© 1990 Oxford University Press
research-article |
‘Clustering’—Real or Apparent?: Probability Maps of Childhood Cancer in the West Midlands Health Authority Region
* West Midlands Regional Children's Tumour Registry. The Children's Hospital Birmingham, B16 8ET, UK
** Departments of Oncology, The Children's Hospital Birmingham, UK
Departments of Histopathology, The Children's Hospital Birmingham, UK
Departments of Haematology, The Children's Hospital Birmingham, UK
§School of Geography, University of Birmingham Birmingham, UK
¶Statistics Division, West Midlands Regional Health Authority Birmingham, UK
Dr K R Muir, WMRCTR, Dept of Oncology, The Children's Hospital, Ladywood Middleway, Birmingham, B16 8ET, United Kingdom.
The West Midlands Regional Children's Tumour Registry collects detailed information on all cases of childhood cancer in the West Midlands Health Authority Region (WMHAR). The distribution by electoral ward of all cases diagnosed in the WMHAR between 1980 and 1984 has been determined. Analysis has also been performed for leukaemias/non-Hodgkin's lymphomas alone. We suggest that this latter grouping should be universally employed, owing to the difficulty of accurately separating out cases of leukaemia.
Both spatial analyses showed several wards with significantly excessive rates on the basis of their cumulative Poisson probability. Observed/expected ratios of 3–35 were seen for cases in significant wards, which are similar to the ratios seen in analasis of incidence around nuclear installations. However, further detailed consideration of the these individual significance levels in the light of the number of statistically significant wards which would occur by chance alone, due to the multiple use of the test, accounted completely for the number of wards obtained in each of the groups considered. Thus, apparent ‘clustering’ of cases could be mere statistical artefact.
In the WMHAR, therefore, using the technique of probability mapping, no true spatial pattern of incidence was found, other than that which would occur by chance alone. This, in a large area without nuclear installations and an even mix of rural and industralised regions, could be seen as control data for those studies which have considered cases of childhood leukaemia around nuclear facilities, where the observation of single point clusters associated with suspected sites restricts assessments of spatial pattern in the rest of the area.
Consideration is given to the problems of using probability mapping alone in analysis of data such as ours, which highlights the low power and inherent limitations of the technique. We conclude that results obtained using this method need to be supplemented using further tests and suggestions are made as to useful future developments.
Received 1 June 1990