Reliability-adjusted disease maps

Bayesian methods for adjusting mortality and morbidity rates to account for variations caused by small numbers are presented. Although such methods produce statistically biased morbidity and mortality rate estimates, these approaches are superior for any applications that depend on a relative ordering of a set of rates because the total error of prediction for the maps taken as a whole is smaller. This approach is especially relevant for identifying cancer ‘hot spots’ for a set of geographic areas. The theory and usefulness of making such adjustments for geographic data sets are described and an example presented, comparing classical and Bayesian rate estimation methods for rank ordering female breast cancer data in the San Francisco-Oakland SMSA.