Article Screening

Our literature search identified 2435 non-duplicate articles [Figure 1]. Two reviewers independently assessed each article for inclusion. The inclusion criteria required that articles 1) be published in an article format (e.g. citations of conference abstracts were excluded), 2) be published entirely in English, 3) describe human NoV outbreaks, and 4) have at least one case from each NoV outbreak confirmed by RT-PCR. Furthermore, the articles were required to have explicitly stated the number of primary cases (the article was excluded if the number of NoV-positive stools was reported instead of the number of primary cases). The exclusion criteria eliminated articles that documented sporadic cases of NoV or were published prior to the development of RT-PCR methodology for NoV detection in 1992. Discrepancies between each pair of reviews were resolved by consensus or a third investigator. Rather than impose a uniform case definition on the articles, we adopted each author’s case definition. Relevant articles were allowed to contain information on more than one outbreak and data was abstracted separately for each outbreak. However, outbreaks reported in multiple publications were only recorded once.

Open in a separate window

Figure 1

Flow diagram of searches, screening, and selection of eligible articles for creation of NoV outbreak database. The 250 articles meeting inclusion criteria contained 900 outbreaks published between December 1993 and May 2011. Outbreaks spanned from December 1983 to March 2010. Arrows indicate the temporal order of the steps represented in the gray rectangles.

NoV Outbreak Database

A standardized Epi Info™ survey (version 3.5.1) was used to abstract data on as many as 74 variables (when available) into a Microsoft Access database. These variables included, but were not limited to: mode of transmission (described below), vehicle (person or object infected or contaminated with NoV that promotes transmission of virus to healthy individuals or contamination of additional objects or surfaces with virus), genetic classification (genogroup and genotype/cluster), outbreak beginning and end dates, location (described below), outbreak setting (described below), incubation period (length of time between viral infection and symptom presentation), illness duration, case ages, number of primary cases and persons at risk (described below), primary attack rates, and description of any action taken to stop the outbreak.

During data abstraction, calculations were limited to addition or subtraction (e.g. adding staff and guest cases to generate a total primary case number for a hotel outbreak). Reported numbers of laboratory confirmed and clinically identified primary cases were also combined. Data were cleaned using SAS™ version 9.2 (SAS Institute, Cary, NC, USA). Outbreak seasonality was derived from the month of occurrence and the hemisphere of the affected country. Modes of transmission were categorized as follows: foodborne (e.g. consumption of contaminated shellfish or foods contaminated during production or preparation by an infected foodhandler), waterborne (e.g. drinking or recreational water), person-to-person (direct contact with infected vomitus or other aerosolized virus particles), and environmental (direct contact with contaminated surfaces). Outbreak settings were categorized as follows: foodservice (e.g. restaurants), healthcare (e.g. hospitals, long-term care facilities, psychiatric institutions), leisure (e.g. vacations, cruises, hotels, recreational activities), school/daycare (e.g. schools, daycare facilities, universities), and other (e.g. residential, community-wide outbreaks, military, and prison settings). Outbreaks were classified as GI or GII if the associated strains were identified exclusively as belonging to one genogroup or the other; outbreaks associated with GI and GII strains combined were classified accordingly, as long as one strain from the outbreak was identified as belonging to GI and another was identified as belonging to GII. To avoid counting an outbreak more than once, an outbreak associated with strains from multiple genogroups was not classified as belonging to GI individually or GII individually. To include the maximum number of outbreaks in the analysis, an algorithm was constructed in SAS to calculate number of persons at risk for direct infection with NoV (primary cases ÷ primary attack rate) and primary attack rate (primary cases ÷ persons at risk) for outbreaks from which those variables were not abstracted.

Transmission Route and Setting

Transmission route data was available for 565 outbreaks; 89 (16%) involved two or more modes of transmission. Multiple transmission outbreaks were replicated and segregated into their individual component modes of transmission, resulting in 666 observations for the transmission route sub-analysis [Table 1]. Setting data was abstracted from 830 outbreaks [Table 1]. Reported outbreaks were most commonly attributed to foodborne transmission (362/666, 54%) and foodservice settings (294/830, 35%), whereas the fewest number of outbreaks was associated with environmental transmission (60/666, 9%) or reported in school or daycare settings (80/830, 10%). As described in the methods, settings were consolidated to enhance statistical power, but within each setting, there were subcategories that comprised more than 15% of the reported setting. “School/Daycare” was comprised of school (64/80, 80%) and daycare (16/80, 20%) outbreaks. Similarly “Healthcare” was comprised of hospitals (107/225, 48%) and long-term care facilities (112/225, 50%). “Other” included private homes (30/91, 33%), community-wide outbreaks (20/91, 22%) and military settings (15/91, 16%). In general, outbreaks due to foodborne transmission or foodservice setting were significantly associated with the lowest number of primary cases and persons at risk, but were significantly associated with the highest attack rate when compared to other modes of transmission or outbreak settings. In addition, waterborne transmission attack rates were significantly higher than those associated with person-to-person or environmental transmission. Healthcare-related outbreaks were significantly associated with fewer primary cases and had the lowest attack rates.

We assessed 580 outbreaks that recorded information on primary attack rate, ranging from less than 1% to 100%, for significant bivariate relationships. Food- and waterborne outbreaks, as well as outbreaks in foodservice settings and outbreaks associated with both GI and GII strains, had significantly higher attack rates, when compared to their referent categories [Table 2]. By contrast, healthcare-related and GII cluster four (GII.4) NoV outbreaks, as well as outbreaks in which authors described a corrective action, had significantly lower attack rates.

In the multivariate linear regression model where multiple transmission outbreaks were excluded [Table 2], or where they were included in a model as an extra dummy variable (data not shown), no covariates were significantly associated with a higher or lower attack rate. However, winter season was borderline significant in the model where multiple transmission routes were classified as an extra dummy variable (p = 0.056, data not shown). In the model that included multiple transmission routes, where we classified outbreaks of more than one mode of transmission as multiple, independent outbreaks of a single transmission type (see Methods), outbreaks in foodservice settings and in the winter were associated with higher attack rates, when compared to their referent categories.

To assess whether additional covariates were significant if the power (sample size) of the models was increased, we included outbreaks that lacked data for the “setting” variable and found that, in either adjusted model, foodborne outbreaks were associated with higher attack rates, when compared to their referent categories (data not shown). In the model where outbreaks with multiple modes of transmission were included, winter outbreaks were also associated with higher attack rates (data not shown). When we included outbreaks that lacked data for the “mode of transmission” variable in either adjusted model, outbreaks in foodservice settings had significantly higher attack rates (data not shown). In addition, the “outbreak year” variable had a significant negative relationship with attack rate—outbreaks that occurred more recently were associated with lower attack rates (data not shown). Furthermore, in the model where outbreaks with multiple modes of transmission were included, spring outbreaks were associated with higher attack rates (data not shown).

Limitations and Strengths

There were three main limitations to this study. First, as is the case with many systematic reviews, our study was influenced by publication bias. We extracted data of several reports from active surveillance that were published. Publications were more likely from outbreaks with novel findings than small, relatively common outbreaks (e.g. healthcare-related settings). However, most infectious disease outbreaks are reported by passive surveillance. Passive surveillance has its own biases. Outbreak data collected may not be as comprehensive in passive versus active surveillance. Some outbreaks may not be recognized because illness is mild or because cases disperse (e.g. restaurant). In contrast, outbreaks may be recognized when cases are in communication with each other (e.g. wedding). Publication bias also resulted from the overrepresentation of outbreak data from five countries (Japan, United States, Sweden, United Kingdom, Netherlands) that accounted for approximately two-thirds of the reported outbreaks. Care should be taken when generalizing these results to countries with limited or extensive surveillance. For example, some countries may only be able to investigate outbreaks in one setting (e.g. foodservice or healthcare-related) or may have better surveillance of food- and waterborne outbreaks because it is easier to implicate a contaminated food item or water source than an environmental or person-to-person source.

Second, because we treated outbreaks with more than one mode of transmission as multiple, independent outbreaks of a single transmission type, we increased the effective sample size in these models and therefore some non-significant variables may have become significant. These findings may still be biologically valid but additional analyses of data from outbreaks with multiple transmission routes are needed to validate these findings. Third, as our data came from different authors, all outbreaks did not have the same variables. This limited the number of analysed outbreaks and may have increased the likelihood of misclassification bias (e.g. an outbreak that reported only genogroup-level information would not have been analysed as a GII.4 outbreak).

There were three key strengths to this analysis. First, we employed a rigorous study protocol that required validation by at least two individuals following article selection and data abstraction. Second, this analysis incorporates a greater number of countries than any other review of published NoV outbreak studies [7, 9, 27]. Third, the size of our database and multivariate modeling allowed us to control for a variety of variables. Furthermore, publication bias likely minimally influenced the multivariate analysis identifying factors associated with either GI or GII outbreaks [Table 4] because an outbreak was neither any more likely nor any less likely to be published simply because it was associated with a GI or a GII strain.

The authors acknowledge assistance with our systematic review from Emory University librarians Barbara Abu-Zeid and Mia White [to J.E.M., J.J.R. and J.R.F.] and statistical assistance from Emory University professor Dr. David Kleinbaum [to J.E.M, B.W.D. and J.S.L.].

We would also like to thank Elisabeth Honorat, MPH, whose master’s thesis “Epidemiology of norovirus outbreaks,” (Emory University, 2007) inspired the current work.

DECLARATION OF INTEREST

This work was partially supported by STAR grant RD83172701 from the U.S. Environmental Protection Agency [to C.L.M. and J.N.S.E.], grant 1K01AI087724 – 01 from the NIAID at the NIH, grant 2010-85212-20608 from the National Institute of Food and Agriculture at the U.S. Department of Agriculture, and the Emory University Global Health Institute [to J.S.L.]. We also acknowledge financial support from the Practical Experience Program at the Emory University Rollins School of Public Health (to J.E.M.), the Federal Work Study program at the U.S. Department of Education (to B.W.D.), and the Scholarly Inquiry and Research at Emory program at Emory University [to J.J.R.].