Spatial clustering of childhood cancers in Great Britain

Study objectives - Firstly, to identify spatially close pairs and triplets of childhood leukaemias and cancers in Britain. Secondly, to compare pair frequencies with random expectations, identify excesses, and measure the diameters of any clusters. Thirdly, to infer possible causes. Design and setting - Stratified Poisson analyses of two comprehensive sets of childhood leukaemia and cancer data in Great Britain: seeking home address pairs within particular census enumeration districts (EDs) or postcodes (PCs). Numbers of pairs/triplets of leukaemia registrations sharing single or adjacent EDs were compared with Poisson expectations in national ED strata with different numbers of households (HHs). Pairs/triplets of leukaemia/cancer deaths (and births) sharing a single PC were compared with Poisson expectations in national PC strata with different numbers of postal delivery points (DPs). Same and adjacent house pairs were identified individually among the same PC death pairs. Areal case densities were estimated around a sample of index cases, using their own PC grid coordinates, and those recorded in the central PC directory. Participants - These comprised, firstly, all cases of childhood leukaemia and non-Hodgkin lymphoma registered between 1966 and 1983 in England and Wales (ED analysis) and, secondly, all childhood leukaemia and cancer deaths between 1953 and 1980, in England, Wales, and Scotland (PC analyses at birth and at death). Main results - Short range spatial clustering was demonstrated (a) far leukaemia at place of registration, and (b) for leukaemia and cancer (separately and jointly) at both birth and death addresses. There was evidence of additional case pairing in adjacent PCs. Both data sets showed a relative local pair excess of about 1.5, within diameters of 300 metres. Secondary case densities, measured within 600 metres of a sample of unpaired index cases, were raised by the same ratio. The raised risk then tapered with increasing distance to about 3 km. Forty-four non-twin pairs had died at exactly the same address, far in excess of random expectation. This same house excess was due entirely to 31 sibling pairs. They also showed a relative excess of central nervous system and other solid tumours; but without the exact tumour type sibling concordances sometimes seen in MZ twins. The sibling pairs were only a small part of the overall excess of same PC pairs. Conclusions - Short range geographical clustering probably reflects two separate causes of childhood cancer, namely (a) an uncommon familial susceptibility to solid cancers, probably inherited, and (b) a group of long standing focal environmental hazards, most effective within a few hundred metres of the source, but detectable as far as 3 km.