MVMRmode: Introducing an R package for plurality valid estimators for multivariable Mendelian randomisation

Abstract Background Mendelian randomisation (MR) is the use of genetic variants as instrumental variables. Mode-based estimators (MBE) are one of the most popular types of estimators used in univariable-MR studies. However, because there are no plurality valid regression estimators, there are no existent modal estimators for multivariable-MR. Methods We use the residual method for multivariable-MR to introduce two multivariable modal estimators: multivariable-MBE, which uses IVW to create residuals fed into a traditional plurality valid estimator, and multivariable-CM which instead has the residuals fed into the contamination mixture method. We then use Monte-Carlo simulations to explore the performance of these estimators when compared to existing ones and re-analyse the data used by Grant and Burgess (2021) looking at the causal effect of intelligence, education, and household income on Alzheimer’s disease as an applied example. Results In our simulation, we found that multivariable-MBE was generally too variable to be much use. Multivariable-CM produced more precise estimates on the other hand. Multivariable-CM performed better than MR-Egger in almost all settings, and Weighted Median under balanced pleiotropy. However, it underperformed Weighted Median when there was a moderate amount of directional pleiotropy. Our re-analysis supported the conclusion of Grant and Burgess (2021), that intelligence had a protective effect on Alzheimer’s disease, while education, and household income do not have a causal effect. Conclusions Here we introduced two, non-regression-based, plurality valid estimators for multivariable MR. Of these, “multivariable-CM” which uses IVW to create residuals fed into a contamination-mixture model, performed the best. This method uses a plurality of variants valid assumption, and appears to provided precise and unbiased estimates in the presence of balanced pleiotropy and small amounts of directional pleiotropy. We developed the MVMRmode R package (available from https://github.com/bar-woolf/MVMRmode/wiki ) to facilitate the use of this estimator. We hope this will further enable the future triangulation of univariable MR studies which have used plurality valid estimators with multivariable MR designs.