01 October 2006
Epidemiology in Silicon
Donald Burke, MD, is the dean and Jonas Salk Chair of Global Health in the Graduate School of Public Health at the University of Pittsburgh.
For two years I have been principal investigator on a grant from the U.S. National Institutes of Health to develop computer models of infectious disease epidemics that may be important to national security. My group decided influenza was a top priority, so we did two types of influenza modeling. For one part of the modeling effort, we worked with collaborators in Thailand to develop simulations of a hypothetical epidemic in Southeast Asia. Then we used the model to determine if intervention strategies could stop an early-stage epidemic in its tracks—what we call “quenching” an epidemic—in Asia before it spread worldwide.
To do that, we created a simulated population for Southeast Asia, focusing on Thailand. Our simulation distributed 85 million individuals onto a map according to population densities. We put them in households, schools, and workplaces—basically creating an artificial society in the computer. We computationally released an influenza virus into the population and studied the transmission patterns that ensued. Then we evaluated what would happen if Thailand treated cases, treated families, closed the schools, or geographically restricted people’s movements. We are testing policies—plans, procedures, and actions designed to bring about a desired governmental goal, in this case epidemic control—in the simulation, and we call it “epidemiology in silicon.”
It isn’t possible to rigorously test policies before an epidemic explodes, but by doing it with a simulation that has some fidelity to natural patterns, you can ask if certain combinations of policies are likely to be more effective under certain circumstances. We published our findings in Nature magazine (September 7, 2005). The main conclusion was that if you responded to a nascent epidemic at a reasonably early stage—fewer than 50 cases—and used an aggressive strategy of treating the cases and everyone in the geographic area with antiviral drugs, it would be possible to contain or quench the outbreak before it became an epidemic.
The second part of our modeling effort, published in Nature on July 26, 2006, was to do the same thing for the United States—create a simulation of population density, movement patterns, households, workplaces, schools, distributed airline travel, and local travel. The difference in the United States is that we don’t expect to be able to completely stop an epidemic. At the height of a global pandemic, such a high percentage of potential travelers would be incubating or ill with influenza that stopping even 99 percent of air travel into the United States would still allow a large number of infected persons into the country by airplane.
These computer models are computationally intensive. We run the models thousands of times because every time we run them, just as chance influences reality, we get somewhat different results. To assess a policy, we have to run a simulation multiple times to see, on average, what effect an intervention-strategy policy option will have on the epidemic. Depending on the simulation, each run can take half an hour on a supercomputer.
In mid-2005, we were just finishing the Southeast Asia quenching modeling work when an opportunity came through the Fogarty International Center to increase Thai involvement. The Thais were very expert from the policy side, but they didn’t have modeling sophistication because most epidemiologists in Thailand don’t have a background in computational modeling and simulation. With Fogarty’s support, we worked with the Thai epidemiology training program through the Ministry of Health and provided training opportunities in modeling. The key collaborator there is Dr. Kumnuan Ungchusak, director of the Bureau of Epidemiology in the Department of Disease Control at the Ministry of Public Health.
Our group is working with the Thais on three levels. First, we worked directly with them as research colleagues to develop models. They were wonderfully helpful in this—we could not have completed our first modeling effort without our Thai colleagues. Second, we have worked on more classroom-oriented types of interactions, where larger groups learn the technology but are also exposed to computational approaches to modeling epidemiology. In June 2006, the Thai students completed a course for field epidemiologists. In addition to a regular epidemiology course, my junior colleague, Dr. Derek Cummings, gave a series of classes on modeling opportunities for 25 or 30 students in the class. Third, which isn’t yet accomplished because we’re still early in the program, we will identify degree candidates to work on projects that are in part related to modeling and simulation.
The opinions expressed in the preceding articles do not necessarily reflect the views or policies of the U.S. government.
From the October 2006 edition of eJournal USA.
