Evaluation of Removable Statistical Interaction in Cancer Epidemiology Studies

• Activity Number: 357
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Jaya M. Satagopan*+ and Robert C. Elston
• Companies: Memorial Sloan-Kettering Cancer Center and Case Western Reserve University
• Keywords: multi-stage carcinogenesis process ; analysis of variance ; score statistic

Abstract

Cancer epidemiology studies have traditionally focused on investigating gene-gene and gene-environment interactions as part of the attempt to understand the role of risk factors associated with tumorigenesis or tumor-related traits. This talk focuses on evaluating the omnibus null hypothesis of no removable statistical interaction between two sets of risk factors in cancer epidemiology studies. The statistical interaction between two or more risk factors measures the change in the effect on the outcome of one risk factor when the value(s) of the other risk factor(s) is (are) altered. A statistical interaction is deemed removable if the relationship between the outcome and the risk factors can be made linear through suitable transformations so that the resulting relationship takes the form of a simple additive model. We will show how the multi-stage carcinogenesis paradigm provides insights about testing for removable statistical interactions in a parsimonious manner, and demonstrate conditions under which this approach can provide a powerful approach to test for interactions relative to an alternative standard approach.