2887

A measured association between an exposure and a disease admits a causal interpretation when the exposure is randomly selected or assigned. However, exposure selection or assignment may be partly determined by genomic variables. This talk shows how to control for confounding by a set of unmeasured genomic variables.
It is sufficient to bound three confounding parameters: genetic correlation, heritability of the disease, and the proportion of variation in exposure attributable to the unmeasured set of genomic variables. From these bounds we can compute an interval to capture the adjusted exposure effect. This computation is exact with negligible run time and based on a mathematical solution to a non-convex, constrained, optimization problem.