How Computer Modeling Can Stem the Spread of Influenza

August 18, 2009

Experts are bracing for an extremely high H1N1 flu infection rate this fall and winter. Joshua Epstein, director of the Center on Social and Economic Dynamics, says computer modeling can help the medical community and policy-makers predict which populations are most susceptible to infection, how great the infection rate will be and how to stem the spread of the virus.


“At the moment we are trying to calibrate these models to existing data on prior pandemics and using the best available estimates of what this disease would look like – that is to say in consultation with the Centers for Disease Control, the National Institutes for Health and other agencies. Of course it is not clear how quickly flu could mutate into a very serious disease so we expect something like a 30% attack rate which is a high level of disease. We are talking about tens of millions of cases worldwide. But, again, the main issue is whether it will mutate into a form more severe than we’ve seen and whether we can develop a well-matched vaccine to that disease, whether people will adhere to distancing measures – to stay home from school, stay home from work and so forth. So that number (the 30% attack rate) assumes business as usual, no interventions, no social distancing, no travel restrictions, none of the things that we would impose so we hope to do way better than that but its needs to understand that if we don’t do those things it could be very severe.”

“…Classical epidemic modeling uses differential equations and assumes very well-mixed populations with no particular diversity in the susceptible, or the infected, or recovered groups. Our version of modeling is called agent-based computational modeling and we, basically, build artificial societies of individuals who have different levels of susceptibility, can go through different stages of the disease, can adapt their behavior, can exist in social networks that we try to capture, so much more of the social richness and behavioral realism is captured in our form of modeling than in classical epidemiology. Although classical epidemiology has given us wonderful and deep insights about disease progression, about the non-linear tipping behavior of epidemics and has also given tremendous insight into what vaccination strategies should be focused on achieving.”

“…You would never base policy on a particular run of the model or a particular evolution. You run these things many, many times and build up a robust statistical portrait of how the disease might progress and address your policy to that. But we know a lot about how people move around – better in the United States than in less-developed countries – but in the U.S. we have good data on movement from zip code to zip code, the distribution of trips by distance, we know a lot about international air travel – we have that completely programmed into our models. So we have good idea how people would move around in a day-to-day, business-as-usual scenario. What is a lot less clear is how people will adapt their behavior under the stress and fear of a large-scale epidemic. And we are seeing, in India for example at the moment, rapid spread of fear itself. And we’ve seen this before. There is a famous incident in 1994 in Surat, India where 300,000 people evacuated Surat, a city in India, out of fear of pneumonic plague and in the end not a single case was confirmed by the World Health Organization. This is also an issue.”