A Bayesian approach for estimating age-adjusted rates for low-prevalence diseases over space and time.

Age-adjusted rates are frequently used by epidemiologists to compare disease incidence and mortality across populations. In small geographic regions, age-adjusted rates computed directly from the data are subject to considerable variability and are generally unreliable. Therefore, we desire an approach that accounts for the excessive number of zero counts in disease mapping datasets, which are naturally present for low-prevalence diseases and are further innated when stratifying by age group. Bayesian modeling approaches are naturally suited to employ spatial and temporal smoothing to produce more stable estimates of age-adjusted rates for small areas. We propose a Bayesian hierarchical spatio-temporal hurdle model for counts and demonstrate how age-adjusted rates can be estimated from the hurdle model. We perform a simulation study to evaluate the performance of the proposed model vs a traditional Poisson model on datasets with varying characteristics. The approach is illustrated using two applications to cancer mortality at the county level.