3.2 Literature review: Legionnaires' disease outbreak detection

Scotland, United Kingdom 1978-1986(Bhopal et al. 1992; Dunn et al. 2007)[1 & 2]

452 apparently sporadic cases of Legionnaires' disease occurred in Scotland between 1978 and 1986 (Bhopal et al. 1992), before the widespread application of GIS. The cases were classified as sporadic as it was not possible to assign them to any wider outbreak using the standard methods that existed at the time. Bhopal et al. (1992) revisit these data and analyse them from a geographic perspective, seeking evidence of clustering using four methods: tabulation of cases by health board area and time of onset of symptoms; mapping of case home locations; identification of potential clusters on listings of cases sorted by residential postcode and date of symptom onset; and statistical testing using an extension of Knox's test for space-time interaction (Knox, 1963). They find that space-time clustering is present and identify numerous groups of cases including an outbreak of 9 cases which had apparently been completely missed. Bhopal et al. note that the clusters were recognised solely on the basis of residential postcode and that if other information on case location had been available, such as workplace, then other clusters would probably have been found.

Dunn et al. (2007) also revisit the Scottish data and apply more recently developed GIS techniques in an effort to foster a better understanding of the spatial epidemiology of Legionnaires' disease. They used GIS to help answer two research questions: (i) is the risk of infection related to distance and direction of home residence from cooling tower locations; and (ii) what is the impact on risk of living in the vicinity of multiple cooling tower locations. GIS were used to define two concentric circles around each cooling tower site, at 0-500m and 500-1000m. These two zones were each sub-divided into four sectors to help capture the effect of wind direction. Standardised rates and ratios were then calculated in order to compare the counts of Legionnaires' disease cases against those of a control group, with somewhat inconclusive results. Dunn et al. also used raised incidence modelling (Diggle and Rowlingson, 1994) to assess the extent of clustering of cases around cooling towers. Their findings support the assertion that distance of home residence from the nearest cooling tower is associated with a higher risk of acquiring Legionnaires' disease. A wind direction effect was also found to be significant but was not precisely estimated.

Denmark, 1990-2005(Rudbeck et al. 2010)[3]

Rudbeck et al. (2010) analyse 606 sporadic cases of Legionnaires' disease that occurred in Denmark between 1990 and 2005. No outbreaks had previously been identified in this period. Each case was geocoded based on residential address and assigned to a corresponding cell on a 10km x 10km grid covering Denmark. Baseline population figures for each cell were then used to calculate the incidence of Legionnaires' disease in each cell for the time periods 1990-1994, 1995-1999, and 2000-2005. It was assumed that the number of cases in each cell could be described by a Poisson model. With a null hypothesis that the true underlying rate for cases in a given cell was the same as the underlying rate overall in the country, it was then possible to identify cells with an incidence higher than the upper limit of the 95% and 99% confidence intervals. Persistent high incidence in two neighbouring cells was detected, as well as two community clusters of two cases each, one of which had not previously been noticed. The study demonstrates that historical data in a grid format may be useful in a GIS-based surveillance model as a continuously updated background layer.

The model has the advantage of being simple but could be improved upon by mapping of workplace address and inclusion of microbiological information and other typing results. This would increase the probability of detecting specific clusters, or an environmental or domestic source.

Netherlands 1999-2006(van den Wijngaard et al. 2010)[4]

van den Wijngaard et al. (2010) use data from the two largest outbreaks of Legionnaires' disease in the Netherlands as controls in a study which simulates prospective syndromic surveillance for lower respiratory infections. The outbreaks occurred in March 1999 among attendees of a flower show (Den Boer et al. , 2002) and in July 2006 in Amsterdam.

A space-time scan statistic was used which compares the observed number of cases in circular areas with variable radii in flexible time periods against the expected number of cases, based on the geographic distribution of cases in the whole dataset (Kulldorff et al. 2005). Cases were defined based on hospital records which had any kind of lower respiratory infection as either discharge or secondary diagnosis. SaTScan software (Kulldorff et al. 2006) was used to implement the computations and visualise the results. Performance was then evaluated by the statistic's ability to detect the known Legionnaires' disease outbreaks, and the earliest detection dates for these outbreaks for daily and weekly analysis was compared.

Both outbreaks were detected by the scan statistic. Daily analysis signalled the 1999 outbreak 4 days earlier than weekly analysis, and 2 days before the national alarm was given at the time. The 2006 outbreak was detected by daily analysis 3 days after the national alarm was given. The authors consequently recommend that real-time syndromic surveillance using a space-time scan statistic is a potentially useful tool for detection of local lower respiratory infection outbreaks. However it is vital to obtain syndromic data with sufficient quality and coverage. It would also be desirable to incorporate epidemiological and microbiological data.

Metropolitan France, 1998-2000(Che et al. , 2003)[5]

Che et al. (2003) describe an ecological study which aimed to quantify the risk of developing Legionnaires' disease from data aggregated on a geographical basis. They compiled an inventory of industrial sites involved in generating water aerosols in France over the years from 1998 to 2000, and obtained summary counts of such sites by postcode area. These were combined with other postcode-level data on Legionnaires' disease cases; population; urban/rural status; and relative geographic location in order to explore the relationship between Legionnaires' disease incidence and exposure to industrial aerosols and plumes of smoke. Exposure to industrial aerosols and plumes of smoke were both found to be significantly associated with an increase in incidence of Legionnaires' disease. Exposure to more than one source of aerosol was also found to be associated with an increase in incidence.

This study has an ecological design and as such does not take individual risk factors for Legionnaires' disease into account. A further limitation is that it only considers residential postcode.

Netherlands, 2010(Euser, 2010)[6]

Euser (2010) describes a GIS developed for detection of Legionnaires' disease clusters in the Netherlands. It is web-based facility which is available to all Municipal Health Services for their own particular mapping purposes. Data are included on patient characteristics such as home address, date of symptom onset, and microbiological diagnosis as well as potential source characteristics such as location, type, and sampling results. Euser argues that the availability of GIS functionality to local health service teams may lead to improved source identification and more rapid cluster detection and response.

1. BHOPAL R. S., DIGGLE P. & ROWLINGSON B. (1992) Pinpointing clusters of apparently sporadic cases of Legionnaires' disease British Medical Journal 304 (6833), pp. 1022-1027. http pdf
2. DUNN C. E., BHOPAL R. S., COCKINGS S., WALKER D., ROWLINGSON B. & DIGGLE P. (2007) Advancing insights into methods for studying environment-health relationships: a multidisciplinary approach to understanding Legionnaires' disease Health Place 13(3), pp.677-90 http
3. RUDBECK M., JEPSEN M. R., SONNE I. B., ULDUM S. A., VISKUM S. & MØLBAK K. (2010) Geographical variation of sporadic Legionnaires' disease analysed in a grid model Epidemiology and Infection 138 (1), pp. 9-14. http
4. VAN DEN WIJNGAARD C. C., VAN ASTEN L., VAN PELT W., DOORNBOS G., NAGELKERKE N. J. D., DONKER G. A, VAN DER HOEK W. & KOOPMANS M., P. G. (2010) Syndromic surveillance for local outbreaks of lower-respiratory infections: would it work? PLoS ONE 5(4): e10406 http pdf
5. CHE D., DECLUDT B., CAMPESE C. & DESCENCLOS J. C. (2003) Sporadic cases of community acquired legionnaires' disease: an ecological study to identify new sources of contamination Journal of Epidemiology and Community Health 57, pp.466-469 http pdf
6. EUSER S. M., PELGRIM M. & DEN BOER J. W. (2010) Legionnaires' disease and Pontiac fever after using a private outdoor whirlpool spa Scandinavian Journal of Infectious Diseases 42(11-12), pp. 910-916. http