The Future Of Biosurveillance Research

Although biosurveillance will benefit immensely from cross-pollination by the fields just discussed, there will remain areas of both applied and basic research that represent the core science of biosurveillance. Applied research in biosurveillance will translate techniques from other fields and address engineering and organizational issues related to the construction of biosurveillance systems. Basic research in biosurveillance will continue to address questions such as: (1) which biosurveillance data to collect, (2) how to analyze the data, and (3) when and how to act in response to the data.

Over the past five years, basic research has developed experimental designs to address each of these questions. It has developed a large number of algorithms and begun to obtain empirical answers to these questions for selected types of biosurveillance data and algorithms. However, a tremendous amount of research remains to be done.

On the question of which biosurveillance data to collect, there is now fairly good understanding of the types of data that are needed to detect outbreaks and cases and the technical and administrative challenges involved in obtaining them in real time. However, only a few dozen experiments have measured the value of selected data for selected outbreaks or diseases. The matrix of data versus outbreaks is sparsely populated at present. Designers (and users) of biosurveillance systems will benefit most when a more complete matrix is available. On the question of which data are needed to characterize outbreaks, there has been almost no work to elucidate the data needed or to assess the feasibility of obtaining these data automatically.

On the question of how to analyze biosurveillance data, there is now good understanding of the types of analyses that algorithms must perform. In particular, algorithms must detect cases; notice anomalous clusters of disease activity (not only in space and time, but also in sociodemographic strata); compute posterior probabilities of disease/outbreak/outbreak characteristic, given available surveillance data; and utilize all available data and knowledge (e.g., wind patterns, observations of astute individuals, and incubation period of disease).

In the area of automatic detection of cases, researchers are adapting algorithms such as diagnostic expert systems developed by the field of medical informatics, but these approaches have not been evaluated in biosurveillance applications. The process of integrating automated case detection into clinical information systems and the infrastructure being built by national health information network efforts in multiple countries has barely begun, if at all.

Researchers have developed many algorithms that detect possible clusters of disease (univariate, multivariate, spatial scanning, search-based methods); however, algorithms that use information about social contacts (social networks) do not yet exist. There is a need for increased emphasis on algorithm evaluation. The literature on laboratory or field testing to measure algorithm performance, however, is still small, and most experiments have used synthetic or semisynthetic data. As a result, the sensitivity and timeliness of these algorithms for detecting most outbreaks are poorly understood. Such understanding is critical if such algorithms are to be used to guide decision making.

Algorithms capable of computing the posterior probability of an outbreak, given surveillance data, are at an earlier stage of development. This capability is also essential if the algorithms are to guide decision making. Multivariate algorithms capable of analyzing all of the surveillance data that are available to a human analyst will become increasingly important because analysts may not have confidence in an algorithm that is ignoring key data. Similarly, analysts may require algorithms that have "explanation" capabilities (a research area in artificial intelligence). Development of algorithms with all these properties should be a research priority because the ability to collect surveillance data is far outstripping the ability of biosurveillance organizations to analyze them manually—a trend that will continue and result in a data-analytic bottleneck. Algorithms will increasingly incorporate the knowledge of experts about disease and outbreak characteristics (e.g., incubation periods). We note that the effort to encode such knowledge for all diseases and types of outbreaks (encoding all that medical, animal, diagnostic knowledge, and epidemio-logical knowledge) is a formidable research and development task that nevertheless must be tackled.

There has been almost no research on methods to animal and human biosurveillance data in an integrated manner to identify potential "hot spots" for crossover of diseases between the populations. Eventually, biosurveillance systems will track the distribution and intensification of animal farming, location of live animal markets, environmental changes, the distribution of human populations, and the food supply chain.

Finally, there has been little work on algorithms designed to support outbreak characterization. There is a need for methods that can accelerate the identification of the source and route of transmission of an outbreak. There is also a need for methods that can estimate the true size and spatial distribution of an outbreak from observable surveillance data and project its future size and distribution.

On the question of when and how to act in response to biosurveillance data, there have been few quantitative studies. Formal methods to address these questions exist (e.g., decision analysis). However, only a handful of studies have applied such techniques to address problems in biosurveillance. Many studies of the cost of illness, the cost of investigation, and the tradeoffs involved in waiting versus acting to provide guidance to decision makers and designers of biosurveillance systems.


The future of biosurveillance will depend not only on scientific advances but also on how the world addresses educational, leadership, data privacy, and organizational issues. Developments in these areas can have as much impact on the level of disease in the world as the emergence of a new scourge or the development of a new technology. Examples of specific issues follow:

• Education: Due to the introduction of new techniques, many individuals working in biosurveillance face a steep learning curve. The public health workforce does not yet have expertise with the newer mathematical formalisms or with information technology. Conversely, mathematicians and information technologists who wish to contribute in this area do not yet have sufficient familiarity with epidemiology and medicine. Without explicit attention to education, even massive expenditures by governments may not produce rapid improvements in biosurveillance capability. Several organizations have recognized the need for extensive retraining of the work force and for curricular changes in the programs that educate the future workforce. The American Medical Informatics Association (AMIA) devoted a four-day workshop in spring 2001 to developing recommendations for training in public health informatics (Yasnoff, 2001, Yasnoff et al., 2001). The Robert Woods Johnson Foundation and the National Library of Medicine recently funded training programs in public health informatics (National Library of Medicine, 2005). Graduate schools of public health—where many individuals destined for careers in biosurveillance are trained—face significant challenges in evolving their curricula.

• Leadership: Perhaps of greater importance, government leaders and funding agencies also are not familiar with many of the techniques discussed in this book, their implications, or how to guide their development and adoption. These individuals exert a profound influence on biosurveillance legislation, assignment of responsibility and authority, and funding. For example, making high-level patterns of telephone calls available for biosurveillance would be very useful and would not seem to threaten individual privacy. However, there has to be societal and governmental insight and will to establish laws and policies that make such data available. The telephone companies are unlikely to make such data available on their own.

• Legal and ethical: There has been insufficient attention to the question of how best to balance the need to collect biosurveillance data about human health with the rights of individuals to privacy and confidentiality. It remains an unfortunate possibility that in the next few years, the world will experience a devastating pandemic, an intentional release of a disease such as smallpox back into nature, or a mass-casualty bioagent release. In the aftermath of any of these scenarios, the nature of biosurveillance might change dramatically with new societal expectations of the level of reporting and individual surveillance that are necessary. Entirely speculative possibilities might include Internet-based systems for all members of a population to provide daily information of their state of health and symptoms, "medical forensics" to detect possible bioterrorists from medical records and lab tests, medical portals for screening international or even domestic travelers, or more intrusive surveillance of personal health-related information than current society would tolerate. The authors of the book sincerely hope that the field of biosurveillance never needs to enter such a "war footing," but it is possible that some advance debate is needed about these kinds of scenarios.

• Organizational: Governments are investing in biosurveillance, but the net result has been multiple uncoordinated efforts within countries and coordination across countries is even less. There have been several unheeded calls for recognizing that the problem at hand may require a "space-race" or "Manhattan-project" level of organization (Wagner, 2001, Wagner, 2002, Frist, 2005). What is needed may be the creation of government organizations with the ability to fund and focus the development of biosurveillance.

• Improving biosurveillance in developing countries: Because of globalization, it is simply no longer possible to erect a defensive perimeter against disease. There really is now only one population on the earth. The problem of infectious disease is a problem for humanity that must be solved for all people. The world's "biosurveillance system" will only be as good as its weakest link, which at present is biosurveillance in developing countries. Fortunately, Moore's Law is delivering cellular technology and the Internet to these countries; however, it will require significant investment to layer advanced methods of biosurveillance on top of this infrastructure.

Although the accomplishments of the past five years are significant, and there are positive trends that suggest that the world is rising to meet the challenge of biosurveillance, there remains a great deal to do and the threat is clear and present. We hope that this book has in some small way invigorated those who are devoting their careers to achieving this humanitarian goal by recognizing their work, and—by bringing the information from diverse fields together under one cover—will facilitate their work and encourage cooperation and understanding.

Was this article helpful?

0 0

Post a comment