Mendelian Randomization Analysis Via The Mode-based Robust Estimate And Its Applications In The Causal Inference

Objectives:The study compared the causal inference performance of mendelian randomization method based on the mode-based estimate,simple median estimator,weighted median estimator,penalized weighted median estimator,the inverse-variance weighted method,MR-Egger regression by simulation with varying violations of instrumental variable assumptions,and apply them to test for causal associations of childhood adiposity on type 1 diabetes mellitus.Methods:We expounded the basic principles of mendelian randomization based on simple MBE,weighted MBE,simple median estimator,weighted median estimator and penalized weighted median estimator.In our four different simulation scenarios,simulation 1 and 2 were aimed at evaluating the performance of the MBE under the causal null(b(28)0)in the two-sample setting,with the proportion of invalid instrumental variables vary from 0 to 100%(per 10%).In simulation 3 and 4,the number of invalid instrumental variables(r)was set to 0,b(28)0.1 and b(28)0,respectively,they were aimed at evaluating weak instrumental variable bias in the two-sample and single-sample settings.In each simulation,average causal effect estimates,standard errors,coverage and power were obtained to assess the merits and demerits of different MR methods based on SMBE?WMBE?SME?WME?PWME?IVW and MR-Egger.We used a genetic instrument of 23 single nucleotide polymorphisms(SNPs)to test causal associations between childhood adiposity and type 1diabetes mellitus risk.Results:Under the causal null(b(28)0)in many situations,the MBE presented less bias and lower type-I error rates than other methods.Its power to detect a causal effect was smaller compared with the IVW and WME methods,but was larger than that of MR-Egger regression,with sample size requirements smaller than the others.Our study found positive support for an effect of childhood adiposity on type 1 diabetes mellitus risk using SMBE,and WMBE analysis,and found the results were consistent with MR-Egger regression analysis(odds ratio,95%CI:SMBE:2.212;1.084-4.512;WMBE:2.233;1.095-4.554;MR-Egger:2.776;1.241-6.208 per standard deviation in body mass index).Conclusions:The mode-based estimate relaxes a greater degree of the instrumental variable assumptions in mendelian randomization study compared to other methods,and should be used in combination with the other approaches in causal inference,not only presents the result of one of the methods.Our study provides the positive causal association support for a link between childhood adiposity and type 1 diabetes mellitus risk.