Brief Report

ED syndromic surveillance for novel H1N1 spring 2009^?

Marc A. Bellazzini MD?, Kyle D. Minor MD

Division of Emergency Medicine, University of Wisconsin school of Medicine and Public Health, Madison, WI 53792, USA

Received 29 July 2009; revised 23 August 2009; accepted 11 September 2009

Abstract

Purpose: The aim of this study is to demonstrate the use of emergency department (ED) syndromic surveillance in the setting of a novel and unexpected H1N1 influenza outbreak.

Basic Procedures: Data collection from ED electronic medical records was used to track initial chief complaint and discharge International Classification of Diseases, Ninth Revision, codes related to influenza-like illness . An alert threshold was generated using cumulative sum sequential analysis technique. The data were retrospectively analyzed to identify alerts that correlated with novel influenza H1N1 illness.

Main Findings: Our system alerted for ILI earlier than both the official national Centers for Disease Control and Prevention (CDC) press release for novel H1N1 and the first laboratory confirmed case in our county. Principal Conclusions: Emergency department syndromic surveillance can be used to detect unexpected ILI before laboratory confirmation and serve as an adjunct to traditional laboratory-guided public health alerts. Early identification may allow for more efficient laboratory testing and early implementation of respiratory isolation precautions.

(C) 2011

Introduction

Syndromic surveillance is an adjunct to traditional public health surveillance that allows epidemiologists and infec- tious disease practitioners to identify Acute disease outbreaks by monitoring groupings of health care data such as Inter- national Classification of Diseases, Ninth Revision (ICD-9) or chief complaints. Examples of such groupings include influenza-like illness (ILI) and respiratory and gastrointes- tinal syndromes [1,2]. Syndromic surveillance also relies on many prediagnostic data sources such as over-the-counter pharmaceutical sales, school or work absenteeism, calls to health care hotlines, and visits to health care providers to identify potential disease outbreaks [3-5]. Historically,

^? Supported from a pilot project grant University of Wisconsin Department of Medicine.

* Corresponding author. Tel.: +1 608 263 1325; fax: +1 608 262 2641.

E-mail address: [email protected] (M.A. Bellazzini).

emergency department (ED) health care data have been shown to be useful for public health disease surveillance [2,6,7]. Emergency department surveillance systems can not only provide early warning to public health officials but also provide advanced warning of unusual trends in illness to ED personnel allowing more timely proactive steps in patient isolation and situational awareness.

Emergency departments are often the frontlines of health care for many who become seriously ill, are unable to be acutely seen by a primary care provider, or are uninsured. Thus, EDs see a wide range of acute illness and will undoubtedly be involved in the evaluation and treatment of those who are acutely ill during a local disease outbreak, epidemic, or pandemic. In addition to caring for a variety of disease, emergency physicians are under increasing pressure to limit laboratory testing to provide more cost-effective medicine.

Activity for the early 2008-2009 influenza season was typical with a rise in activity noted in early January. Activity

0735-6757/$ – see front matter (C) 2011 doi:10.1016/j.ajem.2009.09.009

The purpose of this article is to describe the performance of ED syndromic surveillance for ILI during the novel H1N1 influenza outbreak. The ED syndromic surveillance may serve as a more timely notification system alerting public chealth and medical professionals to novel and unexpected disease patterns. Thus, such systems may allow ED personnel to be more proactive in patient isolation and implement more efficient and appropriate testing.

Table 1 Any one of the listed Diagnostic codes would generate a syndrome count for the patient encounter

ILI ICD-9 syndrome definition

487.00 Influenza with pneumonia

487.10 Influenza w/oth resp ma

487.80 Influenza with other manifest

79.99 viral infections unspecified

79.89 Viral infection others

780.60 Fever

490.00 Bronchitis NOS

486.00 Pneumonia NOS

465.80 URI OTHER MUL SITES

465.90 URI ACUTE NOS

466.00 Bronchitis acute

786.20 Cough

460.00 Nasopharyngitis acute

485.00 Bronchopneumonia organism

461.90 Acute sinusitis unspecified

466.19 Bronchitis acute DT other organism

Methods

peaked from the sixth through the eighth week of 2009, with the predominant types being seasonal A (H1N1) and influenza B. During the 15th week of influenza season, the accounts of influenza continued to trend downward, and it appeared that the influenza season would soon be over.

On April 23, 2009, during the 16th week of influenza season, a CDC press release informed the public of an unusual outbreak of ILI that was identified in Mexico. Two additional cases were identified in the southwestern United States. This causative agent was later identified as a novel H1N1 virus [8-11]. The emergence and rapid spread of this novel virus would later prompt the World Health Organiza- tion to declare a new influenza pandemic.

Influenza pandemics have occurred 3 times in the past century; in 1918-1919, 1957-1958, and 1968-1969. Histor- ically, these pandemics have been preceded by a mild form of the virus in the spring with the more virulent virus disseminating in fall. This early phase of the current H1N1 pandemic may be a precursor to more serious and widespread activity of the influenza virus later this year.

Table 2 Any one of the listed complaints would generate a syndrome count for the patient encounter

ILI chief complaint syndrome definition Fever

Cough Headache Fatigue/Malaise Muscle pain URI symptoms Viral illness Body aches Sore throat Chills

This study was performed in an Urban academic ED with an annual volume of approximately 40 000 patients. Documenta- tion at our facility is done through EPIC ASAP (Epic Systems Corporation, Madison, WI). Patient data including chief com- plaint and diagnosis are entered at the time of initial presentation and discharge, respectively. All diagnoses are automatically linked through the electronic medical record (EMR) to ICD-9 codes resulting in real-time availability of such codes for syndromic surveillance. A standardized list of both chief complaints and diagnostic codes are embedded in the EMR.

A limited data set including patient encounter date, time, age, chief complaint, and ICD-9 discharge diagnosis are automatically and securely transferred to our surveillance server via the Secure File Transfer Protocol by 10:00 AM every morning. Unique Health Insurance Portability and Accountability Act patient identifiers such as medical record number or date of birth are not included in the data set. surveillance data are transferred in comma separated value (CSV) file format and reflects the previous 24 hours of ED patient encounters. The file is automatically imported into a MySQL open source database.

Data were retrospectively gathered from September 1, 2008, through May 27, 2009. The database was queried to count daily chief complaint and discharge diagnosis related to ILI. The definitions used for this study are listed in Tables 1 and 2. The ICD-9 codes were based on and modified from CDC syndrome definitions for ILI. Chief complaints were chosen from a standardized list embedded in the EMR to best reflect ILI. The sum of the chief complaints or ICD-9 codes were reported for each day. For example, 4 complaints of cough, 3 fever, and 3 muscle aches within a 24-hour time frame would result in an ILI syndrome count of 10 for that day. The same was performed for the ICD-9 data stream. This information was exported back into CSV file format.

Fig. 1 Example of calculation of alert threshold using a modified CUSUM technique.

Fig. 2 The ICD-9 data stream for 2008-2009 influenza season. This figure shows the daily counts, the CUSUM line (solid line) and the CUSUM + 2.5 standard deviation alerting threshold line (broken line). Exceeded alert threshold dates are depicted in red. First alert for H1N1 season occurred on April 18; the second alert occurred on April 29. The CDC press release for H1N1 cases in the United States occurred on April 23 (PR), and the first laboratory confirmed case occurred on May 6 (CD).

A modified sequential analysis technique cumulative sum (CUSUM) was used to calculate an alert threshold for the date of interest [12]. This alert threshold was compared to the date of interest’s actual counts. If the date of interest’s counts equaled or exceed the alert threshold, an alert was generated.

To determine the alert value, the previous 7 consecutive day’s average syndrome counts, displaced by 3 days, was used to calculate the threshold. For example, if the date of interest was October 15, then the average of the daily

counts from October 6 through 12 was used in the calculation. The 7-day data range was displaced by 3 days to decrease the dilutional effect of any gradual consecutive rise in counts near the day of interest. Common thresholds of 2.5 to 3 times the standard deviations above the mean are used for heightened and routine syndromic surveil- lance. Values closer to 3 reduce false-positive alerts and also decrease sensitivity of the surveillance system. Our alert threshold was defined as 2.5 standard seviations above the calculated mean (Fig. 1).

Fig. 3 Chief complaint data stream for 2008-2009 influenza season. This figure shows the daily counts of ILI chief complaint, the CUSUM line (solid line) and the CUSUM + 2.5 standard deviation alerting threshold line (broken line). Exceeded alert threshold dates are depicted in red. The only alert for H1N1 season occurred on April 29. The CDC press release for H1N1 cases in the United States occurred in April 23 (PR), and the first laboratory confirmed case occurred in May 6 (CD).

Fig. 4 ILI activity for typical influenza season 2007-2008.

A computer program was written in the PHP program- ming language (www.php.net) to read daily syndrome counts from the CSV file, calculate the alert threshold, and flag if the daily count met or exceeded the threshold. Graphs representing daily syndrome counts and alert dates were generated using the R statistical package version 2.8.0 (R Foundation for Statistical Computing).

Results

Alerts for ILI occurred in April 18 (daily count 17 / alert threshold 12) and in April 29 (daily count 18 / alert threshold 15) using the ICD-9 data stream. These alerts were associated with an overall increase in ICD-9 ILI- associated diagnosis over the previous 2 weeks. The April 18 alert occurred 5 days before the first official CDC press release on April 23, 2009, and 18 days before the first confirmed case of novel H1N1 in Dane county on May 6, 2009 (Fig. 2). An alert for ILI using the chief complaint data stream (daily count 24 / alert threshold 24) coincided with the April 29 ICD-9 signal and occurred 7 days before the first laboratory confirmed case if H1N1 in our county (Fig. 3). Data from the 2007-2008 influenza season are included as an example of disease activity in a typical flu season (Fig. 4).

Discussion

Syndromic surveillance using chief complaint and ICD-9 discharge diagnosis indicated a rise in ILI activity 1 to 2 weeks earlier than laboratory confirmation of cases of novel H1N1 in our county. This has important implications for ED

operation and public health in the detection of unexpected disease outbreaks.

Traditional surveillance methods rely heavily on labora- tory testing and physician reporting of disease activity. Emergency physicians often rely on clinical acumen to minimize testing and expense. In fact, the CDC recommends that testing be prioritized for those with severe respiratory illness and those at highest risk of complications from influenza [13]. This may lead to delayed and underreporting of new disease activity of an unexpected outbreak. Advanced warning generated by Syndromic surveillance systems for ILI may allow health care providers and public health officials to perform more efficient and early laboratory testing for respiratory illness.

Several Hospital systems have advanced integrated syndromic surveillance that use real-time ED or telephone triage chief complaints to alert public health officials to potential disease outbreaks. There has been retrospective demonstration of the early warning benefit of using ED discharge diagnoses compared to traditional laboratory testing surveillance for influenza [14]. With the implemen- tation of statistical methods for space-time cluster detection, such as CUSUM analysis, practitioners and public health officials can prospectively analyze ED data generated in real time [15]. To our knowledge, this article is the first to describe how a syndromic surveillance system can work in an unexpected influenza outbreak.

The CUSUM method was effective in identifying statistically significant spikes in ILI activity; however, the alarms were considered to be true positives only when associated with an increasing trend. There were several false positives throughout the year that were not correlated with a confirmed influenza outbreak. Thus, human analysis of the data associated with an alert is necessary to determine the likelihood of a true outbreak.

Although the CUSUM method was effective in identify- ing aberrant spikes in ILI activity, attention to graphical trends is also central to effective monitoring for disease. A large increase in ILI activity was identified on April 18, 2009, within the ICD-9 data stream. Fortunately, this abrupt increase in counts occurred during the early increase in slope of the graph. This was not the case within the chief complaint data stream where an abrupt rise in activity occurred near the apex of activity. Early identification of aberrant ILI activity could also have been detected by daily visual inspection of the graphs noting a gradual but significant rise in counts.

Increased ED visits for respiratory illness may be attributed to the widespread press coverage of the novel H1N1 virus. However, our system demonstrated an upward trend and also alerted 5 days before the CDC press release. Thus, this makes a false-positive trend of worried well unlikely.

We postulate that the addition of syndromic surveil- lance to traditional surveillance methods can enhance disease outbreak detection. Advanced warning generated by such systems for ILI may allow health care providers to perform more efficient and early laboratory testing for disease confirmation or identification of new strains of Respiratory pathogens.

Early detection may also increase vigilance within the ED by prompting health care providers to place patients in respiratory isolation and increase use of Personal protective equipment. This may decrease transmission of both common respiratory viruses and potentially more serious pathogens both among patients and health care workers. Appropriate safeguards may decrease provider illness and decrease the likelihood of staff shortages during an outbreak. If such systems are linked to public health, early identification of a rise in ILI activity may prompt more aggressive and appropriate investigation of disease outbreaks.

References

1. Cooper DL, Verlander NQ, Smith GE, Charlett A, Gerard E, Willocks L, et al. Can syndromic surveillance data detect local outbreaks of

communicable disease? A model using a historical cryptosporidiosis outbreak. Epidemiol Infect 2006;134(1):13-20.

1. Lemay R, Mawudeku A, Shi Y, Ruben M, Achonu C. Syndromic surveillance for influenzalike illness. Biosecur Bioterror 2008;6(2): 161-70.
2. Marx MA, Rodriguez CV, Greenko J, Das D, Heffernan R, Karpati AM, et al. Diarrheal illness detected through syndromic surveillance after a massive power outage: New York City, August 2003. Am J Public Health 2006;96(3):547-53.
3. Das D, Metzger K, Heffernan R, Balter S, Weiss D, Mostashari F. Monitoring over-the-counter medication sales for early detection of disease outbreaks–New York City. MMWR Morb Mortal Wkly Rep 2005;54(Suppl):41-6.
4. Besculides M, Heffernan R, Mostashari F, Weiss D. Evaluation of school absenteeism data for early outbreak detection, New York City. BMC Public Health 2005;1:5105.
5. Fleischauer AT, Silk BJ, Schumacher M, Komatsu K, Santana S, Vaz V, et al. The validity of chief complaint and discharge diagnosis in emergency department-based syndromic surveillance. Acad Emerg Med 2004;11(12):1262-7.
6. Townes JM, Kohn MA, Southwick KL, Bangs CA, Zechnich AD, Magnuson JA. et al. Investigation of an electronic emergency department information system as a data source for respiratory syndrome surveillance. J Public Health Manag Pract 10(4):299-307
7. Baden LR, Drazen JM, Kritek PA, Curfman GD, Morrissey S, Campion EW. H1N1 influenza A disease–information for health professionals. N Engl J Med 2009;360(25):2666-7.
8. Dawood FS, Jain S, Finelli L, Shaw MW, Lindstrom S, Garten RJ, et al. Emergence of a novel swine-origin influenza A (H1N1) virus in humans. N Engl J Med 2009;360(25):2605-15.
9. Miller MA, Viboud C, Balinska M, Simonsen L. The signature features of influenza pandemics–implications for policy. N Engl J Med 2009;360(25):2595-8.
10. Update: novel Influenza A virus infections–worldwide, May 6, 2009. MMWR Morb Mortal Wkly Rep 2009;58(17):453-8.
11. Hutwagner LC, Thompson WW, Seeman GM, Treadwell T. A Simulation model for assessing aberration detection methods used in public health surveillance for systems with limited baselines. Stat Med 2005;24(4):543-50.
12. CDC H1N1 flu | interim guidance for clinicians on identifying and caring for patients with swine-origin influenza A (H1N1) virus infection. [Internet]. [cited 2009 Jul 24]. Available from http://www. cdc.gov/h1n1flu/identifyingpatients.htm.
13. Zheng W, Aitken R, Muscatello DJ, Churches T. Potential for early warning of viral influenza activity in the community by monitoring clinical diagnoses of influenza in hospital emergency departments. BMC Public Health 2007;1:7250.
14. Sonesson C. A CUSUM framework for detection of space-time disease clusters using scan statistics. Stat Med 2007;26(26):4770-89.