Thanks to a new computer model of the entire U.S. population, agencies soon will be able to test the various strategies, policies and programs for combating epidemics, responding to disasters, improving health habits and more.
Comment on this article in The Forum.The model is big enough to almost encompass the current world population of 6.7 billion (it can house 6 billion individuals, or so-called agents), offering the possibility of testing responses to global pandemics, to worldwide effects of alterations in trade or monetary policy, and the human cost of natural and political disasters, among other global events.
The Large-Scale Agent Model, which resides at the Brookings Institution's Center on Social and Economic Dynamics, hosts 350 million agents that simulate the U.S. population. The model takes into account the age and gender of each agent as it actually occurs in the United States and how those individuals are distributed within the country's 31,255 ZIP codes. These virtual Americans move among ZIP code areas daily, which allows the center to model and track through time and space the progression of, say, an infectious disease outbreak such as pandemic influenza.
The ability to grow artificial societies allows government policy-makers and officials to watch how social, economic, biological and civil events develop based on demographics, and then to see the effects of specific interventions from government or other organizations on the outcomes. Such agent-based modeling took hold in the social and biological sciences and economics in the early 1990s. But only now have they become big enough and flexible enough to model and predict human behavior on a large scale.
The LSAM resides on eight computers at the Brookings Institution's center. It was developed under the auspices of the Homeland Security Department's National Center for the Study of Preparedness and Catastrophic Event Response at Johns Hopkins University.
The large-scale model creates easily understandable visual representations of vast events and takes into account the vagaries of human behavior. Programming into the model the release of a pandemic flu bug in Los Angeles and modeling the rate of infection based on a set variable of interactions that people have with family, work colleagues and schoolmates, produces a spreading scarlet stain across the map of the continental United States as ZIP code area after ZIP code area turns red, signifying that more than 5 percent of the population has become infected.
The model also can take into account a panoply of human responses. Some agents may refuse to be vaccinated, for example, as would a significant portion of the U.S. population in the event of a true epidemic. Adding such realism helps improve the model's predictive power and its depiction of real-world outcomes.
Other variables, such as reducing the interactions between people, can help policy-makers decide what to do in case of a real outbreak. In the LA-based flu scenario, the model showed that reducing interactions among agents by 75 percent for one month prevented the outbreak from becoming an epidemic. Reducing interactions caused the outbreak to fizzle for lack of carriers.
In real life, a government proscription against attending school, going to work, shopping -- what is known as a nonpharmaceutical intervention called "social distancing" -- would seem draconian and be difficult if not impossible to carry out. So what about a 50 percent reduction over six months? Well, many more people die, but the longer period buys time to develop a vaccine.
It's easy to imagine how such modeling could improve policy-makers' understanding of the dynamics of epidemics and therefore help them develop better preventive strategies.
Adding information to the model can make the predictions better. Brookings plans to add the location and the patient capacity for every hospital and emergency room in the United States. It then can test if the resources are distributed properly in case of an outbreak. Ultimately, the plan is to model the global population.
Combining agent-based and other computer models produces richer depictions. Brookings has collaborated with Bharat Soni of the University of Alabama at Birmingham mechanical engineering department to examine transportation options in response to chemical contaminant releases in cities. Their model shows how a toxic plume from a river barge would spread across New Orleans and how people in the city's buildings would react. As the simulation runs, it's immediately obvious that numerous deaths would result simply because people, as they exit buildings by the thousands to escape, would become stuck in the congested streets under the toxic cloud.
The outcome suggests any number of government interventions, including a simple requirement that those in harm's way remain in a building. But the simulation also will model human behavior during disasters, including resistance both to evacuation and remaining in a shelter, the desire to locate and join family members, concern for property and the belief that authorities are unreliable.
The hybrid model, combining a toxic release and the response of cyber-people, is wholly novel, according to Joshua Epstein, director of the Brookings center, who also heads up global epidemic modeling for the National Institutes of Health Models of Infectious Disease Agent Study. "As far as we're concerned, agent-based modeling is an all-terrain vehicle applicable to all sorts of scenarios," he said during a March 12 press event at Brookings.
But Epstein, a pioneer in applying agent-based models in the social studies, says the university center of excellence is autonomous and retains the right to choose what it studies so the model could not be commandeered by a federal agency. For example, "if the government wanted to use it to figure out better means for urban warfare, Brookings could refuse," he said.
On March 11, the Brookings center received the 2008 Modeling and Simulation Award for Outstanding Achievement in Analysis from the National Training and Simulation Association for their computational feat in creating such a large model suited for many research projects and diseases. Written in the programming language JAVA, the model can be rapidly developed and run on computers using different operating systems.