IJE Advance Access originally published online on January 24, 2006
International Journal of Epidemiology 2006 35(2):354-360; doi:10.1093/ije/dyi317
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Article |
What is the probability of successive cases of Legionnaires' disease occurring in European hotels?
1 Health Protection Agency, Centre for Infections, London, UK
2 London School of Hygiene and Tropical Medicine, London, UK
* Corresponding author. Respiratory Diseases Department, Health Protection Agency Centre for Infections, 61 Colindale Avenue, London NW9 5EQ, UK. E-mail: katherine.ricketts{at}hpa.org.uk
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Background Public health officials will normally take action at accommodation sites following an association with a cluster of cases of Legionnaires' disease. This paper seeks to determine the likelihood of such a cluster occurring at a site once it has been associated with a single case of the disease, and therefore whether more should be done at sites following individual cases.
Methods Information for UK residents reported to the EWGLINET system between 1993 and 2000 was included in a dataset. The size and country of hotel visited by the cases were divided into six country groups (France, Italy, Spain, Turkey, other Europe and other World), and eight size groups (<20 rooms, 20–49, 50–99, 100–199, 200–299, 300–399, 400–499, 500+). The data were investigated using Cox proportional hazards regression to model the probability of at least one further case following the first.
Results The dataset comprised 793 cases that had stayed at 605 sites in 51 countries between 1993 and the end of 2000. This included 605 cases that were the first case associated with a site, and 188 subsequent cases. Following the first case, 16.6% of sites were associated with at least one subsequent case during the period under study. The probability of a subsequent case occurring within 6 months of the first varied by country and size group, with some combinations returning a probability >30%; the probability of a subsequent case occurring within 2 years of the first reached over 50% in some instances.
Conclusions There may be support for early intervention at some accommodation sites following a first case of Legionnaires' disease, in specific country and size groups.
Keywords Legionnaires' disease, international surveillance, travellers, survival analysis
Accepted 20 December 2005
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Tourists are at risk of acquiring travel-associated Legionnaires' disease. The Health Protection Agency Centre for Infections in London, co-ordinates a European surveillance scheme (EWGLINET) on behalf of the European Working Group for Legionella Infections (EWGLI),1 which aims to detect clusters of cases of travel-associated Legionnaires' disease linked to accommodation sites and provide a rapid alert for local public health action. The EWGLINET surveillance scheme was established in 1987 and has been co-ordinated by the Health Protection Agency since 1993. Results from the scheme are regularly published.2,3 Cases are defined as residents of the 37 European countries in the scheme who contracted Legionnaires' disease after spending one or more nights away from home staying in hotels, campsites, cruise ships or similar types of accommodation in the 2–10 days prior to their onset of illness. Clusters are now defined as two or more cases of Legionnaires' disease associated with the same place of accommodation and with onset of illness within two years of each other. However in the years 1993–2000 (which are dealt with in this paper), a cluster was similarly defined but with onset of cases within 6 months of each other. The definition was changed at the beginning of 2001. Between 1993 and 2000, over 1700 cases of Legionnaires' disease were reported to EWGLINET in European travellers, and at least 8.7% of these cases died as a result of their infections.
In July 2002, EWGLINET introduced guidelines for the control and prevention of travel-associated Legionnaires' disease.4 These guidelines ensure that control measures and environmental sampling are carried out at cluster sites by the country of infection. In addition, they enable tour operators to fulfil their legal duty of care to clients under The European Council Directive on package travel, package holidays and package tours5 that was implemented in most EU member states by 1996.6 In situations where the measures specified by the guidelines are not carried out, or are carried out inadequately, EWGLINET publishes the name of the accommodation site on its website (www.ewgli.org) stating that the co-ordinating centre has not been assured that measures have been taken to minimise risk at the site. In response to these reports, tour operators must assess what action they should take to minimise risk to current and future clients at these sites.
Prior to the introduction of these guidelines, there was no European wide policy on the response to clusters at public accommodation sites, so that any responses that did occur were country-specific. In addition, the rare interventions that occurred often followed a large outbreak and were frequently initiated by the actions of tour operators. Following a single-case report, tour operators would typically have issued a standard 14-point checklist of good practise in legionella control (an updated version is available in the EWGLI guidelines4), but are unlikely to have done more. There is no record available which shows where the checklist was or was not issued, so for the purposes of this paper, it has been assumed that no interventions were carried out at these hotels following a single case of Legionnaires' disease.
EWGLINET is particularly interested in estimating the probability of a second case occurring within six months of the first (i.e. the probability of a cluster during the 1993–2000 time period under discussion). When a single case is associated with a building, it is difficult to determine whether that building is the source of the case's illness, since the patient could have been exposed to any number of possible sources of infection. However, if two cases are associated with the same place within a short period of time then the likelihood of that building being the source is greatly increased. Therefore, during the period under study, the occurrence of two cases within 6 months was usually the threshold at which public health officials and tour operators would intervene at a site. If the accommodation site was the source of infection, any effective intervention taken should have reduced the chance of further cases occurring. Thus all subsequent cases were, in theory, preventable. If the accommodation sites that had subsequent cases showed distinct characteristics, these could be used in the future to predict which hotels are most likely to have another case after the report of a first. Preventive measures could then be better targeted and action taken after the first case at these ‘high risk’ sites.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
The traditional ‘cohort study’ style of analysis was not possible in this instance because EWGLINET only hears about cases of Legionnaires' disease arising from accommodation sites and does not get information on the numbers of people staying at these sites and not getting ill. Additionally, this study could not include all sites at which there has never been a case of Legionnaires' disease.
Therefore, a subset of EWGLINET surveillance data was analysed to assess how the risk of a site being associated with a case of travel-associated Legionnaires' disease changes over time after it has been associated with a first case. The way in which the risk varied over time by size of accommodation and country was investigated using Cox proportional hazards regression7 to model the probability of at least one further case within 6 months following the first. Time to acquisition of Legionnaires' disease since the index case was used as the dependent variable together with a variable indicating whether the observation was subject to censoring. The Cox proportional hazards assumption was tested and found to be satisfactory (P = 0.249).
Information for cases reported to EWGLINET that occurred in UK residents between 1993 and 2000 inclusive was used. Because people with different nationalities show different travel patterns,8 only travellers from the United Kingdom were used in the analysis, in order to remove this potential confounding variable from the dataset. The proportion of cases reported by each nationality will therefore not need to be taken into account. An additional reason for excluding cases from outside the UK is that the travel history collected for cases diagnosed in other countries may be more or less complete and this could bias the results.
169 cases had stayed at more than one place of accommodation in the incubation period but only one site was included in the analysis. This included 50 cases where the case visited both single and cluster sites and were allocated preferentially to the cluster sites, and 119 cases who visited only single sites and were allocated alphabetically based on site name.
The data were grouped for analysis into 13 country groups (Americas, Caribbean, Cruise, Far East/Asia, France, Greece, Italy, Other Europe, Other World, Portugal, Spain, Turkey, UK) and eight size groups (<20 rooms, 20–49, 50–99, 100–199, 200–299, 300–399, 400–499, 500+), which gave 104 combinations of country and size group. The country groups were then combined in order to make sample sizes larger and analysis simpler, giving six new groups: France, Italy, Spain, Turkey, Other Europe (including Greece, Portugal, UK, and Other Europe), and Other World (Americas, Caribbean, Cruise, Far East/Asia, and Other World).
A number of assumptions had to be made:
- All the cases acquired their infections at the accommodation site in which they were staying and which was included in the study (if the case stayed at more than one site).
- The percentage occupancy of hotel rooms was similar for all accommodation units and did not change markedly with time.
- Seasonality did not affect one country or size of hotel in an unrepresentative manner. Legionnaires' disease does show seasonal patterns, but this paper looks at results over an 8-year period, so it has been assumed that any seasonal effects are similar between the country and size-group categories over this period.
The analyses would have benefited from hotel-level visitor denominator data to add into our model, but unfortunately this was not available.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
The subset of data analysed in this paper encompasses 793 cases that had stayed at 605 sites in 51 countries between 1993 and the end of 2000. This included 605 cases that were the first case associated with a site, and 188 subsequent cases [with time to onset after the first case ranging from 0 days (onset on the same day) to 370 weeks]. In addition, 188 cases stayed in private accommodation or in sites for which no size could be determined, and were excluded from the analysis.
Following the first case, 16.6% of accommodation sites were associated with at least one subsequent case during the period under study (some with more than one). The probability of a subsequent case within 6 months of the first at the same site varied by country and size group, for example in France it ranged from 1.7% (size Group 2; 20–49 rooms) to 7.7% (size Group 6; 300–399 rooms) [size Group 6 was the category with the second highest probability of subsequent cases; size Group 7 gave the highest probabilities, but some of the size-country categories contained no cases, making the data less robust (Table 1)], and in Spain from 8.9% (size Group 2) to 35.7% (size Group 6). When the time period was increased to 2 years, the probability of a subsequent case increased as well, varying from 2.8 to 12.5% in France and from 14.4 to 52.1% in Spain, so that when a hotel of size Group 6 in Spain was associated with a single case of Legionnaires' disease, it was more likely than not to be associated with at least one subsequent case within 2 years. Figure 1 shows these probabilities for each of the combined country categories, using size groups selected to demonstrate two extremes of the data.
|
|
In general, the larger hotels had a higher risk of subsequent cases following a first case (Figure 2). Hotels in size Group 7 (400–499 rooms) had the highest risk, followed by sizes 6, 5, 8, 4, 3, and then size Groups 1 (<20 rooms) and 2 (20–49 rooms) with a notably lower risk. This analysis was independent of country group, with this pattern occurring across all groups. There was insufficient data to reliably investigate whether the risk pattern for the different sizes may vary between country groups.
|
The overall level and risk also varied by country, independent of size group (Figure 3). Some countries (Italy especially, and France) had a very low probability of subsequent cases regardless of hotel size. Spain and Turkey had a relatively high probability of subsequent cases, whilst the combined categories of ‘rest of Europe’ and the ‘rest of the World’ gave probabilities that fell in-between the two extremes.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
This study aimed to determine whether the risk of a subsequent case of Legionnaires' disease following a first case at an accommodation site is sufficiently high to justify some early form of public health intervention. (This paper does not attempt to determine whether any intervention currently available would be effective.) The data analysed covered the period from 1993 to the end of 2000.
Once a hotel has legionella present in its water system at levels to present a risk, the probability of the hotel being associated with a case of Legionnaires' disease can be expected to increase as the size of the hotel increases simply because there will be more people staying there. Additionally, larger hotels have to deal with a bigger load on their water systems and a larger water distribution system across which to regulate the temperature. If the chance of a case occurring per person per week were the same at all hotels, the probability of a further case occurring at a hotel should increase in direct proportion to the size of the hotel. This was true in general, but was not observed for every category of hotel size (e.g. size Group 8 was notably not in direct proportion and Groups 1 and 2 were a much lower risk than the other sizes), and therefore suggests that the risk is not the same at all hotels.
There are two factors that must be acknowledged as potential confounding variables, and which could potentially influence the risk derived from this analysis.
- The age of clients was not similarly distributed at all accommodation sites. When a Kruskal–Wallis One-Way test and an ANOVA test9 were performed, significant differences between age distributions of the cases and country of stay, and age and size category were found. It has been well established that people over 50 years of age2,3 are at increased risk of Legionnaires' disease. Therefore the different age groups staying in different country and size categories could have confounded the results.
- The length of stay ranged from 1 night to 48 nights, and again varied significantly by size category and country of visit when analysed using Kruskal–Wallis and ANOVA tests. The potential for length of stay to confound this analysis is unclear. If a case contracted Legionnaires' disease from the water system of an accommodation site, it is probable that they would have had a high initial exposure, but that the exposure would then have decreased over the rest of the stay (once the bacteria has been flushed out of the room's water system). However, other sources of infection may be present for the duration of the case's stay, such as a water cooled air conditioning system and, if this were the source, an individual with a longer period of stay at an accommodation site would be exposed to infection more than an individual with a shorter duration of stay.
For both of these potential confounders an attempt was made to obtain denominator data. Unfortunately, it was not possible to obtain data on the age or length of stay of tourists in specific hotels or data on the denominators of hotels in each country.
Additionally, the method of case allocation to accommodation site may have led to a small overestimated risk of a cluster occurring, because 50 (out of 793) cases were preferentially assigned to cluster sites rather than single sites. However, this would not have biased the results towards larger or smaller accommodation sites, or towards one country or another.
There are two changes that have taken place to the EWGLINET scheme since 2000 that are relevant to this paper. At the beginning of 2001, EWGLINET's cluster definition changed from two cases associated with the same accommodation site within 6 months to two cases at the same accommodation site within 2 years. This change was intended to allow countries to detect a greater number of problematic sites and, by dealing with them promptly, decrease the number of future cases associated with those sites. For instance, one hotel in Spain was associated with a case in 1991, a second case in 1993, one in 1994, and a fourth in 1995. None of these cases fell within 6 months of each other, and so a cluster was never called under the original EWGLI definitions, despite there being an obvious problem at the hotel. Under the new definitions, this would have been a cluster after the first two cases, and each subsequent case (had they occurred) would have been a cluster update, each leading to a new investigation at the hotel.
The second change to the EWGLINET scheme occurred on July 1, 2002 when the European guidelines4 were introduced, standardising the investigation of cluster sites across Europe. Again the intention was to ensure that, once a cluster site is identified, the actions taken at that site are sufficient to prevent future cases of Legionnaires' disease. The guidelines have only been in effect for 3 years at the time of writing. More time is needed before it is possible to determine whether they are showing any success at preventing further cases of the disease at cluster sites.
It is anticipated that both of these changes to EWGLINET procedures will have an impact on the probability of future cases occurring at cluster sites in all countries using the guidelines. Once 10 years' worth of post-guidelines data has been collected, a repeat of the analysis carried out here should be conducted and could yield some interesting comparative data, which will provide a solid basis for evaluating the long-term effects of the modifications made to the scheme.
The results presented in this paper may be an underestimate of the true incidence of clusters since some hotels were associated with cases from outside the UK and these cases were not included in the dataset. Although the size of hotel was included in the analysis (measured as number of rooms) it was not possible to get this information for a proportion of the sample. These hotels with unknown size might have differed in some way from those for which size was known. If this difference also made them more or less likely to be the source of legionella infections, bias could have been introduced into the dataset.
It is possible that a second type of bias could have affected the data if the smaller hotels were less likely to have websites than the larger hotels, making it more difficult to find their sizes. It also had to be assumed that the proportion of rooms occupied at any time was the same for all the hotels, whereas in some country and size groups it is highly likely that the occupancy varied by season and this would have influenced the results.
The risk to travellers does vary somewhat with the size of the hotel since tour operators tend to associate with bigger hotels, which probably have more stringent health and safety programmes in place. If such a hotel is the source of a cluster of Legionnaires' disease, they have great incentive not to ignore the cluster since the tour operators can threaten to withdraw from the hotel if they feel that their clients are at risk. As well as variation in the probability of tour operators contracting with different sized accommodation sites, there is also variety across countries in as much as France tends to attract independent travellers in comparison with package tourists from the UK who often visit Spain and Turkey. Again, the hotels used by package tourists are likely to be subject to tour operator's health and safety inspections, and therefore maintain higher health and safety standards.
It has been suggested that hotels that have been associated with cases of Legionnaires' disease in the past will be more likely to experience future cases, when compared with hotels that have never been associated with a case. If this were true, it would be an important predictive factor when estimating the probability of a subsequent case, but unfortunately there were not enough data in this dataset for this to be explored. Out of the 605 accommodation sites that had a case, only 188 went on to have at least one subsequent case.
|
Conclusion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Despite the limitations of the data, the results suggest that, in some countries (for instance Spain and Turkey) and especially where accommodation sites with 400–499 rooms (size Group 7) are involved, the risk is sufficiently high to justify the early use of an effective public health intervention at a site associated with a single case of travel-associated Legionnaires' disease. This should help to prevent further cases and deaths from the disease among tourists.
Since the data presented in this paper, EWGLI has introduced guidelines that aim to control risk at accommodation sites. It remains to be seen whether they will successfully reduce the risk of infection in accommodation sites across Europe.
KEY MESSAGES
|
|
Acknowledgments |
|---|
We'd like to thank A Swan and C Bartlett for their assistance in developing an earlier draft of this paper, and A Kuwabara for her assistance in allocating hotels to size groups. This work was partially funded by the European Commission Health and Consumer Protection Directorate-General.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
1 Hutchinson EJ, Joseph CA, Bartlett CLR. EWGLI: A European surveillance scheme for travel associated Legionnaires' disease. Euro Surveill 1996;1:37–39.[Medline]
2 Ricketts KD, Joseph CA. Travel associated Legionnaires’ disease in Europe: 2004. Euro Surveill 2004;9:40–43.[Medline]
3 Ricketts KD, Joseph CA. Travel associated Legionnaires’ disease in Europe: 2002. Euro Surveill 2005;9:6–9.
4 European Working Group for Legionella Infections. European Guidelines for Control and Prevention of Travel Associated Legionnaires' Disease. London: PHLS, 2002. Available at: www.ewgli.org (Accessed on November 14, 2005).
5 European Council Directive 90/314/EEC of 13 June 1990 on package travel, package holidays and package tours.
6 Anon. Legionella and the European Council for Package Travel. CDR Wkly 1996 21st June.
7 StataCorp. Stata Statistical Software: Release 8.2. Survival Analysis and Epidemiological Tables. College Station, Texas: Stata Corporation, 2003.
8 World Tourism Organisation. Yearbook of Tourism Statistics—2003 Edition. 2003. ISBN: 9284406056.
9 Bland M. An Introduction to Medical Statistics. 3rd edn. Oxford: Oxford University Press, 2000.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||