|
Abstract:
|
Instrumental variables (IV) have been used for estimating the causal effect of a treatment on an outcome in observational studies. Using genotypes as IV, these methods have recently been applied in genetic epidemiology. The goal of introducing IV is to remove the effects of unobserved factors that might confound the relationship between the biomarkers and the outcome. A valid inference requires that IV should have a relatively strong association with the biomarkers, but must not have a direct effect on the outcome and not be related to unmeasured confounders. However, a small number of SNPs, mingled with many null SNPs, often provide limited explanation of the variability in a biomarker and can only serve as weak IVs. Therefore, we proposed new methods to select valid IVs or use combinations of SNPs as IVs to increase the variant-biomarker association and thus improve the causal effect inference of the biomarker when the dimensions of candidate SNPs is high. The properties of the proposed methods are demonstrated through simulations and real genomic studies.
|
