Outcome of the investigation
Our results suggest that this was a point-source outbreak, confined to race participants, in which illness was associated with exposures which occurred during the race. Attack rates did not vary significantly with sex but were higher in younger participants. Existing immunity is likely to be a factor in this observation. Univariate analysis identified inadvertent ingestion of mud as the strongest association, a finding that was confirmed by logistic regression. Ingestion of a number of food and drink items was associated with increased risk of illness in the univariate analysis, a finding which is consistent with the widespread soiling of hands, equipment and utensils with mud. Although not statistically significant, eating bananas from the feed station was associated with a reduced risk, possibly due to the fact that they were eaten directly from their skins and so avoided contamination.
It is likely therefore, that heavy rain preceding the off-road event caused mud to form and mix with animal faeces on the course. The mud then acted as an efficient vehicle for pathogenic organisms contained within the animal faeces. Although microbiological confirmation has not been received for the majority of the symptomatic participants, our results are consistent with a point-source Campylobacter outbreak.
The OCT made a number of recommendations aimed at both organizers and riders. These included instructions not to serve open food at feed stations and suggestions to avoid the use of obviously contaminated food and drink containers. Provision of hand and bottle washing facilities was also suggested, as was course re-routing in the case of obvious animal faecal soiling. However, it is important to recognize that the nature of the terrain and the often remote locations may mean that such steps are difficult or impossible to implement. Such interventions therefore aim to reduce rather than eliminate the chance of illness. Mountain biking, along with many other sports conducted in the natural environment, involves a degree of risk. Zoonotic infection is one of these risks. Given that the acceptance of physical risk is necessary to participate in such sports, it is not clear how this additional microbiological risk would be perceived and whether food hygiene recommendations would be followed. This issue would benefit from further research. However, while many might discount the importance of this additional risk, organizers should raise awareness for the benefit of those who might be particularly affected by infective gastrointestinal disease such as the immunocompromised and food handlers.
The outcome of the investigation was fed back directly to the event organizers. The results were then disseminated to riders in a number of ways. An email summarizing the findings of the investigation and containing a link to the final report was sent to all individuals on the original mailing list . This was supported by a press release which was circulated to both general and specialist biking media.
The nature of mountain biking as a sport means that riding across land used by both agricultural and wild animals is unavoidable and indeed desirable. The risk of infection from zoonotic organisms will therefore always be present. In this outbreak, the heavy overnight rain appears to have augmented the risk. This finding is not without precedence. Outbreaks of Escherichia coli O157 have been reported after a scout camp [Reference Howie10] and a music festival [Reference Crampin11], both of which were affected by heavy rain and held on land usually used by livestock. Muddy conditions were also linked with the campylobacteriosis outbreak associated with a mountain-bike event in Canada [Reference Stuart5]. Interestingly, the investigation of the 1998 Wisconsin Triathlon leptospirosis outbreak also blamed preceding heavy rain for causing heavier than usual contamination of the lake water, with a history of having swallowed lake water being identified as the greatest risk factor [Reference Morgan3].
The internet is still a relatively novel tool for conducting outbreak investigations, although its use is increasing [Reference Ghosh12–Reference Raupach and Hundy17]. This was our first experience of using the internet in this way and was prompted by the evident computer literacy of the event participants. In common with other authors we believe our work has demonstrated that efficiency and timeliness are the most significant strengths of the method [Reference Ghosh12, Reference de Jong and Ancker13, Reference Passaro, Scott and Dworkin15, Reference Kuusi16, Reference Castrodale18]. The use of email invitations and an internet-based questionnaire allowed us to rapidly and efficiently access the geographically dispersed population at minimal cost [Reference Rhodes, Bowie and Hergenrather19]. Automated data entry significantly reduced the time and resources needed to conduct the analysis and also reduced the chance of bias due to inaccurate data entry [Reference Rhodes, Bowie and Hergenrather19]. This resulted in the publication of a preliminary outbreak within 4 weeks of the first notification. The automatic download of response data allowed us to conduct a limited analyses of accommodation and catering exposures within 24 h of the questionnaire going live. The non-significant results linked to these exposures corroborated the results of the site inspections and helped reassure us that no ongoing risks were present. This interim analysis did not influence the design of the final analysis but it demonstrates that the internet method is also flexible in its application.
As with most field investigations, our work has a number of weaknesses. Our method clearly excluded those who do not have access to the internet. This may have caused selection bias by allowing over-representation of certain groups, for example younger people and those with greater educational attainment and computer literacy [Reference Kuusi16, Reference Castrodale18, Reference Rhodes, Bowie and Hergenrather19]. In our study, this is less likely to be a problem since it was evident that the majority of our population had internet access. However, it is true that only individuals with an electronically recorded email address were invited to participate. This decision was made to ensure that the investigation was conducted rapidly and with efficient use of resources, but meant that around 30% of entrants were not included in the study population. Although we have no data to suggest there is a systematic difference between these two groups, it is possible that this may have introduced a selection bias. For example, the decision to submit a paper entry form may be associated with age, which in turn would be associated with pre-existing immunity. However, we believe that such biases are unlikely to have significantly affected the results of our investigation due to both the size of the included population and also because the risk of being exposed is unlikely to have been associated with whether participants entered the event online or via a paper entry form.
Our conclusion about the likely responsible organism would have been strengthened if we had further microbiological evidence. The decision not to conduct any environmental sampling was taken since almost 2 weeks had elapsed by the time the outbreak was reported, by which time the faecal contamination was likely to have been dispersed. It was also felt that it would be difficult to identify which parts of the course should be tested and that, as no further events were planned, the value of environmental sampling was not clear. Similarly, although sampling of cases would have been desirable, testing of such a widely dispersed population would have been problematic. As the online questionnaire was completed anonymously, we were unable to verify any reported positive results or request samples on symptomatic individuals. Indeed many had recovered by the time the investigation began. None the less, the outbreak curve is consistent with an outbreak of Campylobacter, as are the most frequently reported symptoms. We are therefore confident that we have identified the likely responsible pathogen.
In common with many retrospective cohort studies, it is probable that those who were ill were more likely to respond and thus elevate attack rates. The response rate of 52·7% is also relatively low. While some investigators have managed to obtain higher response rates using internet-based methods, these have often, but not exclusively, been on relatively small, well defined or easily accessible groups [Reference Stuart5, Reference Ghosh12, Reference de Jong and Ancker13, Reference Raupach and Hundy17, Reference Castrodale18]. In contrast, other outbreak investigations involving larger community-based populations have obtained lower response rates than we achieved, with some as low as 11% [Reference Srikantiah14, Reference Passaro, Scott and Dworkin15]. Although it is therefore possible that we have obtained biased estimate of the risks in this outbreak, we believe that the numbers involved in our investigation are sufficiently large for the overall findings to be robust.
The case definition we chose to use was sensitive and so it is possible that individuals who were ill for other reasons may have been included as cases and caused biased estimates. However, if such misclassification bias did occur it is likely that it would result in a reduced effect size and so make the identification of significant risks less likely. However, even with this sensitive case definition, clear risks did emerge. Sensitivity analyses were conducted with more specific case definitions, but the results were very similar, suggesting that the effect of any misclassification was small.
All survey methods need to address issues of data protection, data security and confidentiality and need to discuss them openly with participants [Reference Castrodale18, Reference Rhodes, Bowie and Hergenrather19]. In our study, data were to be automatically entered onto a secure database which was housed within an NHS system. At the outset of our investigation we were concerned that public uncertainty about the security and confidentiality of electronic methods may have discouraged response to the survey. In view of this potential concern, we elected to allow questionnaires to be completed anonymously and did not collect any identifiable information. This decision meant that we were not able to establish a mechanism for ensuring that individuals did not completed the questionnaire more than once. The possibility of duplicate responses is a recognized problem of electronic survey methods, as is the possibility that invitations are forwarded to unintended respondents [Reference Castrodale18, Reference Rhodes, Bowie and Hergenrather19]. Although it is possible that individuals completed the questionnaire more than once, it is difficult to identify a motivation for such duplication in this situation. We therefore judged that the impact of this potential bias is negligible. Forwarding of emails is also unlikely to have caused a problem and may have been beneficial as it would have reached more of the target population.