|
Abstract:
|
Age-period-cohort models decompose trends in population rates into age, calendar-period, and birth-cohort effects. Methodology to fit parsimonious models to data in geographically-organized regions (e.g., US states) is not well developed. Here, we present a general method for modeling trends in geographically-organized regions. We allow region-specific parameters to be correlated spatially (e.g., among neighboring states), and to one another (e.g., between baseline risk and calendar-period trend), via a random-effects formulation using a generalized multivariate conditionally auto-regressive prior, implemented using a Gibbs sampler in JAGS. We apply our approach to US state-level mortality in young (aged 25-50) white non-Hispanic men and women to assess the impact of fatal drug overdoses (OD) on total mortality. We show that OD fully accounts for rising mortality among women; and that increasing mortality is positively correlated with baseline risk in men and women, suggesting that the disparities between states have grown over time. Our model parsimoniously accounts for spatial heterogeneity in model parameters, providing reliable inference in large national databases.
|
