B1759
Title: Addressing weak instruments in one sample MR analysis with MR-SPLIT
Authors: Yuehua Cui - Michigan State University (United States) [presenting]
Abstract: Mendelian Randomization (MR) is a widely embraced approach to assess causality in epidemiological studies. However, the two-stage least squares (2SLS) method, a predominant technique in MR analysis, can lead to biased estimates when instrumental variables are weak. Focusing on one sample MR analysis, a novel method known as Mendelian randomization with adaptive sample-splitting with cross-fitting instruments (MR-SPLIT) is introduced, specifically designed to address issues related to weak instrumental variables and mitigate estimation bias. It is mathematically shown that the MR-SPLIT estimator is more efficient than its counterpart CFMR estimator. Additionally, a multiple sample-splitting technique is introduced to enhance the robustness of type I error control and improve statistical power. Comprehensive simulation studies are carried out to compare the performance of the method against its counterparts, with the results showcasing its superiority in terms of bias reduction, effective type I error control, and increased power. We further validated its utility through application to a real dataset. The study underscores the importance of addressing weak instrumental variables in MR analyses and provides a robust solution to the challenge.
