Mediation Analysis in the Presence of Partially Missing Data on the Mediator in Genomic Studies

• Activity Number: 359
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Richard Barfield* and Xihong Lin
• Companies: and Harvard School of Public Health
• Keywords: Genetics ; Mediation ; Methylation ; Missing Data ; Causal

Abstract

Mediation analysis is a useful framework to understand whether an exposure affects a disease or a trait through a mediator, such as DNA methylation or expression. Genomic studies involving DNA methylation and/or expression are usually part of a larger study, where genomic data is only collected on a subset of study participants. Mediation analysis performed just on the individuals with complete data (both exposure and the genomic data) fails to incorporate the information from individuals with partial data. We propose a maximum likelihood based mediation analysis method that uses all available data by leveraging the information from individuals with "missing" mediator data. Under standard assumptions for mediation analysis and an additional assumption that the missing data mechanism is ignorable, we estimate the causal direct and indirect effects using all individuals in the study. Simulation results show that under missing completely at random, the proposed MLEs using all the data gained more efficient estimators of the direct and indirect effects and more power to detect their effects compared to complete data analysis, as the amount of individuals with missing data increased.