Linking healthcare associated norovirus outbreaks: a molecular epidemiologic method for investigating transmission

Noroviruses are highly infectious organisms that cause an acute and short-lived gastroenteritis [1]. They are the most common etiologic agent associated with infectious intestinal disease [2, 3]. Akin to many other gastroenteric pathogens, they are transmitted by the faecal-oral route. But, the virus also causes a high frequency of vomiting. During such an episode, virus is aerosolised. It can then be transmitted directly through the air or can settle and contaminate the surrounding environment or foodstuffs, later to be inadvertently swallowed [4]. Thus the transmission pathways of norovirus are complex and the application of molecular typing of the virus holds promise in furthering understanding of transmission [5]/

Molecular data have been used to describe the genetic diversity of norovirus in various national and regional populations [6–14]. These studies have collectively demonstrated these viruses are genetically diverse and dynamic, with new variants regularly replacing predominant strains. Moreover, molecular techniques are increasingly applied to assess suspected transmission links between outbreaks [5] A number of investigations illustrate this. Identical virus has been detected in patients as well as on an implicated food from a delicatessen meal [15]. A multi-state outbreak has been linked to a common source through tracing of common sequences in patients and a widely distributed oyster product [16]. Internationally-distributed raspberries have been linked through identical sequences following the distribution of a frozen fruit product throughout Europe and Canada [17]. These are only a few of the increasing number of reports which demonstrate the value of molecular genotyping. However, this approach has exclusively been used in the investigation of food and waterborne incidents which are a minority of norovirus outbreaks, at least in European countries [18].

Although these potentially linked outbreaks provide interesting anecdotes, from a statistical/sampling sense they may be dubious. The statistical question – what are the chances of detecting two outbreaks of the same genotype? – is purely data-driven, rather than hypothesis-driven.

Another problem with this approach, which is a general phenomenon in tracking pathogens, is that links will more often be hypothesised amongst rare types than common ones. The example of Salmonella highlights this. International outbreaks of Salmonella serotypes including Newport, [19] Anatum, [20] Saphra, [21] Bovismorbificans [22] and Agona [23], which represent relatively rare types, have been reported. In contrast, the linking of outbreaks caused by the major epidemic type, Salmonella enteritidis phage type 4, has proven extremely difficult [24].

In this paper, we will consider the question of using molecular data to assess norovirus transmission events in healthcare settings. The aim is to develop and test a system that is meaningful at the virus population-level, rather than to highlight rare events. The statistical methods that are applied here attempt to assess the significance of the molecular patterns in light of various epidemiological factors. Sound probabilistic criteria for linking of outbreaks and for assessing differences in molecular epidemiologic patterns with reference to place and time are presented.

Ethical approval for this study was obtained from the South West Multi-centre Research Ethics Committee. In nursing homes, executives provided written consent for the study to take place in their institutions; senior infection control nurses as well as microbiologists provided written consent for their hospital's participation.

In this study, all the norovirus-associated outbreaks in a well-defined healthcare environment were sequenced. We have used this data to develop a system for assessing specific transmission links between outbreaks as well as the probability that the observed clusters were not due to chance.

95% of norovirus outbreaks were due to a single genogroup: II4 (phylogentically similar to what has been referred to by others as the Grimsby or Lordsdale clade). This is consistent with Gallimore et al's recognition that genogroup II4 variants are disproportionately associated with hospital outbreaks [30] and Koopmans' study that suggested that this same genotype is more frequently associated with outbreaks than sporadic community cases [13]. But a fundamental question remains unanswered: what are the unique biological characteristics of this genogroup that make it so predominant in healthcare settings?

It is clear, therefore, that the viruses causing these healthcare-associated outbreaks are not necessarily representative of all the circulating strains in the general community. However, we have no specific reason to believe that the strains that were amplified and sequenced were not representative of healthcare-associated norovirus outbreaks. Given the different genetic diversity in the community and healthcare facilities, we selected the healthcare-associated population of viruses as the reference for these analyses.

Historical data demonstrated that within an outbreak viruses were nearly identical; these data were then used to create similarity criteria. The subsequent analyses illustrate that combining virological and epidemiological evidence may give insight into transmission events. There was statistical evidence of greater similarity in clusters of outbreaks than would be expected from a random sample of the entire viral population. This suggests that transmission between hospital units is important.

A number of outbreak pairs that appeared to be linked based on anecdotal evidence proved to be caused by different viruses based on sequencing. Thus, ad hoc reports of links between outbreaks may not always be valid, especially when incidence is high. The opposite is also true: detection of identical viruses does not assure a direct link in transmission. Naturally, specific information on events can and will be taken into account. In this paper, we propose a method working towards systematising such information. We hope such methods will be used and taken forwards, perhaps in a Bayesian framework where one begins with a prior likelihood that events are linked, and the likelihood is updated based on virological or epidemiological findings. In this study we defined a priori what constituted epidemiological linkage, but, of course, many processes in infectious disease transmission are unobserved and therefore, possibility remains that, for example, that there were multiple introductions of the same virus.

In a hospital setting, where there is clearly potential for internal transmission, one may be drawn to the inference that identical virus necessarily implies a link. In fact, we estimate from this study that 15–20% of outbreaks where virus differs by a single base are really of the same outbreak. This limitation of inference based on genetic data applies our study as well as to other analysis. What this study adds is quantification of the level of inaccuracy.

The approach used here was based on extremely high resolution; inferences about transmission events were based on differences as small as a single nucleotide. However, this study is not the first to suggest that very small changes in the norovirus genome remain conserved geographically and temporally. From Maguire et al's analysis of the polymerase gene sequence in norovirus outbreaks in East Anglia, UK, it was clear that most outbreaks caused in a small geographical region were caused by a single variant and outbreaks in different locations formed different clusters [14].

These conclusions are meaningful because they are drawn from the context of a clearly defined population. Sections of the capsid and polymerase genes were sequenced – more than is often used to make inferences about the molecular epidemiology of noroviruses [5]. Furthermore, the sequencing of multiple viruses from within outbreaks provided a baseline to generate probabilistic statements of how likely it is that two viruses really are from a common source.

Despite these advantages, until a robust genotyping scheme is adopted for noroviruses (or whole genomes are analysed) there is always a distinct possibility that viruses that appear closely related are not really, and viruses that appear different by a few nucleotides in the targeted region are, in fact, related. The primers that have been developed for diagnostics target highly conserved regions [31, 32] – less conserved regions may be more appropriate for these types of studies. Also, the rate of mutation is not precisely known for noroviruses, and the error rate in cloning/sequencing may have a strong bearing on results when small regions are sequenced. In this study, both the polymerase and capsid sequence were used to determine if outbreaks were linked whereas the underlying rates of mutation (i.e. uncertainly in the sequence data) were based almost solely on polymerase sequence data. Furthermore, two primer pairs were used to amplify different segments of the genome. The higher levels of variation in products of SG1/D1 primers may simply reflect greater variability in the target sequence as compared to the Ni/E3 primers.

One can only (or, perhaps, should only) make probabilistic statements about links between incidents. The literature is full of reports that assert links between norovirus [15–18] campylobacter [33–35] and salmonella [19–23] incidents. Clearly, an epidemiological link reinforced with characterisation data is highly suggestive. But what is the probability of randomly selected strains being the same? Background data on the diversity of circulating strains is needed to make this statistical assessment. As we have demonstrated here, once these data are available, they can be used to made clear probability statements about the likelihood that events are related.

By systematic investigation of the virological and epidemiological characteristics of norovirus outbreaks in a hospital population, we were able to investigate transmission processes. The evidence suggests that transmission between hospitals units does occur. Using the proposed criteria, certain hypothesized transmission links between outbreaks were supported while others were refuted. The combined molecular/epidemiologic approach presented here could be applied to other viral populations and potentially to other pathogens for a more thorough view of transmission.