Two-sample Mendelian Randomization Methods Using Multiple Genetic Variants And The Application In Causal Association Studies

Objective:Based on the two-sample Mendelian randomization design,this paper will introduce the fundamental theory of three Mendelian randomization methods using multiple genetic variants—the IVW method,MR-Egger method and WME method,and explore their performance in causal association studies with varying violations of the Mendelian randomization assumptions.Methods:During the simulation study,we considered a sample size of 2000;the number of genetic instruments was 25 and the minor allele frequency of all genetic variants was 0.3;the proportions of invalid instrumental variables were 0.1,0.2 or 0.3,respectively.Each scenario is simulated with a null effect(β =0)and a positive effect(β =0.1).Four scenarios with varying violations of the Mendelian randomization assumptions were considered.Simulate data were generated and analyzed using R software.We assessed the quantitative and qualitative performance of each method using bias of the causal estimates and power to reject the causal null hypothesis.Finally,we investigated whether plasma concentrations of low-density lipoprotein cholesterol,high-density lipoprotein cholesterol and triglycerides reflect processes causal for coronary heart disease using 185 SNPs identified from the GLGC and CARDIoGRAM study.Results:Under a weaker set of assumptions than typically used in Mendelian randomization,the MR-Egger regression method can be used to detect and correct for the bias due to directional pleiotropy and Type I error rates are always at nominal levels.Effectiveness effects can also correct the bias of the estimation of causal effects caused by gene pleiotropic effects.Both type I error rates of the casual null hypothesis and the bias of causal estimates are greatly increased when the InSIDE assumption is violated.The WME method has the smallest standard error when estimating causal effects compared to the other two methods.The bias of causal effect estimators and the type I error rate or ability to detect causal effects of the WME method are among the other two methods when the InSIDE assumption is satisfied,while the ability to detect causal effects of IVW method is always higher than the IVW method and MR-Egger regression method.The type I error rates of the casual null hypothesis by MR-Egger regression method are always less than WME method and MR-Egger regression method.when the InSIDE assumption is violated,the performance of WME method is always better than IVW method and MR-Egger regression method as long as the proportion of invalid instrumental variables is not too large.Conclusion:The MR-Egger regression method should be preferred when the InSIDE assumption is satisfied.In case that the InSIDE assumption is violated,the WME method can be selected as long as the proportion of invalid instrumental variables is not too large.The MR-Egger regression method should be selected if the proportion of invalid instrumental variables is large.It can be found that the test of assumed conditions is crucial for the selection of Mendelian randomization methods.All the methods of this paper have their own advantages and limitations for obtaining causal inferences,therefore it is impossible to recommend an authoritative method that is applicable to all Mendelian randomization analyses.Our advice in Mendelian randomization investigations are in doubt for some or all genetic variants,would therefore be to perform and report results from a range of sensitivity analyses using robust methods,including the WME and MR-Egger regression methods,in addition to the typical IVW method.