Modeling Dependent Covariate Subclass Eeects in Bayesian Meta-analysis

Cancers of the breast and endometrium contribute substantially to mortality among women. Studies examining putative links between these cancers and estrogen have reported connicting results. The model presented here provides a natural framework for exploring the eeect of dependent covariate subclasses in explaining the variation in the response. Such subclass eeects can be investigated across several covariates simultaneously. New meta-analyses are performed examining possible links between estrogen exposure and cancers of the breast and endometrium. Markov chain Monte Carlo methods are used to generate posterior samples from parameters of interest. Tables and plots of the results are provided on a covariate-by-covariate basis.