Weak-Instrument and Pleiotropy-Robust Methods for Mendelian randomisation, with Applications to Mental Health

Abstract

This PhD dissertation focused on developing and applying new methods for Mendelian Randomisation (MR), a technique that uses genetic variants as instrumental variables in order to assess causal effects of exposures on health outcomes. The major focus of the applied research is psychiatric research and mental health, with a range of analyses that address the topic of causal risk factors for depression with the use of these genetics-informed methods. The first contribution of this dissertation is the development of new methods for pleiotropy-robust MR by leveraging sex specificity of phenotypes. These methods allow for more accurate and robust estimation of causal effects by cancelling out potential pleiotropic effects of genetic instruments. The second contribution is a new method for appraising high-dimensional correlated variables in multivariable MR. This method allows for the inclusion of multiple correlated variables as exposures in MR analyses, through a transformation to groups of exposures that have attractive statistical properties and biological meaning. Finally, the dissertation provides an applied analysis of how inflammation and BMI affect a range of depression phenotypes with cutting-edge methods. This analysis replicates previous results on the harmful effects of overweight on mood and challenges the independent effect of inflammation as proxied by CRP. The introduction of the dissertation is divided into two parts. The first part provides a walkthrough of the epidemiological concepts of bias, randomisation, and causal inference with observational data. The second part is a specific introduction to MR, including its underlying assumptions and limitations, as well as detailed discussion of developments that make it more robust. Overall, this dissertation contributes new methods and applied analyses to the field of MR, with potential implications for researchers and practitioners.