General Multiple Mediation Analysis with Application to Explore the Racial Disparity in Breast Cancer Survival

• Activity Number: 187
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Ying Fan*+ and Qingzhao Yu and Xiao Cheng Wu
• Companies: Louisiana State University Health Sciences Center and Louisiana State University Health Sciences Center and Louisiana State University Health Sciences Center
• Keywords: direct effect ; indirect effect ; mediation analysis ; racial disparity

Abstract

We propose general definitions of mediation effects with counterfactual outcome context and provide statistical inference methods in different predictive models. The proposed method accommodates various types of treatment variables. We demonstrate the relationship between our definition of direct or indirect effect and natural direct or indirect effect for binary mediator case. Our method is appealing in that multiple mediators of different types can be considered simultaneously. Moreover, the individual indirect effect from each mediator can be identified. This property enables researchers to efficiently manipulate the relationship between exposure and response variables through the most influential interventional variables. Furthermore, we extend the indirect effect estimation to general predictive models in which no specific model form has to be assumed. Two approaches to estimate the variance of indirect effect are investigated and compared. The proposed method is demonstrated by simulations and a real example that examines the mechanism of racial disparity in female breast cancer three-year mortality in Louisiana State.