Designs and analyses for exploring the relationship of magnetic fields to childhood leukaemia: a pilot project for the Danish National Birth Cohort.

Pooled analyses have consistently displayed an association of childhood leukaemia with residential magnetic fields, even after attempting to account for study problems. In light of the serious possibility that the link is indeed causal, a study of possible joint effects (interactions) of magnetic fields and genetic cofactors on childhood leukaemia may be justifiable. Such a study would face serious obstacles of limited numbers for subgroup analysis. To address these obstacles, we describe a design and analysis strategy that combines multiphase (multistage) sampling, measurement error modelling, and Bayesian methods for subgroup analysis that incorporate information from earlier pooled analyses. Special attention is given to prior specification, which would be the potentially controversial element. The approach could be pilot tested on data from the Danish National Birth Cohort, but an informative study would require augmenting these data with case-control sampling. We conclude that the approach outlined may be of value not only for this topic but also in other studies of effects of rare exposures and genetic factors on rare outcomes.