Hey there, human — the robots need you! Vote for IEEE’s Robots Guide in the Webby Awards.

Close bar

Mosquitos Have Brought a Nasty New Disease to the Americas. Can Computer Models Predict Its Spread?

DARPA challenged computer modelers to predict the spread of the chikungunya virus, with decidedly mixed results

3 min read

Mosquitos Have Brought a Nasty New Disease to the Americas. Can Computer Models Predict Its Spread?
Photo: BSIP SA/Alamy

So far in 2015, more than 565,000 people in the Americas and the Caribbean have come down with chikungunya, a viral disease spread by mosquitoes, according to estimates from the Pan American Health Organization. That’s pretty impressive work for a virus that made its first appearance in the western hemisphere in December 2013. 

Public health officials throughout the Americas have been scrambling to contain this unprecedented outbreak of chikungunya. To do so, it would certainly be helpful to be able to predict when and where the next hotspots will occur. But right now, public health officials simply don’t have the necessary tools, DARPA program manager Matthew Hepburn tells IEEE Spectrum.

“If we ask decision makers, whether they’re in the United States, or the ministry of health in one of these other countries, or the Pan American Health Organization, ‘What models or forecasts are you using today to make predictions?’ The answer is, ‘We’re not using any,’ ” Hepburn says.

DARPA sought to correct this situation last year by issuing a challenge to computer modelers, asking them to predict the spread of chikungunya for the six-month period between September 2014 and March 2015.

The results don’t seem very encouraging. “No one did a really good job of forecasting when a disease was going to spread to a new country, and when it would go into an exponential growth phase,” Hepburn told the crowd at a DARPA conference this summer.

The word “chikungunya” comes from an African language and means “bent over in pain,” which gives a pretty good idea of the disease’s symptoms. It causes high fever and intense pain in the joints, and while it’s rarely fatal, the pain can last for months or even years. There’s no vaccine or treatment for the disease.


Image: Pan American Health Organization
In this map of countries with reported cases of chikungunya, the dark purple areas show where people have contracted the disease from local mosquitos.

The virus has caused sporadic trouble in Africa and Asia since the 1950s, but outbreaks have become more severe in the last decade. Researchers think that the virus mutated around 2005 and became more infectious to the mosquito species Aedes albopictus. Commonly called the Asian tiger mosquito, this invasive species has spread throughout the world, and may be responsible for the introduction of chikungunya to the Americas.

Hepburn spoke to Spectrum about his ongoing analysis of the results of the DARPA challenge. There were some successes, he says: Once an outbreak was established in a country and had entered an exponential growth phase, a number of the models did a good job of predicting the future growth curve and when the peak number of infections would occur. (That’s not necessarily an easy task; for more on this, see Spectrum’s recent feature article on the computational biologists who tried—and generally failed—to predict the course of the Ebola epidemic in West Africa.)

But the models weren’t up to the task of predicting when chikungunya would jump to a new country and spread rapidly. Columbia, for example, saw a dramatic rise in cases during the six-month course of the DARPA challenge, Hepburn says, but the models didn’t predict that surge.

So far in 2015, Columbia has seen more than 325,000 suspected cases of chikungunya, more than any other country. But even if public health officials had known that this outbreak was coming, what could they have done about it? In the absense of vaccines or treatment, would that prediction have been useful? Hepburn argues that foreknowledge would have let officials weigh the costs and benefits of expensive mosquito-control programs. “You have to make that deicison early, before you hit a lot of cases,” he says, in order for such programs to be effective in controlling the spread of disease.

The various models used data relating to geography, climate, transportation connections between countries, and the spread of the tiger mosquito in their predictions. Hepburn says DARPA is still analyzing the results to determine which data sources were the most useful, but he has already determined one thing: More data is not necessarily better.

“We assume that more data sources, more info, more details make models more sophisticated, more complicated, and therefore more accurate,” Hepburn says. “That assumption is not necessarily true. By introducing a lot of additional factors, you may just be introducing noise.” 

The goal here is noble: to bring about “ a revolutionary improvement in disease forecasting, in much the way that weather reports transitioned from surveillance to forecasting,” as DARPA put it in a press release.

For another approach to this type of disease forecasting stay tuned for the October issue of Spectrum, which includes the feature article “Building a Better Disease Detective.” In that article, researcher Barbara Han explains how she deploys machine learning techniques to identify wild critters that carry terrible plagues. Here’s a brief intro to her work.

The Conversation (0)