Getting to GRIPS with MR-Egger: Modelling directional pleiotropy independently of allele coding

Area of Science:

• Biostatistics
• Genetic Epidemiology
• Causal Inference

Background:

• Mendelian Randomisation Egger regression (MR-Egger) is widely used for causal inference.
• It relies on single-nucleotide polymorphisms (SNPs) as instrumental variables.
• MR-Egger assumes pleiotropic effects are independent of exposure effects (InSIDE assumption), but is sensitive to allele coding.

Purpose of the Study:

• To address the allele coding sensitivity in MR-Egger.
• To propose a more robust method for causal inference in the presence of pleiotropy.

Main Methods:

• Demonstrated the theoretical limitation of MR-Egger concerning allele coding and the InSIDE assumption.
• Developed a modified MR method, MR-GRIP, with Genotype Recoding Invariance Property (GRIP).
• MR-GRIP is validated under the Variance Independent of Covariance Explained (VICE) assumption.

Main Results:

• Showed that the InSIDE assumption for MR-Egger cannot hold under all-positive coding for both exposure and pleiotropic effects.
• MR-GRIP achieves the goals of MR-Egger while overcoming allele coding challenges.
• Simulations and examples indicate MR-GRIP can resolve discrepancies with other MR methods.

Conclusions:

• The allele coding sensitivity of MR-Egger undermines its soundness.
• MR-GRIP provides a statistically sound and invariant method for causal inference.
• MR-GRIP offers a valuable alternative for genetic epidemiology studies.