Empirical Bayes estimation of cancer mortality rates

Mortality rates for specific cancer types and age-sex groups computed for large numbers of cities typically show extreme fluctuation. This is primarily due to the rare occurrence of specific cancer deaths in most of the small and moderate size cities during a fixed time period. Assuming a Poisson death process, we use an empirical Bayes method to obtain adjusted rates that are more stable for comparison of cities and prediction of future mortality. We have chosen stomach and bladder cancers in Missouri cities to illustrate the problems, techniques and results.