Improving the accuracy of two-sample summary-data Mendelian randomization: moving beyond the NOME assumption

Area of Science:

• Epidemiology
• Statistical Genetics
• Bioinformatics

Background:

• Two-sample summary-data Mendelian randomization (MR) is widely used for causal inference.
• Homogeneity of genetic variant causal effect estimates is expected if MR assumptions hold.
• Heterogeneity suggests potential violations of instrumental variable (IV) or modeling assumptions.

Purpose of the Study:

• To evaluate the impact of different weighting methods on heterogeneity assessment in MR.
• To develop improved weighting methods for more accurate causal effect estimation and heterogeneity quantification.
• To address bias in MR estimates arising from weak instrumental variables.

Main Methods:

• Investigated first-order and second-order weighting schemes for MR.
• Derived and proposed modified weights to mitigate issues with existing methods.
• Utilized Monte Carlo simulations to compare weighting performance.
• Applied the new method to a real-world MR analysis of systolic blood pressure and coronary heart disease.

Main Results:

• First-order weights can inflate false positives for heterogeneity.
• Second-order weights can increase false negatives for heterogeneity.
• Modified weights demonstrated superior heterogeneity quantification in simulations.
• Modified weights corrected regression dilution bias from weak instruments.
• Modified weights may reduce precision and power with few weak instruments.

Conclusions:

• Modified weights are recommended for accurate heterogeneity quantification and outlier detection in two-sample MR.
• Modified weights show promise for causal estimation, particularly with weak instruments.
• Further research is needed to fully understand the application and limitations of modified weights in various MR settings.