The Use of Mendelian Randomization in Causal Effect Estimations

Mendelian randomization is a technique for assessing the causal effects of heritable exposure variables on outcomes with observational data. In practice, data on the exposure variables can be incomplete due to the high cost of measurement and technical limits of detection. In this dissertation, we propose methods for Mendelian randomization with incomplete measurements on the exposures. In the first project, we consider a one-sample Mendelian randomization analysis with a continuous outcome and a continuous exposure, where individual-level data of the exposure can be unmeasured or undetectable. We estimate the causal effect with a maximum likelihood estimator and develop an expectation-maximization algorithm for computation. Simulation studies show that the proposed method performs well in various scenarios. We apply our method to the Hispanic Community Health Study/Study of Latinos and estimate the causal effect of several metabolites on phenotypes of interest. The second project extends the discussion from univariable Mendelian randomization to multivariable Mendelian randomization, which enables simultaneous estimations of direct causal effects of multiple correlated exposure variables on an outcome. We estimate the direct causal effects with maximum likelihood estimation and address the missing data with the expectation-maximization algorithm. We assess the performance of the proposed method through extensive simulation studies. We apply our method to the Hispanic Community Health Study/Study of Latinos; by comparing the results with those in the first project, we yield new insights into the relevant biological pathways. In the third project, we perform a Mendelian randomization analysis that involves multiple correlated outcomes. In the existing literature, there has been interest in using Mendelian randomization to assess the causal effects of branched-chain amino acids on multiple metabolic phenotypes. However, these analyses considered each outcome separately and did not take the correlation among the metabolic phenotypes into account. We apply the multi-response Mendelian randomization method to evaluate the direct causal effects of branched-chain amino acids on a set of metabolic phenotypes among the European population. We identify isoleucine and leucine as key exposure variables for calcium and albumin, respectively. We also find strong evidence of valine being a crucial exposure variable for alkaline phosphatase and glycated hemoglobin.