Identification of Causal Pathways in Studies with a Large Number of Mediators

• Activity Number: 648
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Andriy Derkach* and Joshua Sampson
• Companies: National Cancer Institute and National Cancer Institute
• Keywords: mediation ; multiple testing ; laten variables ; sparse model

Abstract

Modern biomedical and epidemiological studies often measures large number of biomarkers such as gene expressions and metabolite levels that usually play a mediation role between an exposure and an outcome. Typical mediation analysis or causal inference offers numerous methods for testing if a single variable mediates the relationship between a known exposure and outcome). Methods to simultaneously assess multiple mediators have received limited attention. Here we consider statistical methods to fully model a biological pathway between an exposure and disease. We compare methods that build a sparse model of biomarkers that link exposure to disease. The methods can accommodate non-Gaussian distributions of variables, include latent variables and allow the number of variables to be greater than the number individuals. We show compare properties between methods through theory and simulation.