Cluster investigations: are they worth it?

The outcomes of systematic investigation of such disease clusters rarely see the light of day in peer-reviewed journals. In that regard the report in this issue of the Journal by Westley-Wise and her colleagues of an investigation into a cluster of leukaemia cases among residents near a major steelworks is unusual.¹ What is not unusual, however, is the report's failure to produce a persuasive explanation for the observed cluster, although it usefully reports some ambient benzene levels (levels that are orders of magnitude lower than those known to produce health effects and similarly lower than levels inhaled in cigarette smoke).

Of the many studies investigating close case aggregations of any type, only a few have come up with a credible explanation of why the disease cluster exists or have added to knowledge of the causes of disease. Such exceptions would include, for example, the links found between angiosarcoma of the liver and vinyl chloride monomer exposure,²or mercury poisoning in Minamata, Japan.³ These instances demonstrated strong associations that hardly required the sophisticated statistical tools of modern epidemiology to be identified as "possible" causal links.

Sadly, experience tells us that almost all cluster investigations will fail to produce insights into environmental disease interactions, and no amount of time, effort and money will change this. This is for a number of reasons. Clusters are by definition based on tiny case numbers. As such, they may not represent the "typical" disease and are not amenable to epidemiological analyses. Further, most reported clusters tend to be of conditions where little or nothing is known of the common causes. There are often controversial theories relating to the condition (eg, the hypothesis that proximity to powerlines can cause cancer). Finally, the supposed environmental-disease links are often impossible to investigate satisfactorily because of the lapse of time or because it is simply not clear what to investigate. The tenfold excess of childhood cancers in the United Kingdom near the Sellafield reprocessing plant⁴ has been under investigation for a decade, to no avail.

Why then do we continue to spend time investigating these phenomena? Why indeed does every state health department in the USA have its own "cluster investigation protocol"? And why did my colleagues and I produce a similar protocol in the UK in 1997?⁵

Most public health doctors know that having a disease cluster to investigate is a thankless task. Not to have an investigation at all would be regarded with deep suspicion, while investigations with negative results can lead to accusations of a cover-up. On balance, it is better to hold an investigation than not. So the optimistic epidemiologist, hoping that unexpected insights will drop out of the investigation and reveal a truly causal association, adopts a careful protocol to maximise the chances of success. The pessimist feels that at least a sound methodological approach will guard against the reproaches that may follow a negative finding.

All the professionals involved are aware that a bad cluster investigation generates considerable disquiet among the public and tends to give them a bad press. Most therefore take the process to be as important as the outcome -- hence the various guidelines.

Guides to cluster investigation tend to have a common theme of transparency of action and a candid approach to the concerned parties. They usually suggest various strategies for investigation, one of which is to take the local concern about the cause of the disease cluster as a hypothesis to test. Indeed, testing this hypothesis is often the best "outcome" of any investigation. Westley-Wise et al tested the hypothesis that the leukaemia cluster was caused by pollution from the neighbouring coke-making facilities.¹ Their negative findings, if they succeed in allaying local concerns about this point, will be something of a positive outcome.

It has to be said that clusters can and do occur fortuitously. It is possible to calculate how often chance events play a part in clustering. This explanation for a cluster, however, is the one least likely to have credibility with those closely concerned. Perhaps the only sensible way forward is to address the underlying issue that usually starts a cluster investigation -- what are the significant causes of a condition? That is best tackled, not by a cluster investigation, but by the combined efforts of laboratory-based sciences and large epidemiological investigations -- usually cohort or case-control studies. This type of effort has been made (internationally) to elucidate the possible causes of childhood leukaemia (a frequent target of cluster investigation). The next year or so will see the assembly of combined and very large datasets that might tell us why this condition appears to "cluster". Possible small risks (such as powerlines) can be found or discounted, and more plausible, but complex, biological phenomena relating to infection in early life can be properly investigated. The latter will be the focus of much attention.

Ray A Cartwright
Director, Leukaemia Research Fund Centre for Clinical Epidemiology
University of Leeds, UK

1. Westley-Wise VJ, Stewart BW, Kreis I, et al. Investigation of a cluster of leukaemia in the Illawarra region of New South Wales, 1989-1996. Med J Aust 1999; 171: 178-183.
2. Baxter PJ, Anthony PP. Angiosarcoma of the liver in Great Britain, 1963-73. BMJ 1977; 2: 919-921.
3. Tsuchiqa K. The discovery of the causal agent of Minamata disease. Am J Indust Med 1992; 21: 275-280.
4. Draper GJ, Stiller C, Cartwright RA, et al. Cancer in Cumbria and in the vicinity of the Sellafield nuclear installation 1963-1990. BMJ 1993; 306: 89-94.
5. Arrundale J, Bain M, Botting B, et al. Handbook and guide to the investigation of clusters of diseases. London: Leukaemia Research Fund, 1997.