A Bayesian Dynamic Spatio-Temporal Interaction Model: An Application to Prostate Cancer Incidence

During the past three decades, prostate cancer incidence has changed substantially in the United States. A fully Bayesian hierarchical spatio-temporal interaction model is proposed to estimate prostate cancer incidence rates in the state of Iowa. We introduce random spatial effects to capture the local dependence among regions, random temporal effects to explain the nonlinearity of rates over time, and random spatio-temporal interactions. In addition, we introduce fixed age effects because most epidemiologic data are strongly related to age. We find that prostate cancer incidence in Iowa counties increased sharply over age while incidence rates increased initially, then decreased over time. We identify hot spots of high and low rates for age groups and time periods using disease mapping.