Campus & Community

In simulation, bioterrorist warning system passes test:

7 min read

Could also detect outbreaks of natural diseases

Kenneth Mandl tries to isolate the earliest signs of a bioterrorist attack from the ‘noise’ of ordinary coughs, flu, and other medical complaints. In the background, hazardous material handlers train in the emergency room of Children’s Hospital in Boston. (Staff photo by Jon Chase)

Terrorists secretly released smallpox viruses at several locations in Boston. The highly contagious infection silently began to spread.

In about a week, people began showing up at local hospital emergency rooms with flulike symptoms and a florid rash. As patients described their complaints, computers immediately sent their symptoms to a monitoring center. Those checking the information noticed that such cases were rapidly increasing day by day. An early warning was sounded even before many patients developed the distinctive pus-filled blisters that cover the bodies of smallpox victims.

In the one- to three-week interval between exposure to the virus and full infection, medical teams identified all the people with whom the victims had close contact. These people were isolated and put under close surveillance. The outbreak was contained.

This was only a test of an early warning system now under development for detecting biological attacks by terrorists. However, those who conducted it say that, had it been real, millions of lives would have been saved. “We estimate that smallpox cases could have been reduced by as much as six or seven orders of magnitude, from 10 million without the early warning system to 10 cases with it in place,” says Kenneth Mandl, director of emergency medicine at Children’s Hospital in Boston.

Mandl also believes that the new system could decrease the toll from a large-scale anthrax attack or a lethal outbreak of SARS (severe acute respiratory syndrome). “On the basis of simulations, it has been estimated that we could detect an anthrax attack two days earlier than now possible and decrease deaths sevenfold,” he says. In other words, instead of 70 deaths, there would only be 10.

The system now functions at Children’s Hospital and Beth Israel Deaconess Medical Center, both Harvard-affiliated institutions in Boston. Mandl, who is an assistant professor of pediatrics at Harvard Medical School, expects to tie it into a Massachusetts statewide surveillance network before the end of the year. It also might form part of a national warning system now in the planning stage by the federal Centers for Disease Control and Prevention.

The Harvard program and any national plan that it might become part of also would have the power to detect epidemics of natural diseases, such as West Nile virus, food poisoning, asthma, SARS, and environmentally caused cancers. “To be consistently supported and maintained, the system must be useful in periods when bioattacks are not as large a threat as they are today,” Mandl notes.

Working in real time

Many diseases, including anthrax, smallpox, and SARS, begin with so-called “nonspecific symptoms,” such as rashes, stomach and intestinal upsets, or flulike symptoms (chills, fever, headache, muscle aches, loss of appetite, and fatigue). SARS, smallpox, and anthrax all start with symptoms that mimic flu. Post office workers infected with anthrax bacteria in 2001 were first diagnosed with a “flulike illness” and sent home without further attention. About 24 hours after such symptoms show up, patients can go into an irreversible downward spiral. Without an early warning, that time was wasted and some of the people died.

“We want to be able to detect an excess of patients with nonspecific signs immediately,” Mandl says. “In 1998, when we started looking into this, there was no way that could be done.”

The Centers for Disease Control and Prevention and the World Health Organization track the spread of flu, but it takes days or weeks to bring reports from doctors and hospitals together. That was the case with the present SARS epidemic. It’s difficult to see an unusual cluster of cases. A single doctor may see only one or two. Even two doctors in the same hospital, but on different shifts, may not tell each other about the people they have treated. And doctors get too busy to fill out forms.

Some surveillance systems rely on putting information on paper or computer forms, which are sent to public health departments. “Overall these methods don’t work too well,” Mandl allows. “During the World Trade Center attack on Sept. 11, 2001, initial compliance was good, but over time problems arose with consistency of data entry and quality of information.”

Mandl and his colleagues became convinced that the system must be fully automated and run continuously, not just when something is expected to happen. “We need the surveillance to begin as soon as a patient checks into an emergency department, even before he or she sees a doctor,” he notes. “We need to discover without delay any cluster of cases that no one expected to see.”

To spot something that’s abnormal, such as a smallpox attack or a SARS outbreak, you have understand what’s normal. To do that, the researchers at Children’s Hospital and the Harvard School of Public Health started with information from 10 years of normal visits to the emergency department at Children’s, some 500,000 visits.

They tracked the coming and going of colds, coughs, rashes, stomachaches, and sore throats from season to season. Using software originally developed for the national census, they plotted each patient’s address on a local map. Finding geographic clusters of cases would be important in situations like the SARS epidemic where many cases originated in one apartment building in Hong Kong.

“We saw distinct patterns over time and space in the way different sets of symptoms play out over a week, a season, or a year,” Mandl notes. Using the same type of analysis employed to predict stock market volumes, they found that they could forecast tomorrow’s volume of patients based on today’s number of visits to an emergency department.

Most of the cases they tracked were uninteresting. They were, by definition, normal, or mere noise in the system. But the tracking provided a powerful statistical way to set limits on what to expect. When symptoms and locations exceed such limits, that’s when the alarm will go off.

Testing the alarm

The next step was to test the alarm by simulating smallpox and other attacks. Highly contagious diseases such as smallpox or SARS would produce a rapid daily rise in nonspecific symptoms. Anthrax, which is not as contagious, would trigger a surge in emergency room visits over a few days. To detect such variations, the researchers inserted filters that can be tuned to various time periods. In this way, the system can look at a one-day spike in visits, a fixed number of additional visits, a steadily increasing number, or a rapidly rising number of cases.

Mandl published the results of these tests with Ben Reis, who works at the Markle Foundation in New York, and Marcello Pagano, professor of statistical computing at the Harvard School of Public Health. “We showed you shouldn’t just look at what happened today,” Pagano comments. “Rather, you should put what happened today in the context of what happened the day before, and the day before that, etc. It’s a simple idea that improved detection tremendously.”

“The results can be generalized both to outbreaks of natural diseases, and to other detection systems now being built in many regions,” Mandl points out. Such applications promise quicker and more sensitive ways to detect bioattacks whether from terrorists or from nature.