Public health surveillance is a cornerstone of the public health approach. Without surveillance (or some other form of assessment), it is difficult to predict or identify a problem or ascertain whether an intervention has made a difference. However, surveillance data quality are difficult to assure, often because of the distributed nature of data collection, or because the data were collected for other purposes. Appropriate interpretation and use of surveillance data, therefore, remain critical challenges for applied epidemiologists. The increasing emphasis on reuse of data already captured for other purposes, for example, National Health Information Network (http://www.hhs.gov/healthit/) and the National Electronic Disease Surveillance System (NEDSS) (http://www.cdc.gov/nedss), not only presents exciting new opportunities for public health, but also highlights challenges and gaps in the science and methods of surveillance. "Syndromic surveillance," designed to detect disease outbreaks early through real-time monitoring of disease indicators,1 exemplifies such challenges in its embodiment of reuse of existing data2-4; fortunately, the focus on evaluation and exploration of this approach5 should lead to methodologic advances applicable to surveillance in general.
Various surveillance paradigms began shifting in the 1990s and continue to evolve today6-9 (Table 1). The focus of surveillance broadened from an emphasis mostly on diseases and outcomes to include syndromes or indicators that occur earlier in the disease process. For such data, it was insufficient to rely solely on traditional (particularly for infectious disease surveillance) sources of data for ongoing surveillance-that is, clinicians and laboratories-and practitioners began exploring the routine utility of data sources such as pharmacies, school also modifies absentee records or workplace absentee records, and 911 calls. Cognizant of the opportunities provided by advances in information technology, public health officials hoped to collect these data via electronic capture or transfer of existing data, and not through the labor-intensive manual data-collection methods that remain common practice. The Centers for Disease Control and Prevention (CDC) began working with partners to move from a fragmented, largely disease-specific approach to surveillance to a cross-cutting, integrated one (NEDSS), intending not only to increase efficiencies in electronic data capture but also to increase effectiveness with a more comprehensive approach. Throughout the 1990s the need to detect deliberately caused as well as naturally occurring illnesses was paramount.
Given the immediacy of concerns regarding potential terrorist events, CDC worked with state/local public health colleagues and the health care system to put together short-term ("drop-in") surveillance systems, surveillance systems for specific events (eg, 1996 Atlanta Olympics, World Trade Organization meeting in Seattle, the Superbowl in 1999, and the major party political conventions in 2000). The goal of these systems mostly comprised early detection of outbreaks during these specific events, using syndromic surveillance methods, and the systems were operational only for brief periods before/during/after the events. Subsequently, however, with the events of September 11, 2001, and the intentional release of anthrax the next month, the reality of terrorism became starkly apparent. Suddenly, public health practitioners faced a convergence of routine and special event surveillance; the heightened concern and inability to predict the next terrorist event meant that daily life after 9/11 became the equivalent of a special event. It no longer seemed acceptable to pursue enhanced surveillance efforts only in special event settings. The general public and government officials expected similarly heightened surveillance on a continuous basis. Yet, "drop-in" efforts required intense investments of resources that were not sustainable, particularly on a population basis.
Advances in technology and concerns about terrorism preparedness continue to facilitate instantaneous availability of data in electronic format, therefore putting pressure on public health officials to know it all in real time. However, while the powerful statistical and computing tools used in syndromic surveillance systems support rapid detection of aberrations above baseline, the clinical or public health significance of these statistical aberrations remains unclear.2-5,10 Fortunately, the interest in, and exploration of, syndromic surveillance approaches highlights some critical surveillance methodology issues, generally worth resolving, and critical to realize the vision of integrated, real-time surveillance in the 21st century.
Selected Surveillance Methods Issues Exemplified by Syndromic Surveillance
How to determine the optimal surveillance method for a given condition or purpose? What attributes of a surveillance system are most important for a specific goal?
The emphasis on timeliness and early detection has spurred the development of syndromic surveillance systems, underscoring the point that selection of the most important attributes of a system depends on the goals (a point made in Hopkin5). Systems designed to detect outbreaks may differ from those used to evaluate programs and the attributes critical to such systems in such instances necessarily will differ.
Does this new system or data source add value? At what cost?
Because these syndromic systems are specifically intended to detect a problem early, do/will they detect a problem/identify trends in a more timely fashion than existing systems? Given the nonspecific nature of some conditions and events that are the focus of syndromic surveillance (eg, respiratory prodrome, indicators such as over-the-counter sales of cold medications), there are concerns regarding not only the rate of false-positive "outbreak" alarms, but also whether the syndromes under surveillance are likely early indicators of the diseases of interest. Finally, irrespective of the predictive value of the syndrome for a disease of interest, when a patient actually has the syndrome, is the syndrome itself consistently captured in the various coding systems used in electronic medical record systems? After all, it is still people who need to tell computers how to record/code the data.
How to decide among various systems/approaches/data sources (existing or potential)? How to use/make sense of multiple data sources or systems?
What do the data really mean? Evaluating the usefulness of the plethora of data available electronically is crucial for appropriate use of the human resources needed to interpret the data. Further refinement of the scientific basis for using these multiple data sources in concert is also needed. For example, how many indicators from how many systems are needed to take action? What to do when they provide conflicting information? We need to explore the reliability and consistency of different data sources and systems whether they are electronic medical records or over-the-counter medication sales.
When is enough enough? Will public health be able to turn off or ignore data determined to be unreliable?
Just because it is possible to obtain reams of electronic data, it does not necessarily mean that it is useful to do so. More than ever, the ever-increasing interest in myriad electronic data sources and the current explorations of syndromic surveillance reinforce the importance of an old-fashioned human technology: epidemiologic judgment.
The articles by Hopkins11 and Silk and Berkelman12 make practical recommendations to strengthen existing infectious disease surveillance systems. They also underscore the need both to use the existing evidence base in surveillance and to develop it further. Although the jury is still out regarding the value of syndromic surveillance approaches, at minimum the deliberations reinvigorate methodologic explorations critical to the future of surveillance in the electronic era.
REFERENCES