An examination of multivariable Mendelian randomization in the single-sample and two-sample summary data settings

Area of Science:

• Epidemiology
• Genetic Epidemiology
• Statistical Genetics

Background:

• Mendelian randomization (MR) uses genetic variants as instrumental variables (IVs) to infer causal relationships.
• Unobserved confounding is a challenge in observational epidemiology.
• Multivariable MR (MVMR) extends MR to assess multiple exposures simultaneously.

Purpose of the Study:

• To clarify the interpretation of MVMR effects under various exposure relationships (confounder, mediator, pleiotropy, collider).
• To develop methods for assessing instrument strength and validity in MVMR.
• To apply MVMR to estimate the effects of education and cognitive ability on body mass index.

Main Methods:

• Simulations and theoretical analysis to explore MVMR interpretation.
• Development of methods for assessing instrument strength and validity.
• Application of MVMR to UK Biobank data.

Main Results:

• MVMR consistently estimates direct causal effects.
• Methods were developed for assessing instrument validity in single-sample and two-sample settings.
• The study illustrated MVMR's utility with real-world data.

Conclusions:

• MVMR is a robust tool for estimating direct causal effects of exposures.
• It is applicable to both individual-level and summary-level data.
• MVMR aids in understanding complex causal pathways in epidemiology.