A0622
Title: Mediation analysis to infer direct genetic effects on disease risks
Authors: Yildiz Yilmaz - Memorial University of Newfoundland (Canada) [presenting]
Brady Ryan - University of Michigan (United States)
Abstract: Many genetic associations have been identified with disease risks. However, the associations do not infer the causal genetic effects. To distinguish direct genetic effects from indirect genetic effects, a directed acyclic graph is considered to have a direct genetic effect on the primary disease occurrence phenotype and indirect effects through intermediate phenotypes, which are potentially confounded by measured and unmeasured factors. A mediation analysis method is discussed using the odds ratio scale to infer controlled direct genetic effects on disease risks while removing indirect effects through mediators and adjusting the model for measured and unmeasured confounders. The proposed method uses the estimating function methodology with robust sandwich standard errors. The method allows the inclusion of genetic mediator interaction. It provides consistent controlled direct genetic effect estimates and valid tests for testing the absence of the direct effect in both cohort and case-control studies. The proposed method is applied to genome sequence data to estimate and test controlled direct genetic effects on hypertension.
