Early Warning

Stung by anthrax mailings after suicide skyjackings, the United States is hurrying to erect an electronic line of defense against further bioterrorism. At least five sophisticated biosurveillance systems are under development with federal funding to nonprofit and to proprietary ventures; two other groups already have products on the market. The goal is to install a national sentinel network that can detect suspicious trends in medical data and in illness behavior before diagnosis, to help contain a disease by identifying it soon after infections begin. Called syndromic surveillance because it tracks signs and symptoms rather than positively diagnosed disease, this new technology will also accelerate containment of naturally occurring pathogens. The data processed by these systems range from traditional patient charts to such nontraditional sources as Internet health site hits, over-the-counter drug sales, or absences from work and school. A major challenge is accurate analysis of the data, which can involve complex statistical methods, decision-making tools, and even artificial intelligence. But many questions remain to be fully answered, including: How much time between infection and medical treatment can be saved by these methods; what kinds of data are the best indicators of serious disease outbreak; and how well will the systems cope with the problem of raising false alarms? "This was solidly a research field on Sept. 10 and was considered to be a standard technology on Sept. 12," says Kenneth D. Mandl, an assistant professor of pediatrics at Harvard Medical School and research director in the emergency medicine division at Children's Hospital, Boston. Mandl is coprincipal investigator of an academic consortium developing one of the new biosurveillance systems. "It's actually still a research field," he stresses. "We should not forget that a lot of the basic science behind it still needs testing." The first product to reach the marketplace was LEADERS (Lightweight Epidemiology Advanced Detection and Emergency Response System), developed by a consortium that includes EYT, an information technology company in Chantilly, Va. LEADERS can interrogate patient records and compare chief complaints with those of selected syndromes, says EYT vice president Tim Hannon. The system gathers data primarily through continuous feeds from the client hospital's information system. Two other sources are electronic form-based data entered by clinicians and a "heightened surveillance" mode that can place added emphasis on new patient information to generate alerts during major crowd-drawing events. Another for-profit venture, headed by Veridian Corp. of Arlington, Va., will feature artificial intelligence being developed by Mark A. Musen and colleagues in Stanford University School of Medicine's medical informatics division. Veridian program manager Gene McClellan says the prototype system, RPHD (Real-time Population Health Detector), is due for testing later this year. It will routinely acquire diagnostic tests ordered, deduce possible reasons for them, evaluate the physician's hypothesis, and derive an abstraction of such information to place the patient in a syndromic category. The system will integrate relevant data from various patients during a given period. When an unusual-looking pattern emerges, the likelihood of an outbreak or attack can be calculated by algorithmic comparison of the anomalies to normal population health standards. "No one has a magic approach to the issue of false alarms," McClellan allows. He adds that the consortium is continuing to cultivate multiple data sources in an attempt to conquer the signal-to-noise problem. Flu Complicates Detection Part of the reason for this problem is that many deliberately released pathogens—as well as others that occur naturally—manifest as flu-like symptoms. They can be difficult to differentiate in their early stages, not only from each other but from influenza itself. How to detect an upper respiratory infection that's not merely a runny nose is essentially a question of signal-to-noise. Although the new systems are expected to outperform traditional health surveillance methods, there is debate over whether signals from multiple data streams will improve detection or merely increase the noise. Senior scientist Al Zelicoff at the Center for National Security and Arms Control of Sandia National Laboratories in Albuquerque, N. Mex., argues that no evidence has yet been compiled to show that nontraditional data streams will be useful or collectible in real time. Formerly a rheumatologist, Zelicoff masterminded RSVP (Rapid Syndrome Validation Project), which was jointly developed through Sandia, Los Alamos National Laboratory, the University of New Mexico, and the New Mexico Department of Health.1 The latter three collaborators have broken away from Sandia to work on their own biosurveillance product, citing discontent with RSVP's limited data stream and with Sandia's decision to proprietarily market the system. Zelicoff is placing RSVP in groups of hospital emergency rooms in states across the United States and in Singapore. He dismisses the proprietary issue as a red herring. The question, he says, is not whether biosurveillance is offered under a potentially profit-making arrangement but whether the software will interconnect with other systems. (RSVP and most of the other new systems are striving for compatibility.) Zelicoff adds that the decision to become proprietary was based on a need to get the system quickly onto the market. "My fear is that someone is going to put a bad product out there," he explains. The health care industry won't give such technologies more than one chance to prove themselves, he believes. Courtesy of Al Zelicoff, Sandia National LaboratoriesSentinel at the Ready: The RSVP system's zoomable GIS-base (Geographic Information Systems) map that physicians can use to review reports of all similar cases of illness in their area. RSVP relies on Web-based reporting by physicians of new cases that fit into any of six syndromes related to naturally occurring or deliberately caused disease, especially influenza-like illnesses. But physicians are notoriously bad about reporting diseases, because the current system is paper-bound and bureaucratic, he says. He offers a rule of thumb for any reporting software: "It had better be easy, cheap, fast, intuitive and, oh, by the way, give the doctors something back within their attention span, which is about two minutes." The system gives links to geographically based information about disease outbreaks throughout the world, and provides details on local and regional health conditions. The physician enters demographic data that includes age range and zip code but there is no patient name or identifier number. After hitting a series of tabs that indicate symptoms and signs, the doctor gets a best-guess analysis of the case, in text and graphs, based on local conditions. A map tells of similar local reports in the past 30 days and another map offers current, comparative epidemiological information on the six syndromes. "This is the Holy Grail of epidemiology: geographically based analysis of signs and symptoms in the population," Zelicoff declares. He adds that by relying on the expertise of local physicians, "We make the signal as high as we can and keep the noise down." RSVP's former collaborators have different ideas about that. The system they're developing, called B-SAFER (Bio-Surveillance Analysis, Feedback, Evaluation, and Response), will integrate nontraditional data streams into the RSVP data such as calls to the nurse strife line, emergency room, and poison center. Illness behavior begins before people go to the hospital, comments Judith C. Brillman, associate professor of emergency medicine in the University of New Mexico's Health Sciences Center. Detection could be days or even weeks faster, she says. A prototype of B-SAFER, set to be completed by the end of September, will show a photo of a person. Clinicians can click on an organ or system to get a pop-up menu for entering data, allowing them to report signs and symptoms without selecting a syndrome. "It gives all the syndromic information but doesn't make anybody define anything on the front end," Brillman explains. One response to the clinician's data will be a differential diagnosis, based on an analysis of the heterogeneous data. More Systems Developing Other systems in development that will use nontraditional data streams include RODS (Realtime Outbreak Detection System), led by Michael Wagner of the University of Pittsburgh School of Medicine's center for biomedical informatics, and ESSENCE II (Electronic Surveillance System for the Early Notification of Community-Based Epidemics), led by Joseph S. Lombardo of the Johns Hopkins University applied physics laboratory. Both men declined to be interviewed. At Boston's Children's Hospital, Mandl and colleagues are not currently incorporating nontraditional data into their system, which is provisionally called Biosurv. It is being tested at the hospital and by another consortium member, Beth Israel Deaconess Hospital. The next step is to extend it to nine other Massachusetts hospitals, Mandl reveals. Biosurv, which has a focus on pediatric health, works in several modes, including automated surveillance of ER data. Initial patient complaints and hospital billing diagnosis codes are used. Mandl affirms that the main difficulty is to understand what the data signify. Children's Hospital alone gets 50,000 visits annually, he notes. "To say whether you're having an abnormal or a normal day is not a trivial determination." Accordingly, the team is geographically modeling data spanning a decade to determine norms. "I think you'll find that syndromic surveillance will be tying into practical applications in at least 50 different systems, one for each state," he predicts. That could happen within the next two or three years, he adds. "The problem is, how will the regional data be interfaced at the national level? I haven't heard the solution to that yet." Steve Bunk (sbunk@the-scientist.com)is a contributing editor. 1. A. Zelicoff et al., "The Rapid Syndrome Validation Project (RSVP)," Journal of the American Medical Inform atics Association, (Suppl.):S771-S775, 2001. Web Links RSVP demo: rsvp.sandia.gov/SNL B-SAFER demo: openemed.net RODS home page www.health.pitt.edu/rods

Early Warning

Interested in reading more?

Become a Member of