mekaru, who heads up the HealthMap project at Epidemico. There was not enough time to track down and analyze relevant information to predict where a disease would go next, let alone conduct any significant preventative measures -- especially in places without many resources, she explained.

Researchers had to work with “little bits of info about infectious disease outbreaks and events,” some of which came from official reports and others were local news reports, Mekaru said. “For anyone tasked for outbreaks or pandemic preparedness, that wasn’t a realistic way to efficiently track disease.”

HealthMap, an online tool for early outbreak detection, aims to fix that.

Epidemico, a Booz Allen Hamilton subsidiary, has strategic collaborations with the Food and Drug Administration, the U.S. Agency for International Development, the Defense Department the Centers for Disease Control and Prevention and the World Health Organization (WHO). It has contributed to global public health through demonstrated early detection of critical outbreaks, including the ongoing Zika outbreak and recent Ebola outbreak, according to ACT-IAC, which awarded it the Game Changer award at its recent Igniting Innovation conference.

It automatically aggregates more than 200,000 data sources and uses natural language processing and proprietary algorithms to tag, filter, analyze, validate and map real-time surveillance of emerging public health threats.

“The platform pulls in a ton of different articles that analyze where, when, how, who. And then it winnows it down to the ones that are actually current breaking news. That’s what goes on the map. Most of those reports are legit.”

And when sources aren’t legit or relevant? Mekaru cited a 96 percent accuracy rating with HealthMap’s current filters, but she admitted filters sometimes pull in the wrong items. And news outlets do make mistakes, so Mekaru's team tries to review everything that goes on the map.

“Pretty much anything you see on the website,” is intended to be reviewed, she said. Epidemico uses public health students as paid interns to review the information. “The reason we do this is not just because we don’t want incorrect things in the map. We also do it because every time we catch a mistake, it improves the system.”

And what about conflicting reports -- for example, a local news story that says there are a number of Zika cases in a country but a government press release saying no Zika has been found?  Mekaru said HealthMap’s goal is to paint the whole picture. “I have five newspapers and a local ministry of health saying we’re worried about Ebola and a national ministry saying we haven’t seen any evidence,” we put it all up. As long as it isn’t blatantly malicious, she said, they include it on the map.

Pouvez-vous dire Zika en français?

Another hurdle Mekaru’s team faced was the different languages in which the news stories and government reports are often released. And for that, they again turned to staff to review each foreign-language piece. Currently they field information in 15 languages and are in the process of adding Hindi. They hope to add more in the future.

“We have staff that is fluent the language and type of news report. For example, the person who analyzes news reports in Portuguese is from Brazil. It is very easy for her to look at a report and say, ‘Oh you know, this says something about Rio, but it is actually about another Rio,’ and she can easily correct a mistake.”

And while Mekaru said the goal is to one day be completely automatic, cultural savvy and language expertise are still a key part of how HealthMap’s process works -- citing a translation issue within a news article as a recent example.

“There was an article a while ago about smallpox in Africa. The person on our team reviewing it flagged it and said it wasn’t right. The word for smallpox and chickenpox in many languages is the same, and we deeply suspected it was a translation error. We put it on the map but we put a comment on there that we expected it was a translation error and expected a correction soon.”

No plans to replace public health’s slow, but accurate methodology

Despite how much faster the HealthMap’s process is than the traditional on-the-ground research and reporting process, Mekaru is very clear that it is not a replacement for the work local public health workers do to verify an outbreak. Rather, HealthMap is meant to signal that there might be an outbreak on the horizon, even if it is just a rumor.

“Traditional surveillance is very linear," she said.  "You feel sick. You go to the doctor. There are labs and reports and tests. Then you find out you have Zika, and that gets reported to your local authority and then to your national ministry and then to the WHO. The problem is that the person who gets sick is so far removed from the WHO, that by the time [global health officials] get that information, there is often a much larger problem.”

But HealthMap can leapfrog the traditional process to quickly deliver what usually turns out to be fairly accurate information. For Zika, Mekaru was able to overlay its initial path with a similar disease from the same mosquito to predict where it might land next. And while it isn’t a perfect match, she said, it is a good indicator of where to look.

Eventually, HealthMap hopes to include each and every city. It’s a challenge, Mekaru said, to include that in a feed when you have 40 Springfields in the United States. And while Epidemico already tracks animal and plant diseases, they want to put more focus on those areas – especially the ones that crossover to humans.  The systems are flexible to search for any data, she said. They just need the resources to make sure it is correct.