Noroviruses are the leading cause of epidemic gastroenteritis both in the United States and worldwide [1, 2]. In the U.S., noroviruses are the most commonly reported etiology of acute gastroenteritis outbreaks [3] [4]. Person-to-person transmission is the most common route, reported in 66–77% of norovirus outbreaks, while foodborne transmission accounts for 17–26% of outbreaks [1, 5].

While most episodes of gastroenteritis caused by norovirus are mild and self-resolving, severe outcomes, including hospitalization and death, can occur. In the U.S., endemic norovirus illness is responsible for an estimated 56,000–71,000 hospitalizations and 570–800 deaths each year [6]. Vulnerable populations such as young children and the institutionalized elderly are at higher risk for these severe outcomes [6, 7].

Human noroviruses are classified into at least 3 genogroups (GI, GII, and GIV) and 29 genotypes [8]. Since the mid-1990s, genogroup II type 4 (GII.4) viruses have been the predominant genotype worldwide [9, 10] including in the US [5, 11]. New GII.4 variants have emerged every 2 – 4 years and have often been associated with increased norovirus outbreak activity and more severe health outcomes [12–15]. However, this pattern has not been seen with the last two new GII.4 variants (GII.4 New Orleans and GII.4 Sydney) [16, 17].

Despite these observations, it remains challenging to identify which of the myriad host, virus, and setting characteristics (i.e., epidemiologic triad) contribute to more severe outcomes during norovirus outbreaks. For example, it is difficult to distinguish whether the higher mortality rate observed with outbreaks occurring in healthcare facilities [18] is the result of an older, vulnerable patient population [19, 20], or the predominance of GII.4 outbreaks in these facilities [21].

A prior multivariable analysis [18] of data from 843 published norovirus outbreaks from 45 countries found that outbreaks caused by GII.4 viruses had hospitalization rates that were 9 times as high, and mortality rates that were 3 times as high, as outbreaks caused by non-GII.4 viruses, after controlling for other factors. However, because this analysis relied on published outbreaks as opposed to systematic surveillance, it may have been susceptible to publication bias and lack of generalizability.

Since 2009, the Centers for Disease Control and Prevention (CDC) has obtained epidemiologic data on norovirus outbreaks from state health departments through the National Outbreak Reporting System (NORS) [1] and laboratory data through CaliciNet [22]. When data are linked from both surveillance systems, norovirus outbreaks reported to NORS and CaliciNet provide robust epidemiologic data specific to norovirus genotypes, helping to define characteristics of outbreaks that result in more severe health outcomes. This understanding can guide targeted interventions to reduce hospitalizations and deaths from norovirus outbreaks, for instance by informing the genotypic formulation and target populations for vaccines in development [23]. Our objectives were to identify settings, transmission routes, seasonality, symptoms, and patient demographics associated with norovirus genotypes, and to quantify the relationships between these characteristics and hospitalizations and deaths from norovirus outbreaks.

This analysis leverages data from two complementary surveillance systems to provide a comprehensive description of norovirus outbreaks from both the laboratory and epidemiologic perspectives. The matched outbreaks in this analysis represent approximately one-quarter of all NORS norovirus outbreak reports, and over one-half of all CaliciNet outbreak reports from 2009 through 2016. The seasonal distribution of norovirus genotypes in this matched subset is similar to an analysis of the full CaliciNet database [25], and transmission modes of norovirus outbreaks were likewise similar to an analysis of the full NORS database [5], indicating that our results are reasonably representative of all norovirus outbreaks captured in those surveillance systems. Using our linked dataset, we were able to identify several descriptive factors that are associated with norovirus genotype, outbreak hospitalization rates, and outbreak mortality rates.

Notably, we found that GII.4 norovirus outbreaks occurred more frequently in healthcare facilities, affected an older age group, and were more likely to occur during the typical norovirus season (November – April). In non-healthcare settings, GII.4 norovirus outbreaks were significantly more likely to be associated with foodborne transmission, whereas no effect was seen in healthcare facilities. In terms of severe outcomes, outbreaks caused by GII.4 genotypes and occurring in healthcare facilities had higher rates of hospitalization and death. While outbreaks with a high proportion of female cases and outbreaks occurring during the norovirus season were significantly associated with lower hospitalization rates, these characteristics were not significantly associated with mortality rates in multivariable models. However, the analyses of mortality rate may be limited by lower power due to low numbers of deaths in some strata.

Several key findings are consistent with prior studies of the epidemiology of norovirus outbreaks. The associations of GII.4 genotypes with outbreak setting and transmission route have also been observed in a prior analysis of outbreaks occurring in 1994–2006 [26], 2009 – 2013 [12], some prospective studies [27], and systematic reviews [18, 28]. GII.4 was the most commonly reported cause of outbreaks. Further, GII.4 outbreaks were longer in duration and resulted in more severe illnesses, which may be a result of prolonged viral shedding [29] or escape from prior immunity caused by novel strains [30].

The results of our regression models offer some insight in differentiating the inter-related characteristics of genotype, transmission route, setting, and host in impacting severity of norovirus outbreaks. While univariate models of genotype show that foodborne transmission is inversely associated with GII.4 genotype, this relationship unexpectedly reverses in the multivariate model. Stratifying the analysis by setting helps explain this change, as the positive association between foodborne transmission and GII.4 genotype appears to be driven fully by outbreaks occurring in non-healthcare settings; among outbreaks occurring in healthcare settings, the association is negative or null once adjusted for other outbreak characteristics (e.g., demographics). These results are somewhat surprising given previous findings of an association between foodborne transmission and non-GII.4 outbreaks [11]; however, the analysis done by Vega et al. did not stratify by setting, and both our study and theirs noted a small number of foodborne outbreaks in the healthcare setting. Further, the Vega et al. study included all norovirus outbreaks reported to CaliciNet from 2009 to 2013, whereas our study spanned from 2009 to 2016 and included only CaliciNet norovirus outbreaks that were able to be linked to their corresponding report in NORS. While our dataset had a small number of outbreaks in non-healthcare settings in 2009, we found the associations with foodborne transmission and GII.4 to be qualitatively similar in all other years. An important caveat to our results is the low number of foodborne outbreaks in healthcare settings, which may have limited our power to detect associations or reflect biases in reporting which would have affected our results.

With respect to severe outcomes, our multivariable models show that GII.4 genotype and healthcare settings are independently associated with both higher hospitalization and mortality rates. The association of healthcare settings with hospitalizations and mortality is perhaps unsurprising, given the likely co-morbidities of individuals in healthcare settings [6, 7]. However, somewhat unexpectedly, older age and symptoms of fever and vomiting were not consistently significantly associated with these outcomes in multivariable models, though point estimates of incidence rate ratios were >1 for hospitalization. It may be that there was not enough variation in outbreak demographics to show statistical significance. Perhaps surprisingly, outbreaks occurring November – April had lower hospitalization rates after controlling for other factors, despite the fact that most outbreaks occur during this time period, especially GII.4 outbreaks. It could be that outbreaks occurring “out-of-season” are different qualitatively, even within genotypes. Alternatively, perhaps these lower hospitalization rates reflect competing syndromes during the winter season, such as acute respiratory infections (like influenza), which may also disproportionately affect vulnerable populations such as young children and the elderly. Analyses stratified by setting also showed a significant effect of GII.4 on hospitalization and mortality rates (data not shown). These results suggest that GII.4 genotypes cause more severe illness than non-GII.4 genotypes, regardless of the types of symptoms manifested or the location of the outbreak. Our results also further confirm the increased likelihood of severe outcomes in healthcare settings.

The association of GII.4 genotype with higher hospitalization rate adds to the mixed body of evidence of increased clinical severity caused by emergent GII.4 strains. A time-series analysis by Lopman et al. showed increased rates of gastroenteritis hospitalizations in 2006–2007 coinciding with the emergence of GII.4 Den Haag, although the etiologic attribution to norovirus in this model was indirect [31]. Conversely, there was no increase in hospitalization rates in five states participating in a norovirus sentinel surveillance network during the emergence of GII.4 Sydney in 2012–2013 [16]. The association of GII.4 genotype with higher mortality was seen in a systematic literature review [18], and increased norovirus-associated mortality has been estimated to occur during seasons in which pandemic GII.4 variants emerged (i.e., 2001–2002 and 2006–2007) [32]. It may be that some strains are more virulent than others. In a sub-analysis of our data, we found that GII.4 Den Haag was associated with significantly higher rates of mortality and morbidity when compared to GII.4 New Orleans or GII.4 Sydney; however, it should be noted that GII.4 Den Haag was reported in fewer outbreaks, and data were too sparse to control for other factors.

Our analysis is subject to several limitations. Both NORS and CaliciNet are surveillance systems for reporting outbreaks; therefore, these results may not be generalizable to endemic norovirus disease. Furthermore, both systems are voluntary with a minimal set of required data, resulting in incomplete reporting on demographic, symptom, and outcome variables. There may be bias towards reporting outbreaks that are larger in size, longer in duration, or those associated with more severe outcomes, and reporting and coding practices may differ by state. Additionally, our ability to characterize the host is limited to the collection of aggregated data on age and sex. Other host characteristics, such as co-morbidities and genetic predisposition, would be helpful in further clarifying the contribution of host factors to clinical outcomes. Lastly, given the relatively low numbers of reported deaths, our power may have been limited in analyses of mortality, particularly in multivariable analyses.

Despite these limitations, this analysis is unique in its size (3,747 outbreaks) and comprehensiveness of included laboratory and epidemiologic data. The results offer valuable insights into the interrelationships among the epidemiologic triad of norovirus genotype, outbreak setting, and host demographics that contribute to the severe outcomes of hospitalization and death. To reduce hospitalizations and mortality from norovirus outbreaks, GII.4 viruses should be included in norovirus vaccines in development [33], and individuals in healthcare settings should be considered as potential targets. Furthermore, providers and public health officials should be vigilant for the possibility of more severe outcomes when a novel GII.4 variant emerges.