|
Activity Number:
|
262
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #305619 |
|
Title:
|
Semiparametric Bayesian Approach to Prediction of Cancer Mortality Counts in the Presence of Spatial Information
|
|
Author(s):
|
Kaushik Ghosh*+ and Pulak Ghosh and Ram C. Tiwari
|
|
Companies:
|
University of Nevada Las Vegas and Novartis Pharmaceuticals and FDA
|
|
Address:
|
Box 454020, Las Vegas, NV, 89154-4020,
|
|
Keywords:
|
Dirichlet process ; Cancer mortality ; Cancer incidence ; prediction
|
|
Abstract:
|
We propose a nonparametric Bayesian model for jointly predicting the cancer mortality counts at several states and cancer sites based on past observed data. The proposed method uses a multivariate generalization of the local quadratic model developed in Ghosh and Tiwari (2007) with a hierarchical Dirichlet process (HDP) prior for the distribution of the vector of "acceleration" parameters. The inherent two-way clustering property of HDP allows automatic sharing of information among "similar" states as well as years, producing improved estimates compared to when one models each state individually. We illustrate the proposed method with mortality data on selected cancer sites obtained from states in the western United States during the period 1969--2003.
|
