Reevaluation of Epidemiological Criteria for Identifying Outbreaks of Acute Gastroenteritis Due to Norovirus: United States,

Transcription

1 MAJOR ARTICLE Reevaluation of Epidemiological Criteria for Identifying Outbreaks of Acute Gastroenteritis Due to Norovirus: United States, Reina M. Turcios, 1 Marc-Alain Widdowson, 1 Alana C. Sulka, 2 Paul S. Mead, 2 and Roger I. Glass 1 1 Respiratory and Enteric Viruses Branch, Division of Viral and Rickettsial Diseases, and 2 Foodborne and Diarrheal Diseases Branch, Division of Bacterial and Mycotic Diseases, National Center for Infectious Diseases, Centers for Disease Control and Prevention, Atlanta, Georgia (See the editorial commentary by Lopman on pages 970 1) Background. Noroviruses are believed to be the most common etiologic agent of foodborne outbreaks of gastroenteritis, yet diagnostic tests for these agents are not readily available in the United States. In the absence of assays to detect norovirus, several clinical and epidemiologic profiles the criteria of Kaplan et al. (vomiting in 150% of patients, mean incubation period of h, mean duration of illness of h, and no bacterial pathogen) and the ratios of fever to vomiting and diarrhea to vomiting have been used to distinguish foodborne outbreaks of gastroenteritis caused by noroviruses from those caused by bacteria. Methods. To examine how well clinical and epidemiological profiles discriminate between foodborne outbreaks of gastroenteritis due to noroviruses and those due to bacteria and to estimate the proportion of reported outbreaks that might be attributable to noroviruses, we reviewed subsets of the 4050 outbreaks reported from 1998 to Results. The set of criteria of Kaplan et al. is highly specific (99%) and moderately sensitive (68%) in discriminating confirmed outbreaks due to bacteria from those due to norovirus and was the most useful diagnostic aid evaluated. Each individual component of the criteria, the fever-to-vomiting ratio, and the diarrhea-to-vomiting ratio were more sensitive, yet less specific, and therefore less useful, than the criteria of Kaplan et al. We estimated that, at a minimum, 28% of all the foodborne outbreaks reported to the Centers for Disease Control and Prevention may be attributed to norovirus on the basis of these criteria. Conclusion. Until norovirus diagnostic tests become widely available, the criteria of Kaplan et al. remain the most useful and discriminating diagnostic aid to identify foodborne outbreaks of gastroenteritis due to noroviruses. Diarrhea is recognized as one of the most common ailments in humans, causing some 195 million episodes of illness in the United States each year [1], with at least 13.6 million episodes and an average of 550 outbreaks annually associated with foodborne transmission [2, 3]. Despite this large burden of disease, the etiology of most episodes of illness and outbreaks remains unknown [4]. Recent advances in molecular diagnostic testing have not percolated into practice in clinical medicine or public health in the United States. Consequently, clinicians and public health practitioners Received 22 September 2005; accepted 22 November 2005; electronically published 27 February Reprints or correspondence: Dr. Reina M. Turcios, Centers for Disease Control and Prevention, 1600 Clifton Road, Mailstop A-34, Atlanta, GA (RTurcios@cdc.gov). Clinical Infectious Diseases 2006; 42:964 9 This article is in the public domain, and no copyright is claimed /2006/ have limited capacities for determining the etiology of these foodborne outbreaks and thus for implementing more-specific prevention and control measures. This is particularly problematic for noroviruses, formerly called Norwalk-like viruses or small round-structured viruses, which are recognized to be the most common etiologic agents of foodborne outbreaks of diarrhea [5, 6]. New molecular diagnostic tests for noroviruses are being introduced slowly into state public health laboratories but are not available to physicians in clinic or hospital settings [7]. In the absence of such diagnostic tests, it is impossible to fully assess the true burden of norovirus disease, and physicians remain ill-informed as to the etiology of this most common illness. In 1982, Kaplan et al. [8] recognized that the lack of accessible diagnostic tests was a problem in establishing the etiology of outbreaks of gastroenteritis that might be caused by norovirus. They developed a set of criteria to distinguish outbreaks due to Norwalk virus, 964 CID 2006:42 (1 April) Turcios et al.

2 the prototype strain of human noroviruses, from outbreaks of bacterial etiology. The 4 criteria indicative of a norovirus diagnosis are as follows: vomiting in 150% of affected persons in the outbreak; a mean (or median) incubation period of h; a mean (or median) duration of illness of h; and lack of identification of a bacterial pathogen in culture of stool (table 1). Other clinical criteria, such as the fever-to-vomiting ratio to differentiate norovirus outbreaks from bacterial outbreaks and the diarrhea-to-vomiting ratio to differentiate outbreaks due to enterotoxigenic Escherichia coli from those due to norovirus, have been proposed [6, 9]. Of note, Kaplan et al. [8] developed their criteria when diagnostic tests for outbreaks were based on the insensitive tests of electron microscopy and seroconversion to Norwalk virus infection. Since that time, not only have other noroviruses been discovered and categorized into 5 genogroups, but the Norwalk virus has been recognized the prototype of genogroup 1, which is associated with 26% of human norovirus-associated illness [10]. We reexamined how well the criteria of Kaplan et al. [8] (hereafter, referred to as the Kaplan criteria), the fever-to-vomiting ratio, the diarrhea-to-vomiting ratio, and each component of the Kaplan criteria discriminated between outbreaks due to norovirus and those due to bacterial agents. We also estimated the proportion of all outbreaks reported to the Centers for Disease Control and Prevention (CDC) between 1998 and 2000 that could be attributed to norovirus by using the Kaplan criteria alone. METHODS Since 1973, the CDC has maintained the Foodborne Outbreak Reporting System, through which data concerning the clinical characteristics of illness, the size and setting of the outbreak, and the etiologic agent are compiled in a computerized database. In this system, an outbreak is attributed to an etiologic agent if evidence of that agent is detected in stools from 2 affected persons through widely accepted assays listed elsewhere [3]. Using SAS software, version 8.01 (SAS Institute), we segregated reports of outbreaks occurring between 1998 and 2000 by their suspected or confirmed etiologic agents and selected outbreaks with complete information to test the fit of the Kaplan criteria that is, the mean incubation period, the mean duration of illness, and the proportion of vomiting among patients. Our goal was to evaluate how well the clinical Kaplan criteria, the fever-to-vomiting ratio, and the diarrhea-to-vomiting ratio discriminated between outbreaks due to norovirus and those due to bacteria. Thus, from this selection with complete data, we identified only those outbreaks with a confirmed norovirus or bacterial etiology by clinical microbiological testing and molecular biological techniques, respectively, and determined how many of the outbreaks due to norovirus and how many of the Table 1. The Kaplan criteria for the identification of outbreaks of gastroenteritis due to norovirus. Criteria Vomiting in more than half of affected persons Mean (or median) incubation period of h Mean (or median duration of illness of h No bacterial pathogen in stool culture outbreaks due to bacteria did and did not fulfill the Kaplan criteria. In similar fashion, we evaluated each component of the Kaplan criteria, the fever-to-vomiting ratio, and the diarrhea-to-vomiting ratio. To confirm that the cutoff value of vomiting occurring in 50% of affected persons was optimal in differentiating outbreaks due to norovirus from those due to bacteria, we developed receiver operating characteristic (ROC) curves for different values [11] and calculated the areas under ROC curves with SAS software, version 8.01 (SAS Institute). Using EpiInfo 6, version 6.04d (CDC), we calculated for each index the sensitivity, specificity, and positive and negative predictive values with their respective 95% CIs. To further evaluate the usefulness of each index, we also derived the likelihood ratios, which reflect how well a diagnostic test correctly discriminates the condition it is meant to identify, given a positive result. The likelihood ratio is calculated as follows: sensitivity divided by the result of 1 minus the specificity. This value is independent of the prevalence of the condition in the population tested. Thus, a test with a high likelihood ratio is more useful than one with a low value. We repeated this exercise with a small group of outbreaks in which parasites or toxins were specifically identified to test the Kaplan criteria among outbreaks due to nonbacterial causes. To assess differences between outbreaks due to norovirus that fulfilled the Kaplan criteria and those not meeting the criteria, we compared the age distribution and the total number of affected persons in these outbreaks and performed the Wilcoxon rank-sum test to identify significant differences ( P!.05). Last, we estimated the total number of outbreaks reported from 1998 to 2000 that might be attributed to norovirus by applying the results of our reevaluation of the Kaplan et al. criteria to the remaining outbreaks. We summed the number of outbreaks confirmed to be caused by norovirus and the number of outbreaks with sufficient clinical data of unknown or suspected, but not confirmed, etiology that fulfilled the criteria. We then corrected our estimates for norovirus outbreaks by adding those that would not have been identified by the criteria (i.e., false-negative results) and subtracted bacterial outbreaks that met the criteria (i.e., false-positive results). Finally, we added an equivalent proportion of the number of outbreaks of unknown etiology with incomplete clinical data, excluding outbreaks with a suspected bacteria or norovirus etiology but Epidemiological Criteria for Norovirus CID 2006:42 (1 April) 965

3 with incomplete clinical data, and those suspected or confirmed to be caused by other agents (i.e., parasites, toxins, chemicals, or other viruses). RESULTS We analyzed data from 4050 foodborne outbreaks of gastroenteritis reported with the Foodborne Outbreak Reporting System between 1998 and 2000 (figure 1). Noroviruses were suspected or confirmed in 549 outbreaks (14%), bacteria were suspected or confirmed in 1084 (27%), and no etiologic agent was suspected or confirmed in 2108 (52%); the remaining 309 outbreaks (8%) were confirmed or suspected to be caused by other agents, including viruses other than noroviruses (118 outbreaks; 38 confirmed, 80 suspected), toxins or chemicals (176 outbreaks; 110 confirmed, 66 suspected), parasites (13 outbreaks, all confirmed), and other causes (2 outbreaks, both suspected). For our reassessment of the Kaplan criteria, we used only those outbreaks of confirmed bacterial (214) or norovirus (148) etiology for which complete clinical data were available. We tested the ability of the Kaplan criteria, its components (i.e., vomiting, duration of illness, and incubation period), and fever-to-vomiting and diarrhea-to-vomiting ratios to discriminate outbreaks of gastroenteritis due to noroviruses from those due to bacteria (table 2). We calculated sensitivity, specificity, positive and negative predictive values, and likelihood ratios, which we used as an intrinsic indicator of the usefulness of each index. The Kaplan criteria was both the most specific index evaluated (99%) and the most useful for identifying outbreaks caused by norovirus (likelihood ratio, 48.7; positive predictive value, 97.1%). Norovirus outbreaks that did not fulfill the criteria did not differ from those that did either in size of the outbreak or age distribution of those affected. The few bacterial outbreaks that fulfilled the clinical criteria (3 outbreaks) were all caused by a Salmonella species. No outbreak with sufficient clinical data and confirmed to be caused by toxins (21 outbreaks) or by parasites (6 outbreaks) met the criteria. Each individual component of the criteria was more sensitive (86% 89%) than the set of criteria, but each was also less specific (61% 70%), which resulted in a much lower likelihood ratio ( ), a lower positive predictive value (61% 67%), and a less discriminating index. The largest area under the ROC curve calculated corresponded to the curve using the vomiting in 50% of affected persons cutoff value (50%: area under ROC curve, 0.770; 45%: area under ROC curve, 0.750; 55%: area under ROC curve, 0.749; 60%: area under ROC curve, 0.726) to differentiate outbreaks due to norovirus from those due to bacteria. The diarrhea-to-vomiting ratio was the most sensitive test evaluated (97%), but it failed to correctly exclude more than half of the outbreaks of bacterial origin, although no outbreaks due to enterotoxigenic E. coli were included in the data set. The fever-to-vomiting ratio had similar characteristics. We also estimated the total number of the 4050 outbreaks reported that might have been attributable to noroviruses. We applied the Kaplan criteria to the 1141 outbreaks with clinical data available but with unknown etiology (782) or suspected but unconfirmed etiology (norovirus, 160; bacteria, 199) (figure 1). Of these 1141 outbreaks, 289 fulfilled the criteria. Correcting for the limitations of the Kaplan criteria, we included an ad- Figure 1. Characteristics of 4050 foodborne outbreaks of gastroenteritis reported through the Centers for Disease Control and Prevention s Foodborne Outbreak Reporting System between 1998 and 2000 by suspected or confirmed etiology and completeness of clinical data. Other causes of outbreaks include parasites, toxins, chemicals, and other viruses. 966 CID 2006:42 (1 April) Turcios et al.

4 Table 2. Characteristics of indices used to discriminate between outbreaks of gastroenteritis due to norovirus and outbreaks of gastroenteritis due to bacteria reported through the Centers for Disease Control and Prevention s Foodborne Outbreak Reporting System, Index characteristics No. of outbreaks with confirmed etiology Norovirus Bacteria Sensitivity, % (95% CI) Specificity, % (95% CI) Likelihood Positive ratio a predictive value, % (95% CI) Negative predictive value, % (95% CI) Kaplan et al. b 68.2 ( ) 98.6 ( ) ( ) 81.8 ( ) No. of outbreaks that fit the criteria No. of outbreaks that did not fit the criteria % of patients with vomiting 88.5 ( ) 60.7 ( ) ( ) 88.4 ( ) 50% !50% Duration of illness 85.8 ( ) 65.0 ( ) ( ) 86.9 ( ) h Not h Incubation period 89.2 ( ) 69.6 ( ) ( ) 90.3 ( ) h Not h Fever-to-vomiting ratio c 90.1 ( ) 46.6 ( ) ( ) 86.3 ( ) Diarrhea-to-vomiting ratio d 96.6 ( ) 44.5 ( ) ( ) 94.9 ( )! The likelihood ratio was calculated as the sensitivity divided by the result of 1 minus the specificity, and it is an indicator of the inherent usefulness of a test, regardless of illness prevalence. b Excludes the lack of identification of a bacterial pathogen in stool culture. Information regarding fever is unknown for 7 outbreaks due to norovirus and 25 outbreaks due to bacteria. d Information regarding diarrhea is unknown for 5 outbreaks due to bacteria. a c

5 ditional 92 outbreaks representing outbreaks due to noroviruses that were not identified by the criteria (false-negative results, 1 sensitivity) and excluded 4 outbreaks of bacterial etiology that may have been incorrectly attributed to noroviruses (falsepositive results, 1 specificity). This yielded a total of 377 (33%) of the 1141 outbreaks of unknown or unconfirmed etiology attributable to norovirus. We extrapolated this proportion to the remaining 1326 outbreaks of unknown etiology with incomplete data, which resulted in an additional 438 outbreaks due to norovirus. Excluded from this analysis were 82 outbreaks suspected to caused by norovirus but with incomplete clinical data, 181 suspected to be caused by bacteria also with incomplete clinical data, and 309 suspected or confirmed to be cause by other agents. We totaled the outbreaks of known norovirus etiology (307), the corrected number of outbreaks of unknown or unconfirmed etiology attributable to norovirus (377), and the extrapolation to outbreaks of unknown etiology with incomplete data (438). In conclusion, 1122 (28%) of the 4050 outbreaks reported could be attributed to norovirus. DISCUSSION Although it has been 30 years since norovirus became the first virus associated with acute gastroenteritis, we still lack simple and sensitive assays to detect this virus in routine clinical specimens. EIAs with limited sensitivity, because of the genetic diversity of the virus, are available in other countries and are under refinement [12, 13] but are currently neither licensed nor available in the United States. Highly sensitive and specific PCR-based assays, considered the reference standard for the diagnosis of norovirus, are available at the CDC and some public health laboratories, but hospitals and clinics do not have these assays [14]. Until improved diagnostic tests become available, the Kaplan criteria will continue to play a role in the clinical assessment of outbreaks of acute gastroenteritis. They remain a most useful clinical tool for discriminating between outbreaks due to norovirus and those due to bacteria. They have also proven useful in distinguishing outbreaks due to norovirus from those due to toxins and parasites. Application of the criteria to this large data set of outbreaks suggests that foodborne outbreaks attributable to norovirus occur even more often than initially suspected (28% vs. 14%). However, our estimate of 28% is lower than the 41% of foodborne outbreaks attributable to norovirus reported in Minnesota [5] and the up-to-50% estimate from a study in 6 US states [15]. Both of these figures are from states that routinely test for norovirus and report relatively high numbers of outbreaks of gastroenteritis of unknown etiology. Conversely, the latter study found that states that do not test for norovirus also report proportionally fewer outbreaks of unknown etiology and may be biased against reporting outbreaks of virus-like illness [15]. The fact that our study includes states that do not test for norovirus may explain why our estimate is lower than those previously published. Each component of the criteria on its own was more sensitive than the set of criteria, but each was less specific. Thus, in balance, each component was less useful than the set combined. The same was the case for the fever-to-vomiting and the diarrhea-to-vomiting ratios. Our findings confirm the original estimation by Kaplan et al. [8] of a sensitivity of 77% and high specificity for their criteria. Of note is that those authors initial evaluation used a subset of the original data from which the criteria were derived and that data was based on outbreaks due to Norwalk virus alone [16]. This report extends the observation to the more-common genogroup 2 strains. Our evaluation is limited by the nature of the data used. The outbreaks of foodborne illness reported to the CDC may not accurately reflect outbreaks occurring in the community and may be biased toward outbreaks for which an etiology has been specifically sought [4]. Hence, outbreaks without an identified agent are likely to be underreported, and many of these might be found to be caused by norovirus if an etiologic agent were sought. Moreover, the ultimate etiology of outbreaks with unknown or unconfirmed etiology included in the data set analyzed may differ from that of outbreaks with confirmed etiology, because the first probably include a larger proportion of disease due to hard-to-diagnose etiologies such as norovirus. The resulting extrapolations would be conservative estimates of the actual number of norovirus-attributable events. Last, the Kaplan criteria do not permit linking of related outbreaks. Thus, within the group of outbreaks attributed to norovirus, a number of small outbreaks could be present that actually represent a larger common event that might have been recognized had stool samples been collected and submitted for testing for norovirus, sequencing, and comparison of sequences. Our findings are of particular interest to public health practitioners and individual physicians alike. For the former, the Kaplan criteria can help discriminate between foodborne outbreaks due to norovirus and those due to other etiologies and increase awareness of the value of collecting fecal specimens to confirm the diagnosis of an outbreak, possibly linking it to other disease clusters. Results suggest caution in using individual components of the Kaplan criteria, because each is much less discriminating than the full set. In the practice of public health, it would be prudent to obtain stool samples in all outbreaks and to submit samples from outbreaks without a readily identified etiology for norovirus testing. For physicians, our findings highlight the important role that noroviruses play among agents that cause acute gastroenteritis. For both public health practitioners and physicians, they underscore the need for simpler and more accessible diagnostic tests for norovirus in general. 968 CID 2006:42 (1 April) Turcios et al.

6 Acknowledgments We thank Aaron T. Curns for statistical assistance and Claudia Chesley for editorial assistance. Potential conflicts of interest. All authors: no conflicts. References 1. Imhoff B, Morse D, Shiferaw B, et al. Burden of self-reported acute diarrheal illness in FoodNet surveillance areas, Clin Infect Dis 2004; 38:S Mead PS, Slutsker L, Dietz V, et al. Food-related illness and death in the United States. Emerg Infect Dis 1999; 5: Olsen S, MacKinnon L, Goulding J, Bean N, Slutsker L. Surveillance for foodborne-disease outbreaks United States, MMWR CDC Surveill Summ 2000; 49: Jones TF, Imhoff B, Samuel M, et al. Limitations to successful investigation and reporting of foodborne outbreaks: an analysis of foodborne disease outbreaks in FoodNet catchment areas, Clin Infect Dis 2004; 38:S Deneen VC, Hunt JM, Paule CR, et al. The impact of foodborne calicivirus disease: the Minnesota experience. J Infect Dis 2000; 181: S Hedberg CW, Osterholm MT. Outbreaks of food-borne and waterborne viral gastroenteritis. Clin Microbiol Rev 1993; 6: Glass RI, Noel J, Ando T, et al. The epidemiology of enteric caliciviruses from humans: a reassessment using new diagnostics. J Infect Dis 2000; 181:S Kaplan JE, Gary GW, Baron RC, et al. Epidemiology of Norwalk gastroenteritis and the role of Norwalk virus in outbreaks of acute nonbacterial gastroenteritis. Ann Intern Med 1982; 96: Dalton CB, Mintz ED, Wells JG, Bopp CA, Tauxe RV. Outbreaks of enterotoxigenic Escherichia coli infection in American adults: a clinical and epidemiologic profile. Epidemiol Infect 1999; 123: Fankhauser RL, Monroe SS, Noel JS, et al. Epidemiologic and molecular trends of Norwalk-like viruses associated with outbreaks of gastroenteritis in the United States. J Infect Dis 2002; 186: Kirkwood BR, Sterne JAC. Measurement error: assessment and implications. In: Essential medical statistics. 2nd ed. Malden, MA: Blackwell Science, 2003: Burton-MacLeod JA, Kane EM, Beard RS, Hadley LA, Glass RI, Ando T. Evaluation and comparison of two commercial enzyme-linked immunosorbent assay kits for detection of antigenically diverse human noroviruses in stool samples. J Clin Microbiol 2004; 42: Jiang X, Wilton N, Zhong WM, et al. Diagnosis of human caliciviruses by use of enzyme immunoassays. J Infect Dis 2000; 181:S Parashar U, Quiroz ES, Mounts AW, et al. Norwalk-like viruses : public health consequences and outbreak management. MMWR Recomm Rep 2001; 50: Widdowson MA, Sulka A, Bulens SN, et al. Norovirus and foodborne disease, United States, Emerg Infect Dis 2005; 11: Kaplan JE, Feldman R, Campbell DS, Lookabaugh C, Gary GW. The frequency of a Norwalk-like pattern of illness in outbreaks of acute gastroenteritis. Am J Public Health 1982; 72: Epidemiological Criteria for Norovirus CID 2006:42 (1 April) 969