Pesticides and cancer clusters
- is there a relationship?

Pesticides are used because they are toxic to living systems: it is reasonable, therefore, to be concerned about possible adverse consequences for human health. Such issues may be studied from several perspectives, including biological, toxicological and epidemiological- Dr. Kenneth Muir contributes to the debate on the role that epidemiology may play in identifying the impact on exposed populations.

Epidemiology studies are potentially the most relevant of these perspectives, since this subject specifically concerns itself with the study of the causes of disease in human populations. The ethical limitations imposed by working with human beings, however, restrict the nature of the investigations that can be undertaken. Randomised intervention studies, most akin to the experiments that would be carried out in a laboratory situation, are simply not possible when studying the potentially toxicological effects of chemicals. The literature dealing with individual pesticides and their possible associations with cancer is extensive. Moses(1) reviews studies on a number of cancers linked with use of, or exposure to, particular pesticides or groups of pesticides, including malignant lymphoma, leukaemia, multiple myeloma, testicular cancer, cancer of the gastro-intestinal tract, lung cancer and brain cancer. A recent study of the causes of breast cancer in women found that as compared with control groups, women with malignant breast cancer had higher levels of metabolites of DDT in their fatty tissues(2).
    This paper will concentrate on one particular type of descriptive investigation, that of mapping studies.

Mapping disease
Mapping-who gets what disease and where-has a long history, but has recently received considerable impetus following the investigation of supposed clusters of childhood cancers, around firstly the Sellafield reprocessing plant in Cumbria and subsequently other sources of radiation. Since the Sellafield story, the number of allegations concerning cancer clusters has increased dramatically. This partly reflects a heightened awareness of environmental issues, whilst at the same time, no doubt, it manifests lay anxiety and fear about cancer itself. It is interesting to observe that local environmental concerns vary between different regions and that they tend to focus on some readily identifiable landmark. This is only natural, as it is human nature to query possible associations between events and circumstances.
    Given these observations, it is understandable that the people of Lincolnshire recently alleged there may be high rates of breast and childhood brain cancers in their region, which might be related to pesticide usage. In this context, we chose to have a 'first look' to see whether the speculation could be substantiated.
    The starting point for any mapping study lies in the identification of a complete and accurate population-cased set of cancer cases. One result of mapping investigations at Sellafield was a closer examination of the quality of Regional Cancer Registry data as a basis for undertaking analyses.

Market day at Boston, Lincolnshire: the surrounding area is a high-spray location for fruit and vegetable production

Identifying a cluster
Much discussion has centred around the definition of clusters of disease. Suffice to say that it is not possible uniquely to define it. Instead, studies have attempted to identify areas of high and low incidence relative to some norm. The norm that is chosen is usually that of the regional or national cancer rate. It must be borne in mind when using UK baseline data, that the UK has one of the highest breast cancer rates in the world.
    When assessing possible clusters of a rare condition such as childhood cancer, the quality of local data is critical to enable accurate systematic investigation for any 'hot spots'. The simplest method of checking the quality of data in a given geographical area is to compare the overall incidence in that area with that shown in national or other regional data. This invokes a catch-22 situation in that potential problems in identifying all cases in one region may apply universally. This problem is particularly pertinent to any mapping exercise of paediatric brain tumours as some surgically treated or untreated cases may not be included in routine data collection procedures. This serves to underline our first rider in this particular investigation into regional patterns: that we should query the completeness of the original data.
    Assuming that we have an accurate datasheet, we can embark on a statistical assessment of whether there is high incidence of a particular tumour type. One of the main problems for any mapping technique is to allow adequately for the underlying population structure, since, on common sense grounds, we would expect more cases in areas of high population. The simplest solution is to divide the area into the smallest geographical units for which census population figures are available (electoral wards) and calculate the rate in each. This technique is known as probability mapping and is often used as a first investigation (a detailed description of the method can be found in the paper by Muir et al(3).) Another approach compares the distribution of cases to population-weighted but randomly selected control data. We have also applied one such technique in our assessment of geographical patterns(4).

Cluster research in the Trent region and Lincolnshire
Using these two approaches, we undertook an analysis of breast cancer and paediatric brain tumours, based on Regional Cancer Registry data. Owing to the rarity of the latter disease we had to look at its distribution across the whole of the Trent Regional Health Authority (which includes Lincolnshire and neighbouring counties), but since breast cancer is more common, we were able to focus the investigation on the county of Lincolnshire alone. Our analyses revealed no pattern that could reasonably be described as higher occurrence than could be expected by chance alone. Furthermore, there was no apparent relationship between disease incidence and rural areas, where a link with pesticide usage has been suspected.
    Theoretical problems including multiple testing and practical problems such as the simple definition of 'rural' persist and limit the reliability of these conclusions. We might circumvent these problems by restricting our investigation to those areas where a priori defined land classification is used and thereby limit the number of tests that hinder truly meaningful analysis(5). In other words, instead of dividing up the whole region of study into its constituent wards and then testing each ward individually, we would add together all of the wards with expected pesticide usage before undertaking the statistical testing. This would limit the number of tests performed and minimise the chance of false positive results occurring by chance alone.

Lack of data on pesticide use
To undertake this grouping of wards, it would be desirable to have a classification based on actual usage of pesticide or other chemical of interest during the same time span as covered by the disease dataset. Such information is simply not available and therefore one is constrained to use a surrogate classification. One that is readily at hand is that of land usage and, using maps, it is possible to categorise each ward as urban, agricultural or non-agricultural. If we accept that the probability of pesticides use being higher in wards defined as agricultural, then the number of tests that need to be performed is reduced to three. This pragmatic classification obviously has limitations, in that it does not identify variability in total usage, nor the individual products used, within the agricultural areas.
    This also adds weight to the argument that accurate pesticide usage figures must be recorded and made available for this sort of research. Such data would facilitate tracing inappropriate use identified during food or water residues analysis. The Ministry of Agriculture Fisheries and Food produces pesticide usage figures for agriculture, stating which active ingredients are used on what crop. This data is however extrapolated from a sample size that does not provide accurate estimates of pesticide use at the county level. Commercial data is available to the food industry, but the cost is unfortunately beyond the means of academic research. There are no figures in the public domain for pesticides used non-agriculturally. Given these criticisms, however, assuming pesticide use is higher in agricultural areas is probably the best routinely available classification and therefore the one that we have applied. The figures shown in Tables 1 and 2 for paediatric brain tumours and breast cancer respectively, are the result of the sum of the observed and expected numbers of cases for the aggregates of the wards included in the classification as described.

If one ignores the limitations imposed by using mapping studies which utilise routinely collected data and pre-defined surrogate of land type, the results offer no support for a possible association between these cancers and pesticide usage. However, bearing in mind the limitations, this analysis can only offer partial reassurance to the residents of Lincolnshire and Trent regions.

What should be done next?
Ultimately mapping the distribution of a given disease has limitations imposed by underlying population density and imprecise exposure information. A potentially more profitable way to proceed is to study the past exposure histories of individuals. This case control approach circumvents these problems by working with groups or populations of individuals. Such studies are currently being considered for both brain and breast cancer and the results will be eagerly awaited.

References
1. Moses, M., Pesticides-related health problems and farmworkers, AAOHN Journal, 1989, 37:3:115.
2. Moses, M., Pesticides and Breast Cancer, Pesticides News 21, 1993, p. 3.
3. Muir, K.R., Parkes, S.E., Mann. J.R., Stevens, M.C.G, Cameron, A.H., Raafat, F., Darbyshire, P.J., Ingram, D.R., Davis, A and Gascoinge, D.P., Clustering-real or apparent?: probability maps of childhood cancer in the West Midlands Health Authority Region, International Journal of Epidemiology, 1990, 19:853-9.
4. Cuzick, J and Edwards, R., Spatial clustering for inhomogeneous populations, Journal of the Royal Statistical Society, Series B, 1990, 2:73-104.
5. op.cit 3.

Dr. Kenneth Muir is an epidemiologist at the Department of Public Health and Epidemiology, University of Nottingham Medical School, Nottingham NG7 2UH. He was assisted in the writing of this article by his colleagues John Geoghegan (medical researcher), Darren Greenwood (statistician) and Terry Brown (Medical Researcher).

[This article first appeared in Pesticides News No. 30, December 1995, page 6]