(European Commission DG ECHO via Compfight/Flickr)
A digital surveillance program used Twitter feeds and news headlines to pick up on the Ebola outbreak in West Africa a full nine days before the World Health Organization proclaimed it an epidemic.
But that doesn’t mean the outbreak could have been prevented.
Dr. Alessandro Vespignani, a professor of computer science and physics at Northeastern University, uses network science to model and forecast the spread of disease. Like HealthMap, the online tool cited above, Vespignani’s computer simulations cannot anticipate an outbreak before it actually begins.
“They don’t have a crystal ball either,” he says. “HealthMap is really a novel way of doing disease surveillance that can provide a real edge in the early detection of outbreaks by monitoring news articles, journals, Twitter or other digital sources. But they can’t do this before the actual occurrence of the event. There was already a situation in West Africa. HealthMap was just able to pick up the anomaly before anyone else.”
As the death toll climbs over 1,000 in West Africa, I was curious to know what makes this particular outbreak so relentless and what the global community can do to contain its spread. My conversation with Dr. Vespignani, lightly edited:
First of all, what exactly are big data and network science research? And how do you use them to track disease outbreaks?
We create large-scale models for disease forecasting by creating a synthetic world in the computer that integrates all data about human mobility. Then we plug an infectious individual into the model and look at the spread of the disease. You can look at different levels of granularity—whether locally or internationally. Network science is important because most disease now spreads by human mobility. What you hear many times is, “We’re all one hop away from West Africa,” although it’s thousands of kilometers away. No one has a crystal ball, so we cannot say when there will be an Ebola disease outbreak. As soon as we have the data on the outbreak, what we can do is try understanding how it will evolve in the next few weeks or months, which is what we do with this modeling.