| Objectives:The resolution of the problem of pleiotropy is a key issue in Mendelian randomization studies,and the genetic variation found between many traits in the study is correlated,including some unlikely to be causally related,suggesting that correlated pleiotropy is common and may is a significant source of false positives.This study evaluated the performance of IVW,c ML-MA,Conmix,WME,MBE metho and HPMA under different types and proportions of horizontal pleiotropy through simulation.And to explore the causal association between lipids(TC,HDL,LDL,TG)and early AMD.Methods:The basic principles of IVW,c ML-MA,Conmix,WME,MBE and HPMA were introduced.By set four simulation situations,and explored the influence on each method when including all instrumental variables as effective instrumental variables,only the existence of uncorrelated pleiotropic instrumental variables,only the existence of correlated pleiotropic instrumental variables,both correlated and uncorrelated instrumental variables existed at the same time,the fourth situation also included the situation under different quantitative relationship of correlated and uncorrelated invalid instrumental variables.The average causal effect size,standard deviation,mean square error,type I error rate,performance,and 95% confidence interval coverage were used as indicators to evaluate the advantages and disadvantages of each method.In application study,the c ML-MA was used as the main method,and other methods were used as the sensitivity analysis to explore the causal association between lipids and early AMD.Results:The simulation results showed that the c ML-MA was relatively robust and was less affected by correlated pleiotropic effects than other methods.Conmix,MBE,and WME were greatly affected by correlated pleiotropic effects.As the proportion of un correlated pleiotropic effects increases from 0.25 to 0.75,the proportion of processing pleiotropic effects increases,that is,they were greatly affected by the assumption of exclusive.The Conmix as a whole can control the proportion of invalid instrumental variables to about40%,but the coverage rate of 95% was not good,and when the number of SNPs increases,the performance of this method become better,but the coverage rate was still poor.The performance of MBE method was lower than other methods,and the control of type I error rate is too conservative,but this method performed better with the increase of the proportion of invalid instrumental variables.The HPMA was similar to the MBE,but due to the limitations of the method itself,it can only deal with a small number of SNPs.The WME has no outstanding advantages as a whole.This method can deal with correlated pleiotropic effects,but only about 10% of the invalid instrumental variables can be dealt with.According to the results of the example,we found that HDL,LDL,and TG all have effects on early AMD,among which HDL is a risk factor for early AMD(OR: 1.22,95%CI: 1.15,1.29,p<0.001),LDL and TG had a negative causal effect on the early AMD(OR: 0.83,95%CI: 0.76,0.89,p<0.001;OR: 0.78,95CI%: 0.73,0.82,p<0.001).In the analysis of the effect of TC on early AMD,the p-value did not reach the Bonferroni-corrected p-value,but was below the nominal level,suggesting evidence of causality that needs further confirmation.Conclusions:Compared with other MR methods that allow the existence of correlated pleiotropy,the c ML-MA method has higher robustness in the case of different proportions and types of pleiotropy,so it is recommended to be used in two-sample MR analysis.However,in practical applications,different MR methods can be combined to provide robust evidence for investigating whether the causal effect of exposure on outcomes is statistically significant.In the application,we found that HDL is a risk factor for early AMD,and LDL and TG are protective factors for early AMD. |
PDF Full Text Request