Apparent association between benzene and childhood leukaemia: methodological doubts concerning a report by Knox

A recent study by Knox concludes that cases and "clusters" of two or more cases of childhood leukaemia and non-Hodgkin's lymphoma occur closer to many kinds of industrial installation than to supposedly comparable control locations. It is argued that these findings could be largely or entirely artefactual, the apparent differences arising out of the inappropriateness of the control data. Knox used randomly selected postcode units as controls, a procedure that leads to the comparison of individuals located in areas with typically quite different population densities from those for the cases. The resulting potential for bias is explored and the arguments are exemplified by analysing household data based on postcodes.