December 15, 2009 | 1
Rates of infection with H1N1 peaked in the U.S. and western Europe in the beginning of November, months before typical flu viruses from other seasons, according to a November 20 update issued by the World Health Organization. But these early peaks could have been foretold back in September, when Alessandro Vespignani, a professor of informatics and computing at Indiana University in Bloomington, first published a model on predicting H1N1 spread.
Vespignani and his colleagues created a model to predict when and where a H1N1 will strike, based on the patterns of human mobility on both a global scale—focusing on air travel—and a local level—focusing on daily commuting. After correctly predicting how quickly this virus would spread from its origin near La Gloria, Mexico, the researchers have now confirmed their model’s ability to forecast the spread of other pandemic flu strains.
While they found that, not surprisingly, air travel is the metric that predicts the global spread of disease, Vespignani’s group could also pinpoint where, down to a 50-square-mile region, and when, down to the day, an infectious disease would appear. The study was published in the December 13 issue of Proceedings of the National Academy of Sciences.
"You have two components. You have long-range links which are a network of connections and that allows [epidemics] to spread for example from Mexico to Australia or from Mexico to Canada," Vespignani says, "but then when you reach the place…these epidemics spread locally like a wave."
To test how their model could handle a non-H1N1 pandemic, the researchers used the case of the spread of an avian flu strain from Hanoi, Vietnam, in 2001. Along with the global and local travel patterns, the disease itself is the third metric in the model. Information on how easily the disease is spread from one person to another must be entered into the model.
The researchers mapped which cities would first be exposed to this 2001 flu based on the International Air Transportation Association (IATA) 2009 database for the number of people flying and their flight routes for 220 airports, representing more than 3,000 50-square-mile regions. Although, as Vespignani points out, the world is constantly becoming more connected, he does not expect there would have been major differences between 2001 and 2009 air travel patterns. The predictions based on these IATA data were in line with the 2001 and 2002 influenza surveillance data.
If the cities first exposed to a disease form the base of an infection, the regions around those hubs form the branches. To predict the local spread, Vespignani and his colleagues gathered information on the number of people commuting and their commute routes from census data from 29 countries representing 5 continents. The data were taken either in 2006, 2007 or 2008 depending on the country. By comparing areas for which these data were not available to data from regions with similar populations, the group extrapolated worldwide commuting patterns.
Overall, local data show that commuting not only increases the prevalence of an infection in an area, but it also synchronizes different neighboring regions in terms of when they will see the peak of the infection’s prevalence. "[In] New England areas…we were looking at Boston and…Providence and little places around there, and you have all these timings that [the model] predicts better when you introduce the commuting data," Vespignani says.
Using their model, the researchers can play with different scenarios that could inform policies for handling pandemics. For example, the impact of closing airports in an area or regional vaccine efforts could be predicted.
The model could also be extended beyond flu to other infectious diseases that are spread between humans through casual contact. "Diseases in which you have a handshake, the person sneezes or coughs and then you get infected…these are the candidate diseases for a model like this," Vespignani says. To model the spread of sexually transmitted diseases that are usually not spread through daily interactions, he adds that other types of data, such as migration patterns, would have to be considered.
In future work with the model, Vespignani and his collaborators want to explore how, instead of travel affecting disease spread, the disease itself affects travel. While the H1N1 pandemic has been mild so far and not affected travel patterns, Vespignani hopes to empower the model with the ability to project how a more threatening disease could change people’s behavior.
Image of human mobility network courtesy of B. Goncalves et al., Center for Complex Networks and Systems Research, Indiana University