A Geostatistical Approach to Disease Mapping for Aggregated Count Data Using GLMMs

• Activity Number: 249
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Lauren Hund*+ and Brent A. Coull
• Companies: Harvard University and Harvard School of Public Health
• Keywords: generalized linear mixed model (GLMM ; disease mapping ; spatial epidemiology ; additive models ; geostatistics ; small area estimation

Abstract

Aggregate count data arises frequently in disease mapping and is often analyzed using models with conditional MRF priors to account for spatial correlation. We construct a geostatistical model for aggregate count data within the framework of a GLMM using radial splines by assuming an underlying continuous risk surface induces spatial correlation between areas. This framework facilitates fast, easy model fitting with PQL for maximum likelihood estimation. Through simulation, we verify that bias in fixed effects is negligible. Using Boston census data, we assess the relationship between poverty and premature mortality using the GLMM model and a Bayesian model with a CAR prior for spatial correlation; the models give very similar results. Our model is both flexible and computationally fast, and will be useful for boundary misalignment issues and small area estimation in large data sets.