EDMG541

Koenig_and_Schultz_s_Disaster_Medicine_Ch_11.pdf

Syndromic Surveillance

Gary A. Roselle

OVERVIEW

Syndromic surveillance has been defined by the U.S. Centers for Disease Control and Prevention (CDC) as “the collection and analysis of health-related data that precede diagnoses or labora- tory confirmation and signal with sufficient probability a case or an outbreak for further public health response.” Based on its original definition, the purpose of syndromic surveillance would be to prevent morbidity and mortality by early identification of case clusters in which mitigation would affect the outcome of the disease’s natural course. This original definition was designed for early event detection and became prominent in the pub- lic domain after the September 11, 2001 terrorist attacks in the United States and the subsequent anthrax illnesses and deaths.

With a heightened sense of urgency related to the so-called “war on terror,” many systems were put into place within the United States for the protection of the public health. These included such diverse programs as vaccine initiatives (BioShield), static detectors located throughout large cities to identify specific organisms of interest in the air (BioWatch), and the beginning of a national syndromic surveillance system for early detection of outbreaks (BioSense). These three initiatives were designed for the following reasons, respectively: 1) prevention of dis- ease if a terrorist attack occurred; 2) early identification of air- borne pathogens during the asymptomatic phase of such disease; and 3) early identification of illness prior to definitive diag- nosis that would be confirmed either by culture or laboratory tests. These government initiatives were complemented by inde- pendent, nonfederal syndromic surveillance systems that were designed primarily for early identification of naturally occurring illnesses but were adaptable for use in bioterrorism surveillance. Other countries, such as the United Kingdom,1 Canada,2 and Australia,3 have also heightened the evaluation of syndromic surveillance systems for response to early detection of bioter- rorism events. In addition, there are other syndromic surveil- lance initiatives in Europe4–6 and Asia (e.g., Japan7 and Taiwan8). These systems are largely based on surveillance of existing data, such as help line calls and emergency department visits. For such population-based reporting, several factors must be defined if

the surveillance system is to be useful. The system must provide initial detection such as finding an event as early as possible. It must quantify the event by defining the number of people who are potentially ill and identifying the location for the source of infection with enough granularity to allow for specific interven- tion. In addition, it would be useful if the surveillance system incorporated other supportive data such as provider and labora- tory testing, and permitted early, computer-based investigation of possible case clusters by using such items as patient demo- graphics. At least in theory, this should allow for initial outbreak management, such as confirming existing cases and tracking new ones, and timely countermeasure administration such as isola- tion, antimicrobial prophylaxis, or vaccination. In addition, if maximal utilization of data is the goal, then bioterrorism surveil- lance systems should also identify naturally occurring outbreaks and case clusters, because this will be the most frequent use of the data on an ongoing basis.9–17 For such data usage, there are multiple components of a surveillance system that should be defined prior to its implementation (Figure 11.1). Such ques- tions as, “what is the population under surveillance?” “what is the time period for data collection?” and “what data will be collected and who provides it?” are essential components that should be decided in the planning phase. For personal data secu- rity purposes, the issue of information transfer and storage is absolutely critical. Some assortment of personal identifiers are mandatory even if only zip code, age, and sex are used. The issue of data analysis is also critical, particularly who will analyze the data, what methodology will be used and how often, and finally, how will the reports be disseminated, to whom, and by what method. Although all of these factors may appear to be self- evident, there are no accepted and universally available standards for syndromic surveillance that would make the answer to these questions simple. Added to these complexities would be the need for data validation, which may or may not be possible based on the need for reporting timeliness.

With an increased emphasis on surveillance for non- intentional events to augment the usefulness of expensive bioter- rorism surveillance systems, the term “situational awareness” has moved from the military to the public health community. Thus,

165 Koenig and Schultz's Disaster Medicine : Comprehensive Principles and Practices, edited by Kristi L. Koenig, and Carl H. Schultz, Cambridge University Press, 2009. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/apus/detail.action?docID=564432. Created from apus on 2018-03-07 11:37:36.

C op

yr ig

ht ©

2 00

9. C

am br

id ge

U ni

ve rs

ity P

re ss

. A ll

rig ht

s re

se rv

ed .

166 ■ GA RY A. RO S E L L E

SURVEILLANCE SYSTEMS Components

• What is the population under surveillance? • What is the time period of data collection? • What information is collected? • Who provides the surveillance information? • How is the information transferred? • How is the information stored? • Who analyzes the data? • How are the data analyzed and how often? • How often are reports disseminated? • To whom are reports distributed? • How are the reports distributed?

Figure 11.1. Components of a surveillance system to be defined prior to use.

for a system to be fully operational, it should go beyond the pos- sibility of early event detection and define the location, extent, and progression of disease clusters and outbreaks of many differ- ent types and at many different times. For this mission, it may be important to deal with a greater variety of data sources than tra- ditional symptom syndromes. It may require more well-defined geographic locations for individuals (for example, a five-number zip code vs. a three-number zip code in the U.S.). More timely reporting that would approach real time may be necessary, mean- ing instantaneous transmission of any data point as soon as it is available with instantaneous analytic processing and report generation and distribution.

Last, the traditional use of symptoms for syndromic surveil- lance may not be adequate to accomplish all of these missions, including early event detection and situational awareness. A vari- ety of other data sources have been suggested and investigated. Examples include over-the-counter medication sales, prescrip- tion drug purchases, number of phone calls to pediatricians’ offices, absenteeism from schools or work, and ambulance emer- gency runs. This so-called “augmented syndromic surveillance” could allow for greater specificity of signals that define true clus- ters or outbreaks versus statistical anomalies that would oth- erwise require large increments of time by the public health authorities for investigation.18–24

CURRENT STATE OF THE ART

The concept of syndromic surveillance is relatively straight- forward, although the proof of concept and/or value is yet to be shown. In its simplest terms, data that can be immediately obtained prior to definitive diagnostic testing (e.g., microbiol- ogy culture or laboratory serology) are transferred to a central repository. Examples of this type of data include healthcare diag- nostic or procedural coding, such as International Classifica- tion of Diseases (ICD-10) or CPT5 codes, or chief complaints. After receipt at the repository, the data are parsed into groupings related to established syndromes, such as respiratory, neurolog- ical, or gastrointestinal. The philosophy of syndrome grouping such as this rests on the assumption that, although errors may be made in specific diagnostic or procedural codes or chief com-

plaints, the general group of the codes should be correct and allows early analysis of data. In addition, even if ICD-10 or CPT-5 coding is not possible, natural language interpretation of chief complaints can be used with specific trigger words to allow assignment of a syndromic grouping. For identification of the possible spread of avian influenza, syndromic surveillance is considered an important component for early detection of avian flu infecting humans, primarily based on the use of the category of “influenza-like illness” or “respiratory” syndrome using the syndromic surveillance model.

Additional data can be added such as blood pressure or tem- perature to improve the predictive value of any signal derived from statistical analysis of the syndrome groupings. For instance, if evaluating traditional respiratory syndromes, the blood pres- sure of most patients arriving at a healthcare facility would be reasonably normal, even during cold and flu season. On the other hand, if an airborne anthrax attack were to occur, there may be a significant increase in patients with a respiratory syndrome, high fever, and very low blood pressure. A more robust response by the public health community may be required when confronting such a severe syndrome as defined by augmented syndromic surveillance.

There are multiple syndromic surveillance systems in use around the globe and even across the U.S. Often more than one is visible to regional public health entities for use as a stand- alone system or in conjunction with other local alerting systems, such as ambulance runs, to determine priorities for public health investigation and intervention. Syndromic surveillance contrasts with the “knowledgeable intermediary,” the single clinician who, recognizing that a patient or group of patients arriving for care display an unusual set of signs or symptoms, activates pub- lic health authorities. Such knowledgeable intermediaries have been evident in the anthrax attack in the United States and the sarin gas attack in Tokyo. Syndromic surveillance is also differ- ent from standard reporting of notifiable diseases in that such disease reporting is often made after a diagnosis is confirmed such as hepatitis B, meningococcal meningitis, or tuberculo- sis. Although such reporting is important, it generally lacks the timeliness necessary for mitigation in the case of an intentional biological event. Syndromic surveillance, therefore, is a method- ology designed to gain the advantage of earlier detection (by days) of a biological attack or other infectious illness. This may allow for earlier intervention to stop the spread of disease, rapid initiation of appropriate treatment for individuals affected, and, perhaps, by increasing such timeliness, increase the probability of apprehending the perpetrators of an intentional biological event.

Current state of the art in syndromic surveillance is a rapidly moving target. There are a multitude of surveillance systems available. One review of the literature identified 36 systems25

and U.S. health departments alone have implemented syndromic surveillance systems in more than 100 sites since 2003.15 This, coupled with the shift from early event detection to situational awareness, the development of large city systems individualized for specific geographic areas, and the creation of new technolo- gies, necessitates using exemplar systems of syndromic surveil- lance in this discussion. In fact, for public health entities that are better resourced, multiple systems are often used simultane- ously to differentiate true case clusters or outbreaks compared with anomalies identified by the syndromic surveillance system. Improving this signal-to-noise ratio allows for optimum use of public health resources for investigation and mitigation as

Koenig and Schultz's Disaster Medicine : Comprehensive Principles and Practices, edited by Kristi L. Koenig, and Carl H. Schultz, Cambridge University Press, 2009. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/apus/detail.action?docID=564432. Created from apus on 2018-03-07 11:37:36.

C op

yr ig

ht ©

2 00

9. C

am br

id ge

U ni

ve rs

ity P

re ss

. A ll

rig ht

s re

se rv

ed .

SY N D RO M I C SU RV E I L L A N C E ■ 167

U. S. SURVEILLANCE SYSTEMS Examples of

• National Electronic Disease Surveillance System (NEDSS) Centers for Disease Control and Prevention (CDC) and partners

• Epidemic Information Exchange (Epi-X) CDC, state/local health departments, federal agencies, military

• Rapid Syndromic Validation Project (RSVP) Some local, public and private institutions in New Mexico

• Real-time Outbreak and Disease Surveillance (RODS) Western Pennsylvania (13 counties, 14 hospitals, 10 emergency Departments)

• Electronic Surveillance System for the Early Notification of Community-Based Epidemics (ESSENCE)

Department of Defense (DoD) • BioSense

CDC, state/local health departments, DoD, Department of Veterans Affairs (VA)

Figure 11.2. Selected surveillance systems from the United States.

needed. Figure 11.2 gives a brief listing of several U.S. surveil- lance systems, past and present, using a variety of methodologies to achieve the previously stated goals.

Surveillance Systems

National Electronic Disease Surveillance System Although not a true syndromic surveillance system, National

Electronic Disease Surveillance System is a U.S. CDC initiative designed to create standards to ensure uniform data utilization. This includes: data architecture; the user interface; information systems software architecture; tools for interpretation, analy- sis and dissemination of data; and secure data transfer. In fact, there are many surveillance systems in place in U.S. state health departments and the CDC using varied tools that are not inter- operable. The plan is to establish standards and apply them to all the surveillance systems. This would allow electronic transfer of appropriate information from clinical information systems in the healthcare industry to public health entities. This would reduce provider burden in the provision of information, and enhance both the timeliness and quality of data delivery. It would also allow better integration of current datasets by those responsi- ble for public health. This is an extremely complex operation, but, such interoperability will be critical as surveillance systems expand. Gathering information from single or multiple sources for interpretation will be critical because the large volume of information may facilitate statistical analysis and provide better data for the end user.

Global Outbreak and Alert Response Network The World Health Organization (WHO) is in a unique posi-

tion to manage international outbreaks of all sorts. Although this may not be syndromic surveillance, sources of data supervised by the WHO do include a variety of information systems from countries all over the world, each of which uses different surveil- lance systems, different reporting methodologies, and different degrees of voluntary reporting. All of this reporting is influ- enced by global resources, government oversight and political considerations, and includes differences in endemic diseases that may influence the data input into surveillance systems. Although there are multiple networks and institutions around the world, the WHO infrastructure is designed to provide an operational

framework to link the expertise and skills needed to keep the international community constantly alert to the threat of out- breaks. Based on a management system in Geneva and regional hubs around the world, this so-called system of systems allows for general monitoring of outbreaks internationally. As a result of the global economy and the dramatic increase in interna- tional travel, outbreaks in Asia or South America can rapidly spread around the world. Therefore, timely data are important for national security and local health department information needs. In the absence of data uniformity, however, any informa- tion from such a network of networks must be analyzed carefully for validity and value.

Rapid Syndromic Validation Project The Rapid Syndromic Validation Project was developed and

tested by several organizations within the United States, includ- ing Sandia National Laboratories, the University of New Mex- ico Department of Emergency Medicine, and the New Mexico Department of Health. This system is a model that depends on information from practitioners on specific syndromes of interest. Clinicians enter data into a stand-alone, dedicated system. This provides two-way communication between clinicians and pub- lic health authorities. There are advantages and disadvantages. The main advantage of the system is that for syndromes of spe- cific interest, the data should be valid and useful. The provider has defined the patient’s illness to be of sufficient importance that the time required to transmit data to a computer reposi- tory is worth the effort. This should dramatically increase the signal-to-noise ratio and provide important data on specified syndromes. On the other hand, the workload in providing this information is a disincentive to reporting, and bias from uneven data collection and underreporting may be seen. Also, only syn- dromes of interest will be reported, rather than use of a statistical methodology to define what syndromes are occurring at greater than expected frequency. Such a system can be very responsive, however, to the need for situational awareness involving a sin- gle syndrome that requires reporting by clinicians in a timely manner.

Real-time Outbreak and Disease Surveillance The Center for Biomedical Informatics at the University of

Pittsburgh developed a system known as Real-time Outbreak and Disease Surveillance (RODS). It originated as a syndromic surveillance system that included several symptom groups based on chief complaints. RODS developed the use of natural lan- guage processing techniques to define syndromes from free text chief complaints because of delays in coding emergency department or outpatient visits. In addition, RODS has been augmented using such information as laboratory test results, radiology tests ordered, and nontraditional data sources such as over-the-counter drug sales, call centers, and absenteeism. Emer- gency managers used this system for the Winter Olympics in Salt Lake City, Utah, and it has many advocates. Of interest is the use of augmented data in addition to symptom data. Particularly during the 1999 influenza season, public health authorities noted that the earliest predictor of the increase in influenza was ele- vated over-the-counter sales of cough and cold medicines. This, at the very least, showed the concept that augmented syndromic surveillance might be an improvement over the more traditional symptom-based methodology alone. Extensive work has been done on the use of the natural language processor with good validity testing, although opportunities for further improvement

C op

yr ig

ht ©

2 00

9. C

am br

id ge

U ni

ve rs

ity P

re ss

. A ll

rig ht

s re

se rv

ed .

168 ■ GA RY A. RO S E L L E

exist.26,27 This may not be necessary if appropriate coding is done contemporaneous with patient visits for transmission to the cen- tral data repository.

Epi-X Epi-X is a U.S. CDC program that is a secure, web-based

communication network for public health investigation and response. It provides timely information, reports, alerts, and discussions about terrorist events, toxic exposures, disease out- breaks, and other public health events. When U. S. public health officials post reports to Epi-X (or when cross-border surveillance information is received from Canadian or Mexican officials), the information is shared rapidly with public health officials across many U.S. states and jurisdictions. Information is encrypted and secured, an important aspect to maintain data security. Access is limited to specific public health officials. This is a data entry information system with limited access that can be very useful for early reporting of real and suspected clusters and outbreaks of disease across the country. It does involve provider labor in that it requires information be sent in a nonautomated manner to a web-based system. It has the advantage of being a secure system to public health officials. However, this limited access is also a disadvantage in that the information is not widely available in the public domain.

Electronic Surveillance System for the Early Notification of Community-based Epidemics

The Electronic Surveillance System for the Early Notification of Community-based Epidemics is a component of the Defense Global Emerging Infections System with programming from the Johns Hopkins Applied Physics Laboratory.28,29 This was initially a fairly pure syndromic surveillance system generally defined by ICD-9-CM coding with collapse of symptom groups into syn- dromes such as gastrointestinal, respiratory, neurological, coma, and others. This is in use by the U.S. military services, and also is available at the Infectious Diseases Program Office of the Vet- erans Health Administration. This system uses a defined set of mathematical algorithms and a “stop light” methodology with red, yellow and white alert levels for individual facilities. It also allows for multiple onionlike layers of data that can be linked to the individual provider. Such linkage can be important if a group of diagnoses suddenly becomes identified as “red” based on single provider information. It is possible that these data may not define a cluster of true cases but rather coding variability by a single individual. In fact, the goal of such data layering is to allow review of a putative cluster of cases electronically to determine whether further action is required. As with most syn- dromic surveillance algorithms, the use of multiple statistical testing of the dataset generally assures some positive results by chance alone.

Neural Network A neural network is designed as a system that is trainable

to “understand” mathematically data input over time. It uses complex mathematical formulations to define aberrations based on previous complex and apparently erratic behavior. The com- puter reviews all historical data and defines any aberration based on previous “learning.” It looks at increased numbers based on a statistical model without an a priori definition. Although the user selects numerous attributes of a dataset that might help characterize the behavior of interest, the automated detection processor trains on the specific data to look for anomalies. A

major advantage of the system is the definition of items such as days of the week, seasons of the year, weekends, and other selected datasets that may have influence on anomaly detection. This is particularly important in healthcare where seasonal vari- ation and weekday variation may be prominent. Such processing also allows channeling of very large datasets into smaller streams of digestible information. Although there is valid proof of con- cept for this type of data analysis when examining such items as harbor or motor vehicle traffic, medical proof of concept is sparse.

Using information from a Department of Veterans Affairs dataset, a proprietary neural network system tabulated infectious diseases data involving 10 pathogens and 10 variables in greater than 187,000 records. Specific spikes in data were found for hepatitis C antibody when the Veterans Health Administration was conducting a Hepatitis C Surveillance Day. Seasonal varia- tion in diagnosis of Escherichia coli O157 was also determined. This methodology has not been adequately validity-tested but is another example of mathematical modeling for large datasets that involve organisms and other computerized data.

BioSense BioSense was initially designed as an early event detection

methodology assessing and analyzing existing diagnostic health data. It has been extended to ongoing situation awareness, and the U.S. CDC uses BioSense as the designated agency to over- see national biosurveillance data related to human illness. In its initial phase, BioSense data reported to the CDC were pri- marily ICD-9-CM and CPT-4 codes for outpatient and emer- gency department encounters. Most of these data were received from the Department of Veterans Affairs and the Department of Defense. In addition, demographic information was also trans- mitted, particularly related to geography, age, sex, and other items as available. For each patient there was a unique identi- fier assigned, although this was translatable back to an individ- ual patient only by the sending agency and not by the CDC. This maintained patient privacy while allowing for the CDC to perform appropriate sorting and analysis. Once received, the diagnostic and procedure codes were collapsed into as set of syn- dromes (such as gastrointestinal, respiratory, or neurological) for analysis. Both statistical analysis and geographic mapping were available. As time progressed, increased timeliness of data trans- mission as well as expansion to more complete datasets became a major initiative at the CDC. Specifically, such items as vital signs and laboratory tests gained higher interest in order to provide added value to the basic datasets.

Data Integration

To add further value to any syndromic surveillance system, the addition of nonhuman data might also be useful. For instance, data on water systems nationally or internationally, data on unusual illness or deaths in animals,30 or data from unusual occurrences regarding plants or food crops might also increase the positive predictive value of any clusters of human cases found by syndromic surveillance.

The complexity, however, of integrating systems that are dramatically diverse, such as those involving plants, animals, BioWatch sensors, and human data is daunting. In addition to the different types of data noted, there is also the question of dif- ferences in information technology architecture among the vari- ety of datasets or between countries. Although there is variable

C op

yr ig

ht ©

2 00

9. C

am br

id ge

U ni

ve rs

ity P

re ss

. A ll

rig ht

s re

se rv

ed .

SY N D RO M I C SU RV E I L L A N C E ■ 169

architecture in human datasets,31–33 this is highlighted when automated data systems are in place regarding plants, animals, or other more technical datasets such as BioWatch sensors. The information technology architecture of the originating datasets will be important if such large amounts of diverse data are to be electronically delivered, sorted, initially analyzed, and out- liers defined. Developing a specific platform architecture that has been clearly defined for all of these diverse datasets remains a challenge.

At the time of this writing, the U.S. government is developing the National Biosurveillance Integration System that is designed to track and integrate data it will receive electronically from multiple agencies. Such agencies include the CDC, the Environ- mental Protection Agency, the Department of Agriculture, and many other national and international sources. It will use these data to generate reports on the state of risk to the public health. In addition to the difficult task of electronic data interpreta- tion, the National Biosurveillance Integration System will also require human analysts to integrate the algorithmic quantifiable data and the more “fuzzy” threat data obtained from a variety of intelligence sources, such as that from U.S. Embassies and other electronic surveillance traffic. Although this is ground breaking technology, it will take significant time to fully implement and an even longer period to determine whether it is effective.

Data Analysis

There are a variety of mathematical data analysis formulae in place in the extant syndromic surveillance systems.34–48

These include everything from cumulative sum scores, smart scores, and anomaly detection algorithms to trends, proportions, expected to observe frequency, standard deviations, and other more descriptive statistics. At this time, system designers have not demonstrated that any of the statistical methodologies are definitively and clearly superior to any of the others, or that they define specifically which outliers are critical for investigation.

With limited resources provided to the public health com- munity, the critical element in a syndromic surveillance sys- tem is the ability to define which outliers are sufficiently impor- tant to generate an on-the-ground investigation and determine whether a biological event has occurred. This so-called signal- to-noise ratio is one manner of defining the validity of the syn- dromic surveillance system. If alerts are generated very frequently with little or no outcomes from laborious investigation, the sys- tem will go unused and thus have little value. If, on the other hand, the surveillance system is correct every time it defines an abnormal signal, then it is likely that such extreme specificity will not allow for sufficient sensitivity. It may allow other crit- ical signals to go undetected at some unknown rate. Although mathematicians, statisticians and modelers are critical to the process of syndromic surveillance analysis, the most essential element will be in the hands of the public health and epidemi- ologic community where the signal-to-noise ratio will truly be defined.

Value

While inconclusive, there has been some evidence validating syn- dromic surveillance, implying these systems have value. This includes early surveillance of infectious syndromes and trauma after the September 11, 2001 attack in New York City. Such investigations yielded positive findings for certain defined infec-

tious syndromes of interest such as exacerbation of underlying respiratory conditions and diarrhea/gastroenteritis.49,50 Auto- mated syndromic surveillance was also in place for the 2002 Winter Olympics in Salt Lake City, Utah. Although over 100,000 acute care encounters were monitored, no outbreaks of public health significance were detected. This effort does show, however, that an automated syndromic surveillance system can be imple- mented in a setting where thousands of persons gather for a specific event.51 At a large public gathering of more than 40,000 persons in the U.S. state of Virginia, public health authorities conducted surveillance for early detection of disease outbreaks.52

Groups of persons with gastrointestinal disease were found; the system also detected noninfectious events such as those related to heat and physical injury. At the Rugby World Cup in Sydney, Australia, an automated public health surveillance system was also used.53 Although no outbreaks were identified, the system provided situational awareness without an increase in workload for clinicians. All of these efforts underscore the utility of pub- lic health screening and surveillance at mass gatherings where definitive diagnoses may be delayed. In New York City, diar- rheal illness was detected through syndrome surveillance after the power outage of August 2003.54 With regard to influenza- like illness, syndromic surveillance has been used in both eastern Virginia and Connecticut. It appears that the syndromic surveil- lance data were in alignment with the influenza season in both settings.55,56

These are promising examples, but the value of data obtained from any syndromic surveillance system compared to that obtained from a knowledgeable intermediary or the standard public health reporting systems has not been clearly elucidated. Many scholarly articles address this issue, but it is not clear that any have defined a specific, documentable role in early event detection for syndromic surveillance. This should not be con- strued to mean that early event detection is unimportant. Rather, it means it has not as of yet definitively proven itself to be effective. This may be related to inadequate data sources, lack of proper specific items for data transmission, the need for improvement of mathematical algorithms, or the fact that public health authori- ties lack awareness of the case clusters that may be found by such surveillance systems.

Difficulties in proving effectiveness of syndromic surveil- lance must be examined in the context of the system’s initial development challenges. For instance, it is not clear, based on scientific research, which variables are necessary in an expanded syndromic surveillance system to identify abnormal signals and increase the signal-to-noise ratio. A second issue is the accuracy of the data provided to the analysts. It is well known that ICD-9- CM coding is extraordinarily variable. However, this may not be critical based on the large number of cases that would be submit- ted for analysis and the fact that groups of ICD-9-CM codes are collapsed into syndromes rather than as stand alone diagnoses. Another issue is the population for which syndromic surveillance data are provided. Are the patients for whom data are transmit- ted to the repository appropriate for surveillance? For instance, if most of the risk is generated in large metropolitan areas, are all such groups reporting into a national biosurveillance system such as BioSense? Do the data include information from the immunosuppressed populations because they may be the first to become ill if a biological event occurs? Does the system for data transmission, analysis, and feedback reporting provide an adequate timeframe to allow mitigation if an unusual occurrence is found?

C op

yr ig

ht ©

2 00

9. C

am br

id ge

U ni

ve rs

ity P

re ss

. A ll

rig ht

s re

se rv

ed .

170 ■ GA RY A. RO S E L L E

These issues, as well as others, are unresolved at the time of this writing, likely based on the short timeframe during which national biological surveillance has been emphasized. In addi- tion, because intentional biological events are extremely rare, proof of value for the system will likely be made from naturally occurring events such as influenza epidemics or gastrointesti- nal illness. Although these are important public health issues, they may lend themselves more to the knowledgeable interme- diary reporting system than syndromic surveillance or standard public health reporting related to specific diagnoses.

It is, therefore, logical to assume that the syndromic surveil- lance community is in its knowledge acquisition phase to deter- mine the methodology, value, and optimal use of syndromic surveillance data. Although data integration and analytic sys- tems are being developed, it is critical that program evaluation proceed with sufficient vigor to answer these difficult questions in a quantifiable, scientific manner.57–64

RECOMMENDATIONS FOR FURTHER RESEARCH

The future of syndromic surveillance will rest entirely on evi- dence of value over time. Although enthusiasm will remain high for some indeterminate period, only a system with proven value will justify the effort and expense needed for long term, syn- dromic surveillance. Thus, the need for carefully constructed research on a variety of fronts is necessary. These should include, but are not limited to, assessing which variables are most critical for optimizing signal-to-noise ratios. In fact, weighted variables may be necessary because some data points may be more criti- cal than others to identify true clusters or outbreaks. Currently, most of the variables used in syndromic surveillance are cho- sen based on the convenience of electronic reporting for those inputting data. For example, electronic emergency department chief complaints may be immediately available, but discharge diagnoses may be more accurate in determining the appropriate syndrome category. Future systems may need to integrate both types of data for timely and accurate analysis.65 Although this may be necessary in the short term, sustainability is question- able without evidence of efficacy. Another issue that requires careful investigation is the validity of transmitted data, particu- larly because electronic transmission typically proceeds without human involvement to determine accuracy. Furthermore, it is not only important to know the accuracy of the data on an objective level, but also to know if the expected inaccuracies of certain electronic data such as ICD-10-CM coding will have an effect on outcome. Although inaccuracies may exist in ICD-10- CM coding, the ability to collapse codes into syndrome groups may obviate the added expense of improving accuracy at the individual patient level.

In addition to questions regarding system efficacy and valid- ity, the technology that supports mathematical data analysis remains unproven as well. Approaches such as cumulative sums, smart scores, standard deviations, rolling averages, and anomaly detection algorithms are just a few of the methodologies that have been used for syndromic surveillance. However, there are currently insufficient data to define which of these systems pro- vides the best information to protect the public health, provide early event detection, or enhance situational awareness. Clearly, this will be an iterative process requiring a collaborative effort by those in the mathematical, clinical, infectious diseases, and

public health communities. These groups must perform the nec- essary applied and theoretical research that will allow the best use of data gathered by these systems.

Although most of the design and funding for syndromic surveillance is currently based upon early event detection and situational awareness for intentional biological events, proof of value for detection of nonbioterrorism related outbreaks will be necessary. This is particularly true when globalization increases the number of diseases moving from continent to continent and newly identified emerging infectious diseases (EIDs) appear in human populations (see Chapter 6). The number of EIDs per year is difficult to discern, however the emergence of the severe acute respiratory syndrome virus in 2002 exemplifies the speed with which global spread of a new pathogen can occur. The ultimate value of syndromic surveillance rests on its ability to define illness prior to a definitive diagnosis. This is particularly important in the case of EIDs, when diagnosis of disease may be delayed by uncharacterized etiologies. Syndromic surveillance used during the severe acute respiratory syndrome pandemic has revealed challenges in the implementation of these systems, including data acquisition and integration,66 and indicates that major research efforts are needed.

Syndromic surveillance is necessary because of difficulty establishing a diagnosis in a timely manner for human infec- tious diseases. Although these systems can presumably be useful for inhibiting the global spread of new pathogens, this lack of rapid diagnosis certainly is not ideal. It is therefore critical that a major research effort be sustained to develop systems for rapid diagnosis of infectious diseases.67 At best, syndromic surveil- lance is a temporizing measure that uses surrogate information in place of diagnostic data because such data are not available in a timely manner. This affects both the public’s health as well as an individual’s health because treatment is often based on poor information prior to a definitive diagnosis.

There are at least three types of tests that show promise for rapid diagnoses. The first is the specific test for a single entity. This is exemplified by rapid diagnostic testing for influenza and rapid polymerase chain reaction testing for methicillin-resistant Staphylococcus aureus. Although there may be issues regarding the sensitivity and specificity of the rapid diagnostic test for influenza, it does represent a point-of-care test that will identify a specific disease that can spread quickly through the popula- tion. It is important both for the individual patient with regard to therapy, and for the public health, because public health authorities may need to implement broader mitigation strate- gies for an influenza outbreak. Rapid polymerase chain reac- tion testing for single organisms, such as methicillin-resistant S. aureus, can be important for individual patients as well as for identification of case clusters. Although methicillin-resistant S. aureus may not be the most critical organism in screen- ing for an intentional biological event, this technology repre- sents an important tool for detecting a variety of organisms. Intense research is needed to define such testing methodolo- gies, to validate the results, and to define clearly the sensitivity and specificity of the tests for individual patients and public health use.

A second essential area for research is rapid diagnostic test- ing for clinical syndromes. For instance, if patients have a set of symptoms indicating a respiratory syndrome, it might be more useful for surveillance purposes to rapidly establish a diagnosis of each person’s illness. Optimally this would require a single

C op

yr ig

ht ©

2 00

9. C

am br

id ge

U ni

ve rs

ity P

re ss

. A ll

rig ht

s re

se rv

ed .

SY N D RO M I C SU RV E I L L A N C E ■ 171

sample from the patient (such as sputum or swab) that could be immediately tested for a large array of pathogens includ- ing viruses, bacteria and fungi. These microarray tests would be based on components of the organisms themselves. Advan- tages of this type of analysis include 1) the large number of microbes (thousands) that could be detected by a single assay, 2) the ability to use various patient samples for testing (e.g., blood, urine, stool, and tissue samples), 3) providing an unbi- ased disease assessment independent of provider variability, and 4) establishing a disease diagnosis independent of detecting the host response, such as antibody production. However, arrays that include both microbial and host–response probes can also be used to identify disease etiologies. With widespread use, such assays could be relatively inexpensive, developed for a variety of syndrome types, and be of immediate benefit to the patient as well as to public health for surveillance purposes. Further advances in sample preparation and analytic detection can potentially lead to point-of-care testing using microarrays for the rapid diagnosis of infectious diseases outside the hospital or laboratory setting. Research into the feasibility, usability, and validity of diagnos- tic microarray testing is also critical. Based on the results of such studies, further rapid testing could be developed to deter- mine the presence of clinically and epidemiologically impor- tant markers related to specific organisms, such as antibiotic resistance genes, thereby improving both individual and public health. This approach may not always allow for diagnosis of newly emerging pathogens, but it would highlight a group of patients with a specific syndrome for which an array of diseases has been ruled out, and therefore, may require attention by public health entities.

A third critical area for future research is diagnostic testing to detect cases in the postexposure, presymptomatic stage of an infectious disease (i.e., the incubation period). Generally, this utilizes a single diagnostic array to examine thousands of host markers such as proteins, peptides, or nucleic acids, usually from a blood or tissue sample. This can be used to define specific results primarily based on the presence or absence of individual markers that identify a specific illness by using population-based information or previous samples from a single patient. Anoma- lies may represent divergence from a normal state of health or a specific illness. Such a finding could lead to a diagnosis or to specific testing that would establish a definitive diagnosis. These tests may be of particular value if there is evidence of disease spread in the population, a positive finding from a sensor such as a Biowatch monitor, or evidence of unusual plant or animal illness. Although this technology may seem futuristic and less well defined, it is a critical area for research, particularly as new microbes are discovered and advanced biotechnology becomes available to the bioterrorist community.

With all of these research endeavors, the key element for success will not be cost, workload, or political intent. The key element will be value at the individual, public health, or national and international security levels. Sustainability will be dependent on the value of any of these technologies, to prevent and mitigate disease or improve individual and public health. For existing systems, it is often unclear what public health actions should be taken when the technology suggests an emerging infectious disease or bioterrorism event. Detection must be coupled with the initiation of appropriate behaviors that reduce morbidity and mortality. Research and program evaluation will be the hallmarks of this effort.

REFERENCES

1. http://www.eurosurveillance.org/ViewArticle.aspx?ArticleId= 667. Accessed November 29, 2008.

2. http://www.css.drdc-rddc.gc.ca/symposium/symposium/2008/ 06-0234TA-eng.asp. Accessed November 29, 2008.

3. http://www.chi.unsw.edu.au/CHIweb.nsf/pageprintfriendly/ Syndromic%20Surveillance?opendocument. Accessed Novem- ber 29, 2008.

4. http://www.invs.sante.fr/agenda/syndromic surveillance eu/ information.htm. Accessed November 29, 2008.

5. http://www.eurosurveillance.org/ViewArticle.aspx?ArticleId= 668. Accessed November 29, 2008.

6. http://www.ncbi.nlm.nih.gov/pubmed/16796513. Accessed November 29, 2008.

7. http://www.cdc.gov/mmwr/pdf/wk/mm54su01.pdf, pg 47. Ac- cessed November 29, 2008.

8. http://www.biomedcentral.com/1471-2458/8/18. Accessed November 29, 2008.

9. Bravata D, McDonald K, Smith W, et al. Systematic review: surveillance systems for early detection of bioterrorism-related diseases. Ann Intern Med. 2004;140(11):910–922.

10. Green M, Kaufman Z. Surveillance for early detection and mon- itoring of infectious disease outbreaks associated with bioter- rorism. Isr Med Assoc J. 2002:4(7):503–506.

11. Irvin C, Nouhan P, Rice K. Syndromic analysis of computerized emergency department patients’ chief complaints: an oppor- tunity for bioterrorism and influenza surveillance. Ann Emerg Med. 2003;41(4):447–452.

12. Begier E, Sockwell D, Branch L, et al. The National Capitol Region’s Emergency Department Syndromic Surveillance Sys- tem: do chief complaint and discharge diagnosis yield different results? Emerg Infect Dis. 2003;9(3):393–396.

13. Platt R, Bocchino C, Caldwell B, et al. Syndromic surveillance using minimum transfer of identifiable data: the example of the National Bioterrorism Syndromic Surveillance Demonstration Program. J Urban Health. 2003;80(2Suppl 1):i25–i31.

14. Lober W, Trigg L, Karras B, et al. Syndromic surveillance using automated collection of computerized discharge diagnoses. J Urban Health. 2003;80(2 Suppl 1):i97–i106.

15. Buehler J, Berkelman R, Hartley D, Peters C. Syndromic surveil- lance and bioterrorism-related epidemics. Emerg Infect Dis. 2003;9(10):1197–1204.

16. Centers for Disease Control and Prevention. What is syndromic surveillance? MMWR. 2004;53(Suppl):7–11.

17. Centers for Disease Control and Prevention. New York City syndromic surveillance systems. MMWR. 2004;53(Suppl): 25–27.

18. Centers for Disease Control and Prevention. Progress in under- standing and using over-the-counter pharmaceuticals for syn- dromic surveillance. MMWR. 2004;53(Suppl):117–122.

19. Centers for Disease Control and Prevention. Use of Medi- caid prescription data for syndromic surveillance – New York. MMWR. 2005;54(Suppl):31–4.

20. Centers for Disease Control and Prevention. Poison con- trol center-based syndromic surveillance for foodborne illness. MMWR. 2005;54(Suppl):35–40.

21. Centers for Disease Control and Prevention. Monitoring over-the-counter medication sales for early detection of disease outbreaks – New York City. MMWR. 2005;54(Suppl):41–46.

22. Centers for Disease Control and Prevention. Experimental surveillance using data on sales of over-the-counter medi- cations – Japan, November 2003–April 2004. MMWR. 2005; 54(Suppl):47–52.

C op

yr ig

ht ©

2 00

9. C

am br

id ge

U ni

ve rs

ity P

re ss

. A ll

rig ht

s re

se rv

ed .

172 ■ GA RY A. RO S E L L E

23. Centers for Disease Control and Prevention. Increased antiviral medication sales before the 2005–06 influenza season – New York City. MMWR. 2006;55(10):277–279.

24. Vergu E, Grais R, Sarter H, et al. Medication sales and syn- dromic surveillance, France. Emerg Infect Dis. 2006;12(3):416– 421.

25. Hope K, Durrheim, DN, d’Espaignet ET, Dalton C. 2006. Syn- dromic surveillance: is it a useful tool for local outbreak detec- tion? J Epidemiol Community Health. 60:374–375.

26. Chapman W, Christensen L, Wagner M, et al. Classifying free- text triage chief complaints into syndromic categories with natural language processing. Artif Intell Med. 2005;33(1):1– 10.

27. Centers for Disease Control and Prevention. Taming variabil- ity in free text: application to health surveillance. MMWR. 2004;53(Suppl):95–100.

28. Lombardo J, Burkom H, Elbert E, et al. A systems overview of the electronic surveillance system for the early notification of community-based epidemics (ESSENCE II). J Urban Health. 2003;80(2 Suppl 1):i32–i42.

29. Centers for Disease Control and Prevention. ESSENCE II and the framework for evaluating syndromic surveillance systems. MMWR. 2004;53(Suppl):159–165.

30. Vourc’h G, Bridges V, Gibbens J, et al. Detecting emerging dis- eases in farm animals through clinical observations. Emerg Infect Dis. 2006;12(2):204–210.

31. Centers for Disease Control and Prevention. Information system architectures for syndromic surveillance. MMWR. 2004;53(Suppl):203–208.

32. Forslund D, Joyce E, Burr T, et al. Setting standards for improved syndromic surveillance. IEEE Eng Med Biol Mag. 2004;23(1):65– 70.

33. Mandl K, Overhage JM, Wagner M, et al. Implementing syn- dromic surveillance: a practical guide informed by the early experience. J Am Med Inform Assoc. 2004;11(2):141–150.

34. Reis B, Mandl K. Time series modeling for syndromic surveil- lance. BMC Med Inform Decis Making. 2003;3:2.

35. Kleinman K, Lazarus R, Platt R. A generalized linear mixed models approach for detecting incident clusters of disease in small areas, with an application to biological terrorism. Am J Epidemiol. 2004;159(3):217–224.

36. Reis B, Mandl K. Syndromic Surveillance: The effects of syn- drome grouping on model accuracy and outbreak detection. Ann Emerg Med. 2004;44(3):235–241.

37. Feinberg S, Shmueli G. Statistical issues and challenges associ- ated with rapid detection of bio-terrorist attacks. Statist Med. 2005;24:513–529.

38. Hutwagner L, Thompson W, Seeman G, Treadwell T. A simu- lation model for assessing aberration detection methods used in public health surveillance for systems with limited baselines. Statist Med. 2005;24:543–550.

39. Centers for Disease Control and Prevention. Bivariate method for spatio-temporal syndromic surveillance. MMWR. 2004;53(Suppl);61–66.

40. Centers for Disease Control and Prevention. Role of data aggre- gation in biosurveillance detection strategies with applications from ESSENCE. MMWR. 2004;53(Suppl):67–73.

41. Centers for Disease Control and Prevention. Scan statistics for temporal surveillance for biologic terrorism. MMWR. 2004;53(Suppl):74–78.

42. Centers for Disease Control and Prevention. Approaches to syn- dromic surveillance when data consist of small regional counts. MMWR. 2004;53(Suppl):79–85.

43. Centers for Disease Control and Prevention. Measuring outbreak-detection performance by using controlled feature set simulations. MMWR. 2004;53(Suppl):130–136.

44. Centers for Disease Control and Prevention. Benchmark data and power calculations for evaluating disease outbreak detection methods. MMWR. 2004;53(Suppl):144–151.

45. Kleinman K, Abrams A, Kulldorff M, Platt R. A model-adjusted space-time scan statistic with an application to syndromic surveillance. Epidemiol Infect. 2005;133(3):409–419.

46. Centers for Disease Control and Prevention. Use of multiple data streams to conduct Bayesian biologic surveillance. MMWR. 2005;54(Suppl):63–69.

47. Centers for Disease Control and Prevention. Deciphering data anomalies in BioSense. MMWR. 2005;54(Suppl):133– 139.

48. Najmi A-H, Magruder S. An adaptive prediction and detection algorithm for multistream syndromic surveillance. BMC Med Inform Decis Making. 2005;5:33.

49. Centers for Disease Control and Prevention. Syndromic surveil- lance for bioterrorism following the attacks on the World Trade Center – New York City, 2001. MMWR. 2002;51(SI):13– 15.

50. Das D, Weiss D, Mostashari F, et al. Enhanced drop-in syn- dromic surveillance in New York City following September 11, 2001. J Urban Health. 2003;80(2 Suppl 1):i76–i88.

51. Gesteland P, Gardner R, Tsui F-C, et al. Automated syndromic surveillance for the 2002 Winter Olympics. J Am Med Inform Assoc. 2003;10(6):547–554.

52. Centers for Disease Control and Prevention. Surveillance for early detection of disease outbreaks at an outdoor mass gather- ing – Virginia, 2005. MMWR. 2006;55(3):71–74.

53. Muscatello D, Churches T, Kaldor J, et al. An automated, broad- based, near real-time public health surveillance system using presentations to hospital emergency departments in New South Wales, Australia. BMC Public Health. 2005;5:141.

54. Marx M, Rodriguez C, Greenko J, et al. Diarrheal illness detected through syndromic surveillance after a massive power outage: New York City, August 2003. Am J Pub Health. 2006;96(3):547– 553.

55. Centers for Disease Control and Prevention. Syndromic surveil- lance at hospital emergency departments – southeastern Vir- ginia. MMWR. 2004;53(Suppl):56–58.

56. Centers for Disease Control and Prevention. Hospital admis- sions syndromic surveillance – Connecticut, October 2001-June 2004. MMWR. 2005;54(Suppl):169–173.

57. Centers for Disease Control and Prevention. Framework for evaluating public health surveillance systems for early detection of outbreaks. MMWR. 2004;53(No. RR-5):1–11.

58. Mostashari F, Hartman J. Syndromic surveillance: a local per- spective. J Urban Health. 2003;80(2 Suppl 1):i1–i7.

59. Centers for Disease Control and Prevention. High-fidelity injec- tion detectability experiments: a tool for evaluating syndromic surveillance systems. MMWR. 2005;54(Suppl):85–91.

60. Centers for Disease Control and Prevention. Initial evaluation of the early aberration reporting system – Florida. MMWR. 2005;54(Suppl):123–130.

61. Centers for Disease Control and Prevention. Evaluation of syn- dromic surveillance based on National Health Service direct derived data – England and Wales. MMWR. 2005;54(Suppl): 117–122.

62. Centers for Disease Control and Prevention. An evaluation model for syndromic surveillance: assessing the performance of a temporal algorithm. MMWR. 2005;54(Suppl):109–115.

C op

yr ig

ht ©

2 00

9. C

am br

id ge

U ni

ve rs

ity P

re ss

. A ll

rig ht

s re

se rv

ed .

SY N D RO M I C SU RV E I L L A N C E ■ 173

63. Centers for Disease Control and Prevention. Simulation for assessing statistical methods of biologic terrorism surveillance. MMWR. 2005;54(Suppl):101–108.

64. Stoto M, Schonlau M, Mariano L. Syndromic surveillance: is it worth the effort? CHANCE. 2004;17(1):19–24.

65. Chapman WW, Dowling JN, Wagner MM. Classification of emergency department chief complaints into 7 syndromes: a retrospective analysis of 527, 288 patients. Ann Emerg Med. 2005;46(5):445–455.

66. Shih F-Y, Yen M-Y, Wu J-S, et al. Challenges faced by hos- pital healthcare workers in using a syndrome-based surveil- lance system during the 2003 outbreak of severe acute res- piratory syndrome in Taiwan. Infect Control Hosp Epidemiol. 2007;28(3):354–357.

67. Turner K, Shaw K, Coleman D, Misrachi A. Augmentation of influenza surveillance with rapid antigen detection at the point- of-care: results of a pilot study in Tasmania, 2004. Commun Dis Intell. 2006;30(2):201–204.

C op

yr ig

ht ©

2 00

9. C

am br

id ge

U ni

ve rs

ity P

re ss

. A ll

rig ht

s re

se rv

ed .