Comparison And Application Of Mendelian Randomization Methods For Correcting Weak Instrumental Variable Bias

Objectives:In order to evaluate the performance of the three Mendelian randomization methods of Inverse Variance Weighting with Modified Weights（MW-IVW）,Robust Adjusted Profile Score（RAPS）and MR Mixture Model（MR-Mix）under different parameter settings through simulation study.To provide suggestions for researchers to choose the appropriate Mendelian randomization methods when no instrumental variables are available or weak instrument bias in two-sample Mendelian randomization analysis.To explore the causal effects of body mass index（BMI）,high density lipoprotein（HDL）,low density lipoprotein（LDL）,triglycerides（TG）and total cholesterol（TC）on serum urate.as well as for providing genetic support for then prevention and treatment of gout.Methods:We expounded the basic principles of mendelian randomization based on MW-IVW,RAPS and MR-Mix.In this study,three different simulation schemes were set up.In the case of no pleiotropy,balanced pleiotropy,and directional pleiotropy,respectively,the impact of weak instrumental variables on each method was investigated by changing the strength of instrumental variables.In addition,under the condition that both directional pleiotropic effects and weak instrumental variables exist,the impact of different numbers of instrumental variables on the MR-Mix method is simulated.And mean causal effect value,standard error,typeⅠerror rate power and 95%confidence interval coverage were used as indicators to evaluate the advantages and disadvantages of MW-IVW,RAPS and MR-Mix under different parameter settings.The main analysis was performed using the MR-Mix method,while MW-IVW and RAPS methods were used as sensitivity analyses to estimate the causal association of BMI,HDL,LDL,TG and TC on serum urate.Results:Based on the results of the simulation study,In the case of no pleiotropy and balanced pleiotropy,the MW-IVW method outperformed RAPS and MR-Mix methods.Although the strength of instrumental variables is weakening,the causal effect estimates of the MW-IVW method are always close to unbiased,the standard error is small,the type Ⅰ error rate and 95%confidence interval coverage can be kept near the nominal level,and the statistical power is greater than 90%.The RAPS method performed close to MW-IVW in the no pleiotropic scenario,but its type Ⅰ error rate was significantly inflated in the balanced pleiotropic scenario.The MR-Mix method was too conservative in controlling for type Ⅰ error in the no pleiotropy and balanced pleiotropy cases,and its statistical power was less than 70%in most cases.Among directional pleiotropic cases,MR-Mix performed best.The simulation results showed that the RAPS methods can correct the weak instrumental variable bias to a certain extent,but its 95%confidence interval coverage tends to decrease gradually with the weakening of the instrumental variable strength,and when directional pleiotropic effects exist,its I The type error rate also inflated.Besides,we found that BMI（β=0.280,95%CI:0.023,0.537,p=0.003）and TG(β=0.370,95%CI:0.192,0.548,p=4.436×10^-5)were risk factors for elevated serum urate.HDL was a protective factor for elevated serum urate（β=-0.250,95%CI:-0.64,0.140,p=0.002）.At the same time,this study did not find causal relationship between LDL,TC and serum urate.Conclusion:In the case of no pleiotropy and balanced pleiotropy,the MW-IVW method has better statistical performance than the other two methods;but in the presence of directional pleiotropy,the MR-Mix method has better robustness.In two-sample MR analysis,if researchers cannot obtain instruments or can only obtain a small number of instruments by traditional methods,the threshold for including instrumental variables should be relaxed to allow weak instrumental variables to enter the model.At this time,if there is no pleiotropy in MR model,it is recommended to use the MW-IVW results;if there is directional pleiotropy in MR model,it is recommended to use the MR-Mix method.In this study,MR-Mix was used to find that BMI and TG were risk factors for elevated serum urate,which provided a theoretical basis for the prevention and treatment of gout.