MR-Egger Regression

The defining assumption of MR-Egger is the Instrument Strength Independent of Direct Effect (InSIDE) condition. This requires that the pleiotropic effects of the genetic variants on the outcome are independent of the variant-exposure associations. In other words, the magnitude of a variant's confounding pathway must not be correlated with how strongly it predicts the exposure. This is fundamentally different from the standard IV assumption and is the mechanism that allows the intercept to capture directional bias.

Unlike IVW regression, which forces the regression line through the origin, MR-Egger fits an unconstrained intercept term. This intercept provides a formal statistical test for directional pleiotropy:

• Non-zero intercept: Indicates the presence of directional pleiotropy, meaning the average pleiotropic effect across all variants is not zero.
• Zero intercept: Suggests no evidence of directional pleiotropy, and the MR-Egger causal estimate should converge toward the IVW estimate.

MR-Egger is particularly vulnerable to regression dilution bias caused by imprecise variant-exposure associations. The NOME assumption states that the genetic associations with the exposure are measured without error. Violations of NOME lead to attenuation of the causal effect estimate toward the null. This is quantified by the I²_GX statistic:

• I²_GX > 90%: NOME violation is negligible.
• I²_GX < 90%: Requires the SIMEX (Simulation Extrapolation) correction to adjust the causal estimate for measurement error.

The robustness of MR-Egger to pleiotropy comes at a significant cost in statistical power. By estimating an additional intercept parameter, the standard error of the causal effect is typically much larger than that of an IVW analysis. This means MR-Egger requires a larger sample size and a greater number of genetic instruments to detect a true causal effect. It is generally used as a sensitivity analysis rather than the primary discovery method, with wide confidence intervals expected.

Applying MR-Egger to binary outcomes requires careful methodological consideration. The standard linear MR-Egger model assumes a continuous outcome with constant variance. For case-control studies, the MR-Egger method must be adapted using a multiplicative random-effects model or a generalized MR-Egger framework that accounts for the binary nature of the data. Failure to do so can lead to biased intercept estimates and inflated Type I error rates for the pleiotropy test.

The validity of MR-Egger depends heavily on instrument selection. Best practices include:

• Genome-wide significance: Use variants reaching p < 5×10⁻⁸.
• LD clumping: Ensure instruments are independent (r² < 0.001) to avoid correlated pleiotropy.
• F-statistic filtering: Exclude weak instruments (F < 10) to mitigate weak instrument bias.
• Steiger filtering: Confirm the direction of effect is from exposure to outcome, not reverse causation. A robust instrument set strengthens the InSIDE assumption and improves the reliability of the causal estimate.