Mendelian Randomization when Many Instruments are Invalid: Hierarchical Empirical Bayes Estimation

Abstract: Estimating the causal effect of an exposure on an outcome is an important task in many economical and biological studies. Mendelian randomization, in particular, uses genetic variants as instruments to estimate causal effects in epidemiological studies. However, conventional instrumental variable methods rely on some untestable assumptions, which may be violated in real problems. In this paper, we adopt a Bayesian framework and build hierarchical models to incorporate invalid effects of instruments. We introduce an empirical Bayes estimator for which some of the instruments are invalid by utilizing a Gaussian mixture prior. Theoretical performance and algorithm implementations are provided and illustrated. The reliable performance of the proposed method is demonstrated in various simulation settings and on real datasets concerning the causal effects of HDL cholesterol and LDL cholesterol on type 2 diabetes.