For his latest research, Neil Ferguson had to face an event that could spell disaster for the world. The epidemiologist at Imperial College London wanted to know what would happen if the avian influenza virus H5N1 mutated so that it could pass readily from human to human. How fast would the flu spread? What, if anything, could be done to stop a pandemic?

To find out, Ferguson, with fellow epidemiologist Don Burke at Johns Hopkins Bloomberg School of Public Health, and their colleagues, built the largest computer simulation of infectious-disease epidemics yet published.

The model simulated an outbreak starting in Thailand, so the first thing the team needed was detailed data on that country's population. “The sizes and locations of households, workplaces and schools, and how far people travel between each are key,” Ferguson explains. Collecting these data and turning them into model parameters, such as how many people one person might contact in a certain time period, was harder than writing the program's code, Ferguson says.

The process was further complicated by a lack of background information. “We had to make some assumptions about how a new influenza virus would behave,” says Ferguson. “These had to be based on what was seen in past influenza epidemics and pandemics.” But that sort of information proved hard to come by. “Less detailed statistical work had been done on past pandemics than we hoped,” Ferguson says. Making up for this shortfall was an important part of the team's research.

Once they had the data and the computer model, Ferguson and his team set out to make sure they covered all possibilities. They used ‘sensitivity analysis’, which involves running the model over and over again using different assumptions about unknown parameters, such as incubation times, and looking at how the outcome changes.

This meant running the model hundreds of thousands of times. To do these runs quickly, the model needed to be coded efficiently, and required computers with huge amounts of memory — 20 times that found on a typical PC. In fact, the team hooked up ten high-powered computers in parallel, but even then the final runs took more than a month of computer time.

The outcome (see page 209) was worth the wait. The team found that on average one person infected with a new pandemic virus might infect 1.8 other people, that people are likely to be highly infectious for only 1 or 2 days after they develop symptoms and, most importantly, that we have a chance of preventing a pandemic if we can detect the first few cases and act fast enough.

Ferguson says that the results argue for improving disease monitoring, creating international stockpiles of antiviral drugs and vaccines, and planning detailed strategies for a rapid response to suspicious clusters of human cases. The advanced online publication of the paper has already helped prompt Roche to stockpile drugs to enable the World Health Organization to tackle flu outbreaks using similar methods to those modelled by Ferguson's group.

Meanwhile, Ferguson and his team are working on a model of what would happen if containment failed and a pandemic spread from Asia to Europe and the United States.