Medical Informatics Realties in
Bioterrorism Surveillance
By Judith Broadhurst

Cities have been running disaster-preparedness drills for years, long before 9/11/2001, including scenarios for intentionally contaminated water supplies and such. Fortunately, a massive chemical release in a major population center and communicable diseases intentionally spread on a wide scale are unprecedented. So the closest we've come to testing the real-world effectiveness of technology-based early detection was at the Olympics in Salt Lake City (after 9/11) and, more recently, with scattered West Nile virus outbreaks.

Yet many people in the Bush administration, as well as in major cities and medical centers, seem to be hanging onto the hope that sophisticated medical informatics networks will detect a bioterrorism attack in the US soon enough to protect most of us. Agencies of the federal government, private companies and cities, with and without federal assistance, are steadily pouring more money into development of systems to track everything from consumers' purchases of over-the-counter remedies at drugstores to emergency room visits to job absenteeism.

In their December roundup article, "2002: The Year in Healthcare," HealthLeaders magazine summarized several relevant federal government efforts last year:

• In January 2002, the US Health and Human Services agency allocated $2.9 billion in fiscal 2002 to combat bioterrorism and related threats — a tenfold increase over the $296 million appropriated in 2001.
• The Centers for Disease Control and Prevention received a $918 million increase in supplemental funding to bolster lab capacity and communication.
• Large numbers of physicians received training to help them quickly diagnose and treat various infectious diseases that could be part of a biological attack.
• In the early fall, the Bush administration floated the idea of inoculating all 280 million Americans with a vaccine against smallpox.
• Officials also planned to stockpile enough anthrax vaccine to protect 25 million people.

"Despite these and other proposed changes," they concluded, "many public health observers believe there's still a long way to go before the healthcare system is truly prepared to handle a wide-scale bioterror incident."

Skepticism is valid

That skepticism is healthy, says Justin Graham, MD, an infectious disease specialist who also has a master's in medical informatics from Stanford University. His research (funded by DARPA) on automated decision support and surveillance for bioterrorism highlights several problems with the technology itself, the data collection, and the interpretation of the data. (See the Stanford University BioSTORM project home page for more information)

"There's an unsubstantiated belief that if we are able to find out that lots of people are suddenly getting mildly ill and in some suspicious pattern, we would have early warning to activate our public health system and see what's going on, as we've done for West Nile virus, for instance," he says. "The widespread adoption of information technology with data from a variety of sources now makes it possible to combine data, electronically, in ways we never could before. It's conceivable that, with the right algorithms and networked data, we could look at all this data in real time and come up with a usable signal.

"Projections show that early detection of just hours could make an enormous difference. But it's very hard to detect a pathogen early. There's no big cloud that people can see, nothing to announce it, as there is for a bomb. Communicable diseases are even more difficult. How do you pick up what's going on between the release of a pathogen and people getting sick?"

Problems in both theory and implementation

Scientists are gambling on two techniques for that, says Graham:

• Detecting the actual pathogen by monitoring the environment and
• Detecting patterns among people who are sick, but still only mildly ill.

Collecting data about people is a major problem, however. "It has been really, really hard," says Graham. For example:

• Pharmacy sales data can tell competitors and shareholders how a drugstore chain is doing, so they want to control what, when and how anything is disclosed.
• For the same reasons, pharmaceutical companies don't want to provide information that researchers need.
• Monitoring medical records raises the all-important privacy issue.

"The detection phase is anonymous," says Graham, "but once an outbreak is detected, we need to notify the patients and doctors. For public health data in general, that's a problem. In this kind of thing, you're doing broad-scale detection and scanning for information every day, looking at many people who aren't sick and who aren't contagious. So there are privacy tradeoffs."

Collecting the data to detect agroterrorism — attacks on our food supply and economy through animals or crops — carries similar problems, he says. "We study animals as proxies for human health. But with [large, commercial growers and breeders], says Graham, "Nobody wants anyone to know their cows have died. That's propriety information, so they're very guarded about that data."

Even when we can gather, compile and correctly interpret the data, there are still problems that may be insurmountable, at least at this point.

"It has never been proven that these kinds of systems will give us early detection," says Graham.

"All we can say is 'we think we're looking for this or that' — that people buy Nyquil, for instance, when they can't sleep because they're coughing. There's an influenza outbreak every November through March, so everybody trains their systems on that data. Anthrax has flu-like symptoms — cough, fever, and so on — but all that happens before people go to the doctor or show up dead or dying in emergency room, when it's too late to treat them. A couple of grams of anthrax could kill a million people. How would we know the difference at first?"

The most serious problem, though, and the reason the hope seems like hype when you start looking behind the news headlines is this single statement from Dr. Graham: "According to some models of worst-case scenarios, the casualties from a bioterrorist's attack can be astronomical if it isn't detected until 48 hours after its release. Prototype automated surveillance systems have never been able to prove they can detect a pathogen that quickly."

A hastily prepared list of arbitrarily selected reading and sites related to medical informatics and bioterrorism detection