[Paper Review] A More Robust Approach to Multivariable Mendelian Randomization

Yinxiang Wu, Hyunseung Kang+1 more|arXiv (Cornell University)|Feb 1, 2024

Gene expression and cancer classification6 citations

TL;DR

The paper develops a general asymptotic regime for many weak instruments in summary-data MVMR and introduces debiased IVW estimators (MV-dIVW and MV-adIVW) with robustness to weak instruments and balanced pleiotropy.

ABSTRACT

Multivariable Mendelian randomization (MVMR) uses genetic variants as instrumental variables to infer the direct effects of multiple exposures on an outcome. However, unlike univariable Mendelian randomization, MVMR often faces greater challenges with many weak instruments, which can lead to bias not necessarily toward zero and inflation of type I errors. In this work, we introduce a new asymptotic regime that allows exposures to have varying degrees of instrument strength, providing a more accurate theoretical framework for studying MVMR estimators. Under this regime, our analysis of the widely used multivariable inverse-variance weighted method shows that it is often biased and tends to produce misleadingly narrow confidence intervals in the presence of many weak instruments. To address this, we propose a simple, closed-form modification to the multivariable inverse-variance weighted estimator to reduce bias from weak instruments, and additionally introduce a novel spectral regularization technique to improve finite-sample performance. We show that the resulting spectral-regularized estimator remains consistent and asymptotically normal under many weak instruments. Through simulations and real data applications, we demonstrate that our proposed estimator and asymptotic framework can enhance the robustness of MVMR analyses.

Motivation & Objective

• Motivate the challenges of weak instrumental variables (IVs) in multivariable MR (MVMR) with summary data.
• Propose a general asymptotic regime that permits different IV strengths across linear combinations of exposures.
• Develop debiased IVW estimators (MV-dIVW and MV-adIVW) with theoretical guarantees.
• Extend methods to handle balanced horizontal pleiotropy.
• Provide theoretical properties, simulations, and real-data demonstrations, with software in R.

Proposed method

• Formulate a two-sample summary-data MVMR model with p SNPs and K exposures.
• Introduce Assumptions 1–2 to model many weak IVs and define a new IV strength metric via S_n and the matrix sum_j gamma_j gamma_j^T sigma_Yj^{-2}.
• Analyze the MV-IVW estimator and derive conditions for consistency and asymptotic normality under varying instrument strength regimes.
• Propose MV-dIVW by replacing the biased term with its unbiased counterpart to debias MV-IVW, and derive its properties.
• Define MV-adIVW as a data-adaptive adjustment improving finite-sample robustness and establish its consistency and asymptotic normality.
• Extend the framework to balanced horizontal pleiotropy.

Experimental results

Research questions

• RQ1 How does MV-IVW behave under many weak IVs with heterogeneous instrument strength across exposures?
• RQ2 Can we design debiased estimators (MV-dIVW, MV-adIVW) that remain consistent and asymptotically normal under weak-IV regimes?
• RQ3 How can the method be extended to account for balanced horizontal pleiotropy in MVMR?
• RQ4 What are practical metrics to assess IV strength in summary-data MVMR with multiple exposures?
• RQ5 How do the proposed estimators perform in simulations and real GWAS datasets compared to existing MV-IVW methods?

Key findings

• MV-IVW’s asymptotic behavior depends critically on instrument strength across linear combinations of exposures.
• MV-dIVW reduces weak-IV bias and has broader consistency and asymptotic-normality conditions than MV-IVW.
• MV-adIVW provides improved finite-sample robustness through data-adaptive adjustment and is consistent and asymptotically normal under milder conditions.
• The estimators can be extended to handle balanced horizontal pleiotropy.
• The authors provide theoretical results, simulations, and a real-data demonstration, with software in R (mr.divw).