Burgess, S., Bowden, J., Fall, T., Ingelsson, E., & Thompson, S. G. (2017) discuss sensitivity analyses for robust causal inference in Mendelian randomization (MR) studies using multiple genetic variants. They emphasize that while MR is becoming more powerful due to larger genome-wide association studies (GWAS) and summarized data, the instrumental variable (IV) assumptions are rarely fully satisfied. Thus, relying solely on a single IV is insufficient for causal inference. The authors propose sensitivity analyses to assess the validity of MR conclusions, focusing on methods that can be applied using summarized data. These include testing for pleiotropy, heterogeneity, and directional pleiotropy, as well as robust analysis methods like penalization, median-based methods, and Egger regression. They argue that these approaches help ensure that causal estimates are not biased by violations of IV assumptions. The study uses CRP and coronary artery disease (CAD) as an example, showing how different methods can yield conflicting results, highlighting the need for careful interpretation. The authors conclude that while MR is a valuable tool, its conclusions should be viewed cautiously without proper sensitivity analyses, especially when using multiple genetic variants. Key points include the importance of testing for pleiotropy, heterogeneity, and the limitations of standard MR methods. The study underscores the need for robust methods to assess the validity of causal inferences in MR studies.Burgess, S., Bowden, J., Fall, T., Ingelsson, E., & Thompson, S. G. (2017) discuss sensitivity analyses for robust causal inference in Mendelian randomization (MR) studies using multiple genetic variants. They emphasize that while MR is becoming more powerful due to larger genome-wide association studies (GWAS) and summarized data, the instrumental variable (IV) assumptions are rarely fully satisfied. Thus, relying solely on a single IV is insufficient for causal inference. The authors propose sensitivity analyses to assess the validity of MR conclusions, focusing on methods that can be applied using summarized data. These include testing for pleiotropy, heterogeneity, and directional pleiotropy, as well as robust analysis methods like penalization, median-based methods, and Egger regression. They argue that these approaches help ensure that causal estimates are not biased by violations of IV assumptions. The study uses CRP and coronary artery disease (CAD) as an example, showing how different methods can yield conflicting results, highlighting the need for careful interpretation. The authors conclude that while MR is a valuable tool, its conclusions should be viewed cautiously without proper sensitivity analyses, especially when using multiple genetic variants. Key points include the importance of testing for pleiotropy, heterogeneity, and the limitations of standard MR methods. The study underscores the need for robust methods to assess the validity of causal inferences in MR studies.