Hierarchical Spatial Regression Models for Change Point Analysis

• Activity Number: 291
• Type: Topic Contributed
• Date/Time: Tuesday, August 2, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Chae Young Lim*+ and Taps Maiti and Sarat C. Dass
• Companies: Michigan State University and Michigan State University and Michigan State University
• Keywords: Joinpoint analysis ; Disease mapping ; Bayesian nonparametrics ; Dirichlet process priors

Abstract

It is well known that cancer incidence rates are heterogeneous across geographical regions. The aim of this paper is to supplement the existing tools for analyzing cancer rates from the Surveillance, Epidemiology, and End Results (SEER) database that are able to find the change points over time. Subsequently, the model can cluster the geographical subregions based on the magnitude and direction of changes of the disease risk. The proposed model to find change-points over time and cluster spatial locations is based on Dirichlet process priors where we consider temporal functions as the random quantities arising from the Dirichlet process prior. Through the analysis of age adjusted lung cancer mortality rates from 1969 to 2006, the proposed model nicely characterized local data features, namely, the local change points, the rate of changes, and clusters of states that exhibited similar trends of cancer incidence rates. The procedure to extend this model to include covariates therefore enabling selection of meaningful covariates are also discussed.