Dr. Mohsen Rezaeian (PhD, Epidemiologist, Associate Professor)
Social Medicine Department, Rafsanjan Medical School, Rafsanjan, Iran.
Tel: +98 391 5234003
Fax: +98 391 5225209
Email: moeygmr2@yahoo.co.uk

The examination of disease clustering has become a flourishing area in medical research during recent decades⁽¹⁾. Generally speaking, the search for disease clusters is one of the branches of geographical epidemiology⁽²⁾. The term cluster usually refers to "uncommon diseases of non-infectious origin (e.g., leukaemia, spontaneous abortion, suicides), which are often perceived to be due to environmental exposures"⁽³⁾.

Clusters of health events are often reported to health authorities especially within developed countries⁽⁴⁾. Although evidence suggests that only a small fraction of such reports are likely to lead to the identification of a real disease cluster⁽⁵⁾, health authorities should investigates cautiously such reported clusters further for two obvious reasons: Firstly, time and space clustering may suggest that there would be some social, economic, cultural, etc. predisposing factors, which affect the occurrence of disease. Secondly, it can also guide an appropriate response for relieving communities from the fear of a perceived or real disease cluster.

Given the importance of cluster studies in both developed and developing countries, the aim of the present article is to discuss some of the most important fundamental issues surrounding this concept for the public health specialists especially within the Middle East region. To fulfil this demand, the article begins with the existing definitions of cluster, and then it moves to discuss different types of cluster and their related statistical issues, and also diverse scenarios in cluster investigations. Finally, it ends by providing some guidelines for dealing with a cluster more appropriately.

There are different definitions of cluster. For instance, Knox (1989) defined a cluster as: "a geographically bounded group of occurrences of sufficient size and concentration to be unlikely to have occurred by chance⁽⁶⁾." Whilst spatial cluster is the main focus of this definition, Last (1995) tries to define this term more generally to include both temporal and spatial aspects of a cluster. He defined a cluster as: "aggregation of relatively uncommon events or diseases in space and/or time in amounts that are believed or perceived to be greater than could be expected by chance⁽⁷⁾."

Similarly, Gerstman (1998) defined cluster as: "A close grouping of disease or disease-related events in space, time, or both space and time and is usually reserved to describe the aggregation of rare diseases such as specific forms of cancer⁽⁸⁾." Rothman (1990) also states that: "Clustering in both space and time, or space-time clustering means that the incidence rates are temporarily higher in some places than in others, with the places that have a high incidence rate changing with time⁽⁹⁾." Furthermore, the definition which refers to it earlier in the first paragraph of the introduction, focuses on the non-infectious origin of a cluster⁽³⁾.

Different types of cluster and their related statistical issues

It would be possible to classify types of clustering studies into general and specific. General or non-specific clustering is the analysis of the overall clustering tendency of the disease incidence in a study region. It should be noted that the investigators of the general clustering do not seek to determine the exact locations of clusters but simply to assess if clustering is noticeable in the study region. On the contrary, specific studies are designed to determine the precise location of the clusters⁽¹⁰⁾.

During recent years there has been a rapid expansion in the number of statistical tests for detecting of both general and specific diseases clusters. These tests have become more specially designed to encompass the particular disease-environment interactions^(11&12). However, caution is required in the application of such tests in order to avoid 'false positives' results i.e. detecting an unreal cluster as a real one⁽¹³⁾.

The types of statistical tests for both kinds of disease clusters are dependent on the types of data which might include point and area data⁽¹⁾. Each item of health data, such as population or environmental exposure, may be connected with a point e.g. a home or an area e.g. a district⁽¹⁴⁾.

General cluster in area data implies that given an event e.g. suicide, the rates of it within neighbouring areas are likely to be more similar than those in distant ones^(15&16). In such situations detecting a cluster is accomplished by the use of spatial autocorrelation statistics⁽¹⁷⁾. The two most commonly used spatial autocorrelation statistics for detecting general clustering in area data are the I statistic, developed by Moran⁽¹⁸⁾ and Geary's c statistic⁽¹⁹⁾.

On the contrary, specific clusters in area data search for local clustering e.g. hot spots of high or low values by finding any association between a value at a specific area and values of neighbouring areas⁽¹⁾. There is also a number of spatial autocorrelation statistics available e.g. Getis and Ord's G* statistic for detecting such specific clusters^(20&21).

Tests for the clusters detection in point format data are more frequent than those for area data⁽¹⁾. To name a few, Cuzick and Edwards'⁽²²⁾ method examines the k nearest neighbours of each case in order to determine global clustering. The geographical analysis machine⁽²³⁾ and the spatial scan statistic⁽²⁴⁾ also try to detect the localised clustering by drawing predefined circles over the area of study and compare the risk of disease inside and outside of each circle.

Diverse scenarios in cluster investigations

It would also be possible to categorise diverse scenarios in cluster investigations into three situations as follows⁽²⁵⁾:

Within the first scenario no clustering has been already detected within the population under study. Therefore, the question of whether or not a cluster is occurring is being approached a priori.

The second scenario is similar to the first in that nothing is known about the occurence of clusters in the population. However, there is a specific hypothesis to be investigated e.g. leukaemia risk is associated withcloseness to a nuclear power plant.

In the third scenario, a disease cluster has already been detected within the population under study. Therefore, a posteriori or post hoc approach is selected in order to determine the realness of the cluster and/or to provide an explanation for it.

One should bear in mind that the problems of interpretation of each cluster, is crucially dependent on its scenario. For instance, it can be only possible to infer the conventional P value in relation to a priori hypotheses⁽⁵⁾.

How to deal with a cluster?

In order to appropriately respond to the reports of the clusters a comprehensive approach is needed. For instance, the recommended approach by the US Centers for Disease Control and Preventions (CDC) consists of a four-stage process, which includes: primary response, evaluation, major feasibility study, and etiologic study. It should be noted that each step provides opportunities for collecting data and making informative decisions in order to stop or carry on the investigation⁽⁴⁾.

It is also suggested that to implement such comprehensive approach successfully each health authority should have an interior management system. Such a system involves the establishment of a central point of responsibility and control. Furthermore, written working procedures and devoted resources might have immense value⁽⁴⁾.

The investigations of suspected disease clusters due to environmental exposures are often originated in response to public anxiety within developed countries^(4&5). This makes public health specialists within such countries examine the alleged clusters from different perspectives in order to prepare a more appropriate plan for dealing with these events and their perceived risks among the community^(26&27).

Given the recent environmental changes within developing countries including countries within the Middle East region, it seems that public health specialists in these countries also have to take the issue of cluster investigation more seriously. They should be aware that when such investigations become informative, that the following criteria are met: "chemical exposures are documented, routes of human exposure are traced, sub-populations at highest risk are identified, reliable denominator data are available, the diagnosis of the outcome has been consistent over time, and specific health outcomes are studied⁽⁸⁾."

The author would like to appreciate the valuable comments of Ian Enzer on the earlier draft of this paper.

1. Rezaeian, M. Dunn, G. St. Leger, S. Appleby L. Geographical epidemiology, spatial analysis and geographical information systems: a multidisciplinary glossary. J Epidemiol Community Health 2007; 61 : 98-102.
   
2. Lawson, AB. Bohning, D. Biggeri, A. et al. eds. Disease mapping and its uses. Disease mapping and risk assessment for public health. Chichester: Wiley, 1999.
   
3. California Dept of Health Services. Investigating possible Non-infectious Disease Clusters. Berkeley, CA: Environmental Epidemiology and Toxicology Branch. California Department of Health Services, 1989.
   
4. CDC. Guidelines for investigating clusters of health events. MMWR. Morb Mortal Wkly. 1990; 27:39(RR-11):1-23. Atlanta.
   
5. Olsen, FS. Martuzzi, M. Elliot, P. Cluster analysis and disease mapping- Why, when, and how? A step by step guide. BMJ 1996; 313:863-6.
   
6. Knox, EG. Detection of clusters. In: Elliott P, ed. Methodology of enquiries into disease clustering. London: Small Area Health Statistics Unit, 17-20, 1989.
   
7. Last JM. A Dictionary of Epidemiology. Oxford University Press. New York, 1995.
   
8. Gerstman, BB. Epidemiology Kept Simple. An Introduction to Classic and Modern Epidemiology. Willey-Liss. USA, 1998.
   
9. Rothman, KJ. A sobering start for the cluster busters' conference. Am J Epidemiol 1990;132(suppl):S6-S13.
   
10. Besag, J. Newell, J. The detection of clusters in rare disease. J R Stat Soc A 154: 143-155.
    
11. Thomas, R. Geomedical systems, Interventions and Control. Routledge, London, 1992.
    
12. Alexander, FE. Boyle, P. Methods for Investigating Localised Clustering of Disease. International Agency for Research Cancer. Lyon, France, 1996.
    
13. Elliott, P. Martuzzi, M. Shaddick, G. Spatial methods in environment Epidemiology: a critique. Stat Methods Med Res 1995; 4: 137-159.
    
14. Elliott, P. Wartenberg, D. Spatial epidemiology: current approaches and future challenges. Environ Health Perspect 2004;112:998-1006.
    
15. Jerrett, M. Burnett, RT. Goldberg, MS. et al. Spatial analysis for environmental health research: concepts, methods, and examples. J Toxicol Environ Health A 2003;66:1783-810.
    
16. Tobler, WR. A computer movie simulating urban growth in the Detroit region. Econ Geogr 1970;46(Suppl):234-40.
    
17. Odland, J. Spatial autocorrelation. California: Sage Publications, 1988.
    
18. Moran, PAP. The interpretation of statistical maps. J R Stat Soc B 1948;10:243-51.
    
19. Geary, RC. The contiguity ratio and statistical mapping. Incorporated Stat 1954;5:115-45.
    
20. Cromley, EK. McLafferty, SL. GIS and public health. New York: The Guilford Press, 2002.
    
21. Gatrell, AC. Bailly, TC. Diggle, PJ. et al. Spatial point pattern analysis and its application in geographical epidemiology. Trans Inst Br Geogr, 1996;21:256-74.
    
22. Cuzick, J. Edwards, R. Spatial clustering for inhomogeneous populations. J R Stat Soc B 1990;52:73-104.
    
23. Openshaw, S. Charlton, M. Wymer, C. et al. A mark 1 geographical analysis machine for the automated analysis of point data sets. Int J Geogr Inf Syst 1987;1:335-58.
    
24. Kulldorff, M. Statistical methods for spatial epidemiology: tests for randomness. In: Gatrell A, Loytonen M, eds. GIS and health. London: Taylor & Francis, 1998:49-62.
    
25. MacMahon, B. Trichopoulos, D. Epidemiology Principle and Method. Little, Brown and Company. USA, 1996.
    
26. Trumbo, CW. McComas, KA. Besley, JC. Individual- and community-level effects on risk perception in cancer cluster investigations. Risk Anal 2008; 28:161-178.
    
27. Trumbo, CW. McComas, KA. Institutional trust, information processing and perception of environmental cancer risk. Int J Glob Environ Issues 2008; 8:61-76.