Estimating Cancer Mortality Rates Using Bayesian Area-Age-Period-Cohort (AAPC) Model and Multistage Carcinogenesis Models

• Activity Number: 479
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Zhiheng Xu*+ and Vicki S. Hertzberg
• Companies: Emory University and Emory University
• Keywords: AAPC ; carcinogenesis ; DIC

Abstract

AAPC models have been widely used as a general framework to study the temporal and spatial pattern of the risk of disease. It is well known that there is a non-identifiability problem associated with the age, period and cohort effects because of the exact linear relationship among them. Bayesian procedures have been implemented to handle the identifiability issues in the APC models through prior setting and model constraints. However, some constraints lack of a strong biological explanation. In this study, we plan to introduce the multistage carcinogenesis models into AAPC model to incorporate more biological meaning to the age effects. The age effect in the AAPC model will be replaced by the hazard function derived from carcinogenesis model and new priors will be assigned on the carcinogenesis model. Spatial correlation will be taken into account in the AAPC model. Colon cancer death rate in IOWA will be used in this study and we will apply spatial effect to the county level. Deviance Information Criteria (DIC) will be used to compare different models in fitting cancer mortality data.