Supercomputer's Model of Human Contact Simulates Swine Flu
A group at Virginia Tech is working with the U.S. Department of Defense to tackle the H1N1 outbreak
6 May 2009—An extravagantly detailed computer model of the U.S. population is taking a crack at understanding the H1N1 ”swine flu” outbreak. The model, built by researchers at Virginia Polytechnic Institute and State University, in Blacksburg, Va., is composed of realistic representations of the major ways that people in the United States come into contact with one another—in other words, real-world social networks. Last Thursday, the U.S. Department of Defense began using the model to provide recommendations to the Department of Health and Human Services, according to the Virginia Tech engineers.
In the model, called EpiSimdemics, real cities are represented as groups of artificial people whose demographic attributes match data from the last census and land-use databases. By seeding the model with a handful of infected individuals in a manner that mirrors the real cases—say, 45 teenagers in one part of New York City—the model can run hundreds of simulations to illustrate possible future infection patterns across a population of between 50 million and 60 million in nine regions, according to Madhav Marathe, a deputy director of Virginia Tech’s Network Dynamics and Simulation Science Laboratory (NDSSL). In one experiment, for example, the model was asked to determine the impact of school closures on flu transmissions.
The major innovation of EpiSimdemics is the ease with which it allows users to manipulate a wide array of variables in the simulations. ”We’ve built a tool that lets [public health officials] design experiments on how interventions will affect outcomes,” says Christopher Barrett, the director of NDSSL, which is in charge of the project. The tool is essentially an interface that hides the high-performance computing platform that runs the model, enabling public officials and health experts to tap into the university’s computing power without requiring any technical expertise.
One question that the model may help answer in the upcoming months is whether to release an H1N1 vaccine, assuming one is developed while the flu is still active. ”Suppose the current outbreak goes away in the summertime,” says Stephen Eubank, a physicist on the project. ”There are going to be a lot of questions about what to do with the vaccine if it’s ready by the fall.” The model can bring to light the full range of possible outcomes from a vaccine intervention, including changes in the flu’s virulence and the number of people who might become infected under different scenarios.
Another use is to help understand the impact of antiviral medicines. The use of antivirals places a specific type of pressure on a virus, which could cause a more virulent strain to evolve. Public health officials may need to decide whether to try to aggressively snuff out the strain with antivirals. The alternative, if the H1N1 virus proves not to be particularly virulent, is to hope that the strain dies out on its own.
An active research direction for the Virginia Tech team involves modeling the impact of sequestering a small group of critical individuals, such as transportation workers and people in the power industry, to limit the economic impact of disease outbreaks. Recent systems engineering research shows that the U.S. freight-rail system is particularly vulnerable to a flu pandemic.