Numbers of observed deaths were closer to those expected after known risk factors were allowed for

Editor—Busby and Cato report a significant excess of deaths from childhood leukaemia during 1981–95 in the county districts of Newbury and south Oxfordshire and suggest that this might be linked to discharges of radionuclides from nuclear installations in the vicinity.1 This study has two major limitations: the fact that survival from childhood leukaemia during the period studied was substantial2 and the authors' failure to take into account any of the known risk factors for childhood leukaemia. In particular, the incidence of acute lymphoblastic leukaemia (which accounted for 64% of childhood deaths from leukaemia in the period considered) is higher in county districts that are rural and have a higher socioeconomic status and a higher level of child migration.3 4

Using the survival rates presented by Stiller,2 we estimate that Busby and Cato are likely to have excluded from their study about two thirds of the cases that were diagnosed during 1981-95. While the incidence of leukaemia may reflect environmental factors, the death rate is likely to depend on genetic and clinical factors.5

Despite the difficulty of interpreting a study of childhood leukaemia based on deaths rather than incidence, we investigated the possible effect of known risk factors. Draper et al reported the incidence of childhood lymphocytic and unspecified leukaemia in county districts classified by urban or rural status and socioeconomic score.3 We derived these indicators from 1981 census data for each of the county districts of interest and recalculated the expected number of deaths and the corresponding Poisson probability, allowing for these factors. We assumed that the risk of non-lymphocytic leukaemia and the survival of patients with leukaemia did not vary between county districts. We carried out a similar calculation using incidences of acute lymphoblastic leukaemia in county districts classified by level of child …