Biosurveillance in Outbreak Investigations

Abstract

Following the terrorist attacks of September 11 and the anthrax attacks in 2001, public health entities implemented automated surveillance systems based on disease syndromes for early detection of bioterror events and to increase timeliness of responses. Despite widespread adoption, syndromic surveillance systems' ability to provide early notification of outbreaks is unproven, and there is little documentation on their role in outbreak response. We hypothesized that biosurveillance is used in practice to augment classical outbreak investigations, and we used case studies conducted in 2007-08 to determine (1) which steps in outbreak investigations were best served by biosurveillance, and (2) which steps presented the greatest opportunities for improvement. The systems used in the case studies varied in how they functioned, and there were examples in which syndromic systems had identified outbreaks before other methods. Biosurveillance was used successfully for all steps of outbreak investigations. Key advantages of syndromic systems were sensitivity, timeliness, and flexibility and as a source of data for situational awareness. Limitations of biosurveillance were a lack of specificity, reliance on chief complaint data, and a lack of formal training for users. Linking syndromic data to triage notes and medical chart data would substantially increase the value of biosurveillance in the conduct of outbreak investigations and reduce the burden on health department staff.

The ability of syndromic surveillance systems to provide early notification of outbreaks is unproven, and there is little documentation on their role in outbreak response. The authors used case studies to determine (1) which steps in outbreak investigations were best served by biosurveillance, and (2) which steps presented the greatest opportunities for improvement. Key advantages of syndromic systems were sensitivity, timeliness, and flexibility and as a source of data for situational awareness. Limitations of biosurveillance were a lack of specificity, reliance on chief complaint data, and a lack of formal training for users.

Following the terrorist attacks of September 11 and the anthrax attacks in 2001, Congress allocated funding to federal agencies and states to develop and implement automated disease surveillance systems. These systems had 2 main purposes: (1) to serve as early-warning systems for bioterrorism events, and (2) to increase the timeliness of public health responses to such attacks.¹ The surveillance systems initially relied on hospital emergency department (ED) prediagnostic (chief complaint) data, which were coded into groups of signs and symptoms associated with bioterrorism agents. The resulting syndromes, such as fever and rash and neurological, gastrointestinal, and respiratory symptoms, were then tracked over time. The syndromic surveillance systems issued “alerts” for further investigation when cases exceeded a threshold level set by mathematical algorithms. Over time, syndromic surveillance systems included data from pharmacies, laboratories, poison control centers, and other sources, and they commonly have been referred to as “biosurveillance systems” to encompass these additional data types.² Biosurveillance systems have been widely adopted under this federal funding by federal, state, and local public health entities—mainly the Centers for Disease Control and Prevention (CDC), health departments, and some universities—for public health surveillance.^3–5 We use the term biosurveillance to mean “automated monitoring of existing health data sources to identify trends that may indicate naturally occurring or intentional disease outbreaks.”⁶ The focus of this article is primarily syndromic surveillance systems based on hospital emergency department data.

Despite high rates of adoption, the utility of biosurveillance/syndromic surveillance has been questioned.^{2, 7–11} The absence of large-scale bioterror events has resulted in biosurveillance systems' early-detection capabilities being untested for their primary objective. Simulations and reviews have been conducted to evaluate the likely success of biosurveillance in detecting outbreaks earlier than traditional methods of disease surveillance, with unclear results.^10,12–15

Less attention has been given to evaluating biosurveillance for improving public health response to outbreaks, regardless of how the outbreaks are first identified.¹¹ Investigation of disease outbreaks is one of the central responsibilities of public health authorities. A standard set of steps comprise a disease outbreak investigation (Table 1). Depending on the characteristics of a particular outbreak, the order of the steps may change, multiple steps may occur simultaneously, or steps may not occur at all.¹⁶ First, the diagnosis is verified to rule out error or misdiagnosis. Verification typically requires some combination of medical chart review, laboratory test results, and follow-up with patients. Second, the outbreak is confirmed as a greater-than-expected number of cases within a specified time period, population, and geographical area. Third, a case definition is applied, and the maximum number of cases meeting that definition is identified. Fourth, the data are described by identifying the population affected, the geographical dispersion, and the number or rate of cases over time. Fifth, one or more hypotheses to explain the epidemic are generated, and a quantitative observational study is conducted to test the hypothesis. Sixth, control measures to prevent additional exposures and protect at-risk populations are implemented. Finally, findings and recommendations are shared with stakeholders, possibly through the media or written reports.

Table 1.

Steps in an Outbreak Investigation

Step in Outbreak Investigation ^a	Typical Occurrence During the Course of the Outbreak Investigation ^b
1. Verify the diagnosis	Early
2. Confirm the outbreak	Early
3. Verify and count cases	Early
4. Orient the data	Begin early to middle
5. Formulate and test hypothesis	Middle to late—may not occur if etiology, cause, manner of transmission are known
6. Implement control measures	Early, middle or late—may not occur in point-source outbreak
7. Communicate findings	Early, middle, late, and postoutbreak

Modified from Ref. 16.

Depending on outbreak characteristics, the steps may occur in a different order, simultaneously, or not at all.

We anticipated that biosurveillance could be used effectively to improve outbreak investigations and that, in practice, biosurveillance systems were achieving part of their original purpose despite a lack of known bioterrorism events. We used in-depth qualitative interviews and case studies to determine which steps in outbreak investigations were best served by biosurveillance and which steps presented the greatest opportunities for improvement. State and local health departments can use the results of the case studies to identify additional ways in which relatively simple biosurveillance systems can be used in identifying and investigating outbreaks.

Methods

This study was funded under a cooperative agreement with CDC, with the goal to assess the effect of biosurveillance on public health preparedness, early detection, situational awareness, and response to public health threats. In addition to objectives regarding the implementation, costs, and data quality from biosurveillance systems, the evaluation addressed the operational components and effectiveness of biosurveillance systems to respond to public health threats.

Three case studies were conducted to examine how local public health systems are using biosurveillance systems. Case studies were purposely chosen for maturity of the system and for recent use of the system during a significant public health event. Case studies were conducted of the North Carolina Disease Event Tracking and Epidemiologic Collection Tool (NC DETECT), the Cook County Department of Public Health's system (Chicago, Illinois), and the regional reporting network in north central Texas that is housed in the offices of the Tarrant County Public Health Department (Fort Worth, Texas). The biosurveillance systems were examined for general routine uses of biosurveillance and for the use of biosurveillance during a significant public health event. These case studies were part of a larger study encompassing an economic evaluation and data quality evaluation. NC DETECT is both a state-level and local system, and both uses were considered for the study. The Cook County case study was chosen in part because of the extensive media attention to a 2007 itch-mite outbreak and therefore focused only on the outbreak investigation, so a narrower group of informants was interviewed.

Case studies are used to investigate entities in their real-world environment using numerous sources of data.¹⁷ The use of the case studies was intended to describe examples of the myriad ways in which state and local public health entities were using biosurveillance systems to address public health events, including the investigations of disease outbreaks.

For each case study, RTI evaluators worked with a lead contact in each agency to identify key informants who were currently using or supporting biosurveillance activities for that agency (eg, epidemiologists, health officers, public information officers, public health preparedness coordinators, hospital infection preventionists). A select number of key informants were invited to participate in interviews.

The evaluators developed a general interview guide that included open-ended questions organized around the key areas of inquiry for the evaluation. Topics focused on the routine and event-based use of biosurveillance systems in the health department and key informant perspectives on the utility of systems for event detection, situation awareness, and response. The interview guide was customized to include the most relevant questions based on the position and role of the key informant.

We conducted interviews with NC DETECT during summer 2007, for the Tarrant County case study during winter 2007, and in Cook County during fall 2008. Interviews were approximately 60 to 90 minutes long and were conducted in person when possible or otherwise by telephone. Group interviews were conducted for users of similar types who worked in the same establishment; for example, several county epidemiologists who are regular public health users of Tarrant County's biosurveillance system were interviewed as a group. All interviews were recorded, with the interviewees' permission, to ensure the accuracy of the transcripts. In total, 28 interviews were conducted involving 44 individuals (Table 2). This study was deemed exempt by RTI International's institutional review board. Informed consent was obtained orally from all respondents before the interviews commenced.

Table 2.

Type and Number of Case Study Informants

	Number Interviewed
Key Informant Roles	NC DETECT	Tarrant County	Cook County
Biosurveillance system staff	4^a	3
Hospital infection control managers/epidemiologists	7	5
County epidemiologists		7	1
County public health information officers			1
County surveillance/public health preparedness coordinators	1		1
County public health directors/section chiefs			2
Regional epidemiologists		2
Regional health directors/authorities		3
State epidemiologists/section chiefs/directors	4
Other	2		1
Total	18	20	6

University-based developers and managers of the system.

The evaluators created written transcripts from the interview recordings, which were reviewed by the key informants prior to analysis to ensure accuracy and completeness. The evaluators analyzed the transcripts qualitatively by applying a technique known as content analysis, which groups or categorizes large sections of text into smaller content categories based on explicit rules of coding.¹⁸ In this study, codes based on the themes and research questions of interest were established a priori. Data were analyzed using N'Vivo version 7 (QSR International), a software application used for qualitative data analysis. One analyst coded the entire transcript set, while a second checked the coding for consistency and relevance.

Results

Automated Surveillance Systems Used

The 3 case-study sites used a variety of automated systems for biosurveillance. North Carolina developed NC DETECT, beginning in 1999, as previously described.¹⁹ At the time of the study, NC DETECT received chief complaint data from 100 of 112 hospital EDs in the state, triage notes from a portion of EDs, and wildlife disease data and was in the process of adding poison control center data to its twice-daily automatic data feeds. At the time of the study, NC DETECT did not report to BioSense. The Cook County Department of Public Health used the Electronic Surveillance System for the Early Notification of Community-based Epidemics (ESSENCE) (Johns Hopkins University, Baltimore, MD), which fed data to BioSense, although BioSense was not routinely monitored by the health department staff. The Tarrant County Public Health Department used both ESSENCE and the Real Time Outbreak and Disease Surveillance system (RODS) (University of Pittsburgh, Pittsburgh, PA), as well as BioSense.

In all cases, the biosurveillance systems used were monitored daily by users, who were generally epidemiologists at the county and state levels. Users at hospitals were either epidemiologists or infection preventionists. Access rights to the data were defined by the jurisdiction and role of the informants, which influenced the amount and level of data that each could view. Access rights varied from data for a single hospital to data from the entire state. The level of detail also varied, in that some users could view only case counts, while others could view medical records of individuals who were case-patients associated with a system-generated alert.

Outbreak Investigations: Early Detection

Multiple informants shared experiences with specific outbreaks identified by biosurveillance systems. For example, in North Carolina, university students were exposed to a foodborne pathogen at a function immediately preceding a holiday vacation period of several weeks. The students presented with gastrointestinal (GI) symptoms to EDs across the state, and the outbreak was identified through written comments in the triage notes mentioning that the patients were students at the university.

In Cook County, an outbreak of extremely itchy rash caused by itch mites affected thousands of people in the Chicago suburbs in the late summer of 2007. ESSENCE showed a dramatic increase in rash illness days before it was reported to the health department. Epidemiologists began the investigation when the health department was notified of the number of cases of an unusual rash illness by an ED physician. Because many of the chief complaint entries lacked specificity and were limited to the word “rash,” and triage note data were not available, the increase in cases was not investigated when first detected by biosurveillance. Rather, the public health investigation was initiated after the health department was alerted to the problem.

Despite some successes in which outbreaks were identified more quickly by biosurveillance systems, users were hesitant to fully endorse the systems for early event detection (Table 3). Biosurveillance's reliance on syndrome data derived from ED chief complaint data had high sensitivity but low specificity. While the systems were good at identifying true positive cases for a syndrome, often a large number of false positives were identified as well. Small outbreaks that would be hidden in reports of common syndromes were perceived to be less likely to be captured by current biosurveillance methods. Although several foodborne illness outbreaks had been identified by biosurveillance systems, many users thought that GI illness outbreaks were unlikely to be identified through biosurveillance because GI illness is so widespread. For diseases in which a single case would constitute an outbreak (eg, anthrax), users thought the health department was more likely to be informed by a physician than by a biosurveillance alert. Syndromic surveillance was generally considered more useful for outbreaks of severe disease than for milder diseases, because data were coming from EDs. One informant noted that in areas in which EDs also provide primary care for segments of the population without primary care providers (especially in rural areas), and outside of normal business hours, mild cases of illness would be detected by ED-based surveillance.

Table 3.

Advantages and Barriers to the Utility of Biosurveillance Systems in the Investigation of Outbreaks

Step in Outbreak Investigation	Advantages of Biosurveillance	Barriers Encountered
Early detection	• For dramatic increases in case counts, biosurveillance can detect outbreaks earlier than traditional surveillance.• Automated alerts issued when case counts reach above a mathematic threshold.• Syndromes have high sensitivity.	• Syndromes, particularly gastrointestinal, are nonspecific.• Small outbreaks do not produce an alert.• Utility of ED data may limit detection to severe illness.
Verify the diagnosis	• Additional automated data from triage note, medical chart, laboratory increase specificity.• Cases can be updated automatically.	• Some systems lack more specific data sources.
Confirm the outbreak	• Use of text queries enables quick case counts.• Algorithms can be modified to increase specificity of alerts.• Easy to rule out false alerts with access to triage note or medical record.• Background incidence rates of syndromes are known.	• False alerts can be numerous.• Investigating alerts requires hospital staff time.
Count cases	• Use of filters and queries increases ease.• Filters and queries can be applied prospectively and retrospectively.	• Syndrome data lack specificity, leading to inclusion of false positives.• Diagnostic data not immediately available.• Less severe cases may not be identified.
Orient the data	• “Situational awareness”• Geographical dispersion is viewable.• Presentation of data in time series is standard.	• Static unit of time for time series may not be meaningful.• Age, race, ethnicity data are not available in all systems.
Formulate and test hypothesis	• Triage note and ability to query on keywords enhances ability to determine unusual circumstances and commonalities.• Mapping of cases enables geographic considerations in hypothesis generation.	• Unknown; health departments did not conduct formal hypothesis testing for the outbreaks discussed in the case studies.
Implement control measures	• Situational awareness is improved at state, local, institutional levels.	• None identified.
Communicate findings	• Increased speed of identifying, confirming, and investigating outbreaks allows information to be shared more rapidly.• Decrease burden on hospitals from mild illnesses.• Improves credibility with the public.• Ongoing collaboration improves communication between hospitals and health departments.	• Syndromic data are difficult to interpret and should not be released to the media or public.

Verifying the Diagnosis

Biosurveillance systems that relied on ED data were variably useful for verifying diagnoses. In systems in which the triage note was available, users were better able to verify a syndrome of interest than users who relied solely on chief complaint data. The advantage of using biosurveillance to verify the diagnosis depended on the user type in North Carolina. Because of jurisdictional boundaries and privacy concerns, some users in North Carolina hospitals were not able to identify medical record numbers from the systems even when cases were at their hospital. Other users at the state level were able to directly view an electronic medical record for case-patients linked to an alert. Some hospital data were updated through the system automatically, so as laboratory tests were completed they were viewable in the updated records by a very select set of users. Incorporating this level of data into biosurveillance was unusual. Although data on laboratory tests ordered by ED physicians would not have provided a diagnosis to biosurveillance users, several of them indicated that those data would help them with the possible diagnoses while still providing near real-time data.

Confirming the Outbreak

The utility of biosurveillance in confirming the presence of an outbreak depended on many of the same factors as for early identification of outbreaks. In confirming an outbreak, epidemiologists must rule out false alerts, or alerts that do not represent true case clusters. In general, ruling out false alerts did not place a large burden on users in terms of effort or time required. One user reported that there had been a month during which the system he used produced 90 alerts; no other informant reported high numbers of alerts (the informants did not define “high”). Baseline rates of syndromes are a key factor in the issuance of alerts and can require years of data gathering to be well developed. Users were pleased that, as baseline data were refined over time, the number of false alerts decreased. The algorithms also affected the issuance of alerts. When algorithms were modified, the number of alerts noticeably changed. When fewer alerts occurred, epidemiologists who followed up on the alerts had more success in receiving detailed information on cases from hospitals, possibly because the hospital employees perceived that the alerts were more likely to indicate true outbreaks.

Users did not always know exactly what the threshold for an alert represented or how it was calculated. This was seen as a barrier to investigating possible outbreaks by some epidemiologists, who thought that they would have more cooperation from hospitals if their hospital contacts better understood what the alerts signified. In systems in which users could query the data on keywords, epidemiologists reported more success confirming outbreaks because the system's capacity to screen out false alerts increased.

Counting Cases

Cases of illness are counted during an outbreak to (1) estimate the magnitude of the outbreak, (2) identify potential cases, and/or (3) count confirmed cases, depending on the particular outbreak situation. Despite the lack of specificity of biosurveillance system syndromes, users considered them very helpful in counting cases. In Cook County, biosurveillance enabled epidemiologists to estimate the increase in the number of cases of rash illness over time. In this outbreak, the ability to query the data based on text searches improved the positive predictive value (the proportion of identified cases that actually were affected by the itch mites) of the system.

Across the health departments in the case studies, the ability to filter biosurveillance data using keywords was critical to the success of identifying cases. For example, in North Carolina this ability was used to find cases associated with a nationwide outbreak of salmonellosis associated with peanut butter and to search for cases of Escherichia coli O157:H7 among children who had visited a petting zoo. Text queries or filters can quickly be set up and used to look for cases prospectively and retrospectively. These advantages helped offset the inherent problems that arose from using syndromic data to identify cases of a particular illness, mainly the lack of specificity and unavailability of diagnostic data. Epidemiologists who investigated the itch-mite outbreak in Cook County noted that, during the height of the outbreak, the incidence of cases rose every weekend and fell during the week. This phenomenon was attributed to the illness being mild enough that when primary care physicians' offices were open on weekdays, fewer patients presented to hospital EDs. Some users reported not knowing how to do queries, and multiple users said that formal training in using the systems was inadequate.

Orienting the Data

Biosurveillance systems were considered invaluable for describing the data in terms of the population affected, the number of cases over time, and the geographical dispersion of the cases. This step includes “situational awareness,” which was mentioned by most users in all of the case studies as a primary use of biosurveillance systems. All of the systems could present data in time series, allowing the user to visualize peaks and valleys in disease incidence. Systems that did not allow the user to define the unit of time were considered less useful for orienting the data, because the most meaningful unit of time for analyzing outbreak data is highly dependent on the incubation period (for infectious diseases) and other factors.

The mapping function of biosurveillance systems was highlighted by multiple users as valuable. During the itch-mite outbreak, epidemiologists generated maps of patients' residential ZIP codes, which in turn was credited with saving the health department days in generating a hypothesis regarding the diagnosis and source of the outbreak. Epidemiologists in Tarrant County used the mapping function of their biosurveillance systems for multiple outbreaks, including identifying a school as a possible amplifier in an outbreak of shigellosis.

Biosurveillance systems also allow for concurrent analysis of data in terms of person and place. Hospital users were able to avoid surprise outbreaks when high-level data visualized geographically over time illustrated nonreportable infectious diseases like norovirus and influenza spreading across the state and toward their jurisdictions. Identifying populations in terms of age and race/ethnicity is possible, as in the foodborne illness outbreak among university students on holiday break. One informant noted that not all systems automatically include such demographic data on case-patients, which was considered a weakness.

Formulating and Testing Hypotheses

The availability of the triage notes and ability to conduct queries on keywords helped epidemiologists to formulate a hypothesis to explain a given outbreak. Unusual circumstances are likely to be included in the triage note, which generally contains far more information than the chief complaint field. The mapping function used for the itch-mite outbreak greatly increased the speed with which a hypothesis of itch mites as the source of the outbreak was developed. The mapping function also helped public health entomologists with the environmental investigation by assisting them in determining where to place traps for the mites for identification. Informants either did not mention or denied the conduct of a formal observational study to test hypotheses in the outbreak investigations mentioned in the case studies.

Implementing Control Measures

Biosurveillance helped state and local health departments track epidemics over time, and it was instrumental in the decision-making processes of health departments and hospitals to implement control measures to protect at-risk populations and prevent future cases of illness. Several hospital users reported that they used biosurveillance systems to track influenzalike illness in their institutions and that they used the data to determine when to provide measures to prevent illness spread to and from visitors. Such measures included providing hand sanitizer for visitors, prominently displaying respiratory hygiene signs, and restricting children from visiting patients. Biosurveillance was used in a similar way for norovirus outbreaks that occurred at many hospitals in North Carolina.

Communicating Findings

Multiple users reported that they routinely either read or generated reports for dissemination to stakeholders using their biosurveillance systems. The systems increased timeliness of sharing information with the public and saved health department resources. One health director reported that he could quickly respond to media inquiries, thereby avoiding having to delay his response because he was unaware of the issue in question. The system not only saved health department and hospital personnel time, it also underscored his credibility as a health authority for his jurisdiction.

In the itch-mite outbreak, health department personnel were able to use biosurveillance to confirm the event, map the patient residences by ZIP code, formulate a hypothesis, and, even without a confirmed diagnosis, hold a press conference within 48 hours of being notified of the first cluster of patients by an ED physician. The health department staff reported that they were confident of the data and found communicating with the hospitals and media quickly to be helpful in their management of the event. Once the public was aware that the rash was more of a nuisance than a communicable life-threatening disease, the weekly incidence of ED visits leveled off and then dropped. The informants suggested that as the public realized that the syndrome was mild, fewer people went to the hospitals for treatment.

Discussion

In this study, we conducted in-depth interviews to describe how biosurveillance systems have been used by state and local public health practitioners to identify and investigate outbreaks. We found evidence that biosurveillance can be successful in detecting disease outbreaks earlier than other surveillance methods. A rapid increase of rash illness caused by itch mites in a defined geographic region was first detected by syndromic surveillance, before it was reported by other means. Additionally, a small foodborne illness outbreak was identified by biosurveillance, despite the affected individuals having dispersed statewide before presenting to EDs for care. The populations covered by the case-study systems number well over 10 million individuals, and the health departments had been using biosurveillance systems for years, but the number of examples provided as early detection successes was small. The small number of early detection successes suggests that, in practice, biosurveillance in its form during these interviews may not have been well suited to meet its primary goal of early detection, as has been indicated in previous work.^1,2,7–9,14

Biosurveillance was also used as a tool for investigating outbreaks, once possible epidemics were detected by any means. All steps in classical outbreak investigations were at least in part amenable to biosurveillance. The strengths of biosurveillance were its sensitivity, timeliness, and flexibility and as a provider of situational awareness. Barriers were a lack of specificity, reliance on ED data, and the need for formal training. Both the strengths and weaknesses of biosurveillance confirmed previous findings.^3,13,20 Biosurveillance can therefore complement traditional, specialized data collection and analysis activities to inform outbreak investigations in myriad ways.

A limitation of the study was the small number of cases, which may not represent the full range of biosurveillance systems and users. The systems used by the case-study participants varied in their attributes, so it is difficult to make inferences on how biosurveillance systems are used nationwide, but we evaluated more and less sophisticated systems that show the range of what biosurveillance systems are capable of. Using qualitative data collection methods enabled a deeper exploration of how biosurveillance is used, although there are undoubtedly important factors that were missed. In addition, biosurveillance systems have evolved rapidly since the interviews and currently have more advanced capabilities and functions. Sophisticated current systems are probably even more useful for outbreak investigations than presented here.

The use of qualitative interviews resulted in some unexpected findings. In the itch-mite outbreak, in part because of the geographic data available in ESSENCE, the health department was able to communicate information about the outbreak to allay public fears even though there was no confirmed source of the outbreak. Additionally, biosurveillance users indicated that they do not spend a large portion of their time investigating false alarms, because false alarms were generally not frequent and they were easily investigated and dismissed. The need for a confirmed diagnosis and the investigation of false alarms were not important burdens, as has been suggested.^1,9,10,13

Recurring attributes that affected the usefulness of biosurveillance to investigate outbreaks were running queries, more detailed patient data, and education and training. Queries enabled users to more quickly confirm an outbreak or dismiss false alarms, verify the diagnosis, count cases, orient the data, and formulate a hypothesis. An inability to conduct text queries because of either system limitations or lack of training could slow investigations and lead to outbreaks being missed altogether. Training and education were needed for both users and triage nurses. Users who did not know how to conduct queries could not take full advantage of the tools and data available. Others opined that if infection control preventionists and nurses better understood how alerts were generated and how data flow, the chief complaint data in the system would be more complete and infection preventionists would be more willing to help investigate alerts. Nurses are an integral part of ED-based surveillance systems and should be empowered to use and improve them.^11,21

When data were limited to basic demographics and broad syndrome categories, biosurveillance systems were less useful. The users who were most supportive of their systems were able to query not just the chief complaint data but also the information-rich triage note. Filters could also be applied to the triage note quickly, with the filters then applied both prospectively and retrospectively. The few users who were able to obtain patient chart data linked to syndrome cases reported high levels of satisfaction with the system and perceived biosurveillance to be very useful.

In at least one case, a broad syndrome category was a strength of the systems rather than a weakness. Although the identification of a rash caused by itch mites could be considered a failure of the biosurveillance system because it was such a mild health event, we believe it instead illustrates that these systems can perform as intended. The Cook County experience demonstrated that a biosurveillance system can detect an outbreak of an unexpected and locally unknown health threat.

Health departments have made commendable efforts to increase the types of data sources that are incorporated into automated surveillance systems, including ED visits, outpatient visits, electronic medical records, over-the-counter medicine sales, poison control center calls, school or work absenteeism, 911 and emergency medical services calls, medical hotline calls, prescription medication sales,³ and wildlife diseases.¹⁹ Adding more data sources may be helpful for situational awareness and possibly even early event detection, but it will result in little improvement to public health capacity to investigate outbreaks. Most of the sources listed above remain truly syndromic data sources, with the possible exceptions of ED visits, outpatient visits, and health information exchanges. What is needed to improve outbreak investigations is automated access to detailed, patient-level data. Although patient data for syndrome cases are obtainable in some institutions, the systems are siloed so that valuable staff time is used to either call hospitals or check separate systems themselves. Linking patient-level clinical or laboratory data to syndromic data in a single, secure system would likely improve early event detection as well.^{8,15,20,22–24}

There are multiple barriers to linking the most useful data sources and systems. Jurisdictional boundaries, institutional and legal policies, systems that lack interoperability, and privacy concerns will require time and cooperation to overcome, but it has been done. Interoperable systems and access to data across jurisdictional lines will decrease time requirements for staff (eg, one biosurveillance user reported checking 3 different systems almost daily, because each had features the others lacked) to monitor systems and investigate outbreaks and false alarms.

Biosurveillance systems were not created specifically for use in outbreak investigations beyond early detection, but they are being used for them. Further improvements to biosurveillance system capabilities for use in outbreak investigations should be made by federal and state governments and system creators to address the weaknesses identified through these case studies. Such improvements would include making detailed data available to improve specificity, increasing the quality and availability of training on the use of biosurveillance systems, and improving the ability of the systems with regard to conducting text queries. Additionally, other novel ways in which biosurveillance systems are likely being used to improve public health services without increasing resource requirements should be widely shared among public health practitioners. Health departments are continuously being required to increase their duties while their available funding, staff, and other resources decrease.^20,25 Federal initiatives to encourage adoption of electronic health records among healthcare providers and sharing of data through health information exchanges represent an important opportunity to improve the utility of biosurveillance and capacity in health departments.^20,26 The consideration of public health data needs for biosurveillance in these initiatives is critical to our population's health.²⁷

Footnotes

Acknowledgments

The authors thank the key informants who were interviewed for the study, particularly the staff at the Carolina Center for Health Informatics at the University of North Carolina at Chapel Hill Department of Emergency Medicine, the North Carolina Division of Public Health, the Cook County Department of Public Health, and the Tarrant County Public Health Department. This study was funded by Cooperative Agreement No. U38 HK000009 from the Centers for Disease Control and Prevention. The opinions and assertions in this article are the private views of the authors and are not to be construed as official or as necessarily reflecting the views of the Centers for Disease Control and Prevention.

References

Buehler

, Berkelman

, Hartey

, Peters

. Syndromic surveillance and bioterrorism-related epidemics. Emerg Infect Dis, 2003 Oct 9,10:1197–1204.

Buehler

, Whitney

, Smith

, Prietula

, Stanton

, Isakov

. Situational uses of biosurveillance. Biosecur Bioterror, 2009; 7,2:165–177.

Buehler

, Sonricker

, Paladini

, Soper

, Mostashari

. Syndromic surveillance practice in the United States: findings from a survey of state, territorial, and selected local health departments. Adv Dis Surveill, 2008; 6,3:1–20.

Trust for America's Health. Ready or Not? 2010: Protecting the Public's Health from Disease, Disasters, and Bioterrorism, December 2010. http://healthyamericans.org/assets/files/TFAH2010ReadyorNot%20FINAL.pdf. 2011 December 13.

Uscher-Pines

, Farrell

, Cattani

et. al. A survey of usage protocols of syndromic surveillance systems by state public health departments in the United States. J Public Health Manag Pract, 2009; 15,5:432–438.

National Association of County and City Health Officials (NACCHO). Compendium of NACCHO Policy Statements. August 2011. http://www.naccho.org/advocacy/positions/upload/compendiumpolicystatements-final.pdf. 2012 November 5.

Buckeridge

. Outbreak detection through automated surveillance: a review of the determinants of detection. J Biomed Inform, 2006; 40:370–379.

Chretien

, Tomich

, Gaydos

, Kelley

. Real-time public health surveillance for emergency preparedness. Am J Pub Health, 2009; 99,8:1360–1363.

Reingold

. If syndromic surveillance is the answer, what is the question? Biosecur Bioterror, 2003; 1,2:77–81.

10.

Stoto

, Schonlau

, Mariano

. Syndromic surveillance: is it worth the effort? Chance, 2004; 17,1:19–24.

11.

Uscher-Pines

, Farrell

, Babin

et al. Framework for the development of response protocols for public health syndromic surveillance systems: case studies of 8 US states. Disaster Med Public Health Prep, 2009; 3,Suppl 1:S29–S36.

12.

Bravata

, McDonald

, Smith

et al. Systematic review: surveillance systems for early detection of bioterrorism-related diseases. Ann Intern Med, 2004; 140,11:910–922.

13.

Hope

, Durrheim

, d'Espaignet

, Dalton

. Syndromic surveillance: is it a useful tool for local outbreak investigation? J Epidemiol Community Health, 2006; 60:374–375.

14.

Nordin

, Goodman

, Kulldorff

et al. Simulated anthrax attacks and syndromic surveillance. Emerg Infect Dis, 2005; 11:1394–1398.

15.

Van den Wijngaard

, Van Asten

, Van Pelt

et al.

Syndromic surveillance for local outbreaks of lower-respiratory infections: would it work?

PLoS ONE, 2010; 5,4:e10406.

16.

Gregg

. Conducting a field investigation. Gregg

. Field Epidemiology, 2nd. New York: Oxford University Press, 2002; 62–77.

17.

Yin

. Case Study Research: Design and Methods, 4th. Thousand Oaks, CA: Sage, 2009; 18.

18.

Stemler

. An overview of content analysis. Practical Assessment, Research & Evaluation, 2001; 7,17 http://pareonline.net/getvn.asp?v=7&n=17. 2010 September 29.

19.

Kaydos-Daniels

, Rojas Smith

, Ising

et al. Biosurveillance and public health practice: a case study of North Carolina's NC DETECT system. Kass-Hout

, Zhang

. Biosurveillance: Methods and Case Studies. Boca Raton, FL: Chapman and Hall/CRC, 2010; 195–213.

20.

Toner

, Nuzzo

, Watson

et al. Biosurveillance where it happens: state and local capabilities and needs. Biosecur Bioterror, 2011; 9,4:321–330.

21.

Persell

, Robinson

. Detection and early identification in bioterrorism events. Fam Community Health, 2008; 31,1:4–16.

22.

Dembek

, Cochrane

, Pavlin

. Syndromic surveillance. Emerg Infect Dis, 2004; 10,7:1333–1334.

23.

Kaufman

, Pesik

, Meltzer

. Syndromic surveillance in bioterrorist attacks. Emerg Infect Dis, 2005; 11,9:1487–1488.

24.

Sintchenko

, Gallego

. Laboratory-guided detection of disease outbreaks: three generations of surveillance systems. Arch Pathol Lab Med, 2009; 133:916–925.

25.

Trust for America's Health. Ready or Not? 2011: Protecting the Public's Health from Disease, Disasters, and Bioterrorism, December 2011. http://healthyamericans.org/assets/files/TFAH2011ReadyorNot_09.pdf. 2011 December 30.

26.

Nuzzo

. Developing a national biosurveillance program. Biosecur Bioterror, 2009; 7,1:37–38.

27.

International Society for Disease Surveillance (ISDS) Meaningful Use Workgroup. Preliminary recommendation: Core EHR data requirements for syndromic surveillance. Brighton, MA: ISDS, September 2010. https://sites.google.com/site/biosenseredesign/file-cabinet. 2010 September 29.