A comparison of robust Mendelian randomization methods using summary data.

Area of Science:

• Genetics
• Epidemiology
• Statistical genetics

Background:

• Mendelian randomization (MR) analyses are increasingly used for causal inference.
• The growing number of genetic variants in genome-wide association studies (GWAS) enables more precise MR estimates.
• A key challenge is the presence of invalid instrumental variables among the genetic variants used.

Purpose of the Study:

• To compare the performance of nine robust Mendelian randomization methods using summary data.
• To evaluate methods based on theoretical properties, simulation studies, and empirical examples.
• To provide recommendations for selecting appropriate MR methods.

Main Methods:

• Comparison of nine robust Mendelian randomization methods.
• Theoretical property review.
• Extensive simulation study assessing performance metrics like mean squared error and Type 1 error rates.
• Empirical example analysis.

Main Results:

• The contamination mixture method demonstrated the best performance in simulations, with well-controlled Type 1 error rates even with up to 50% invalid instruments.
• Outlier-robust methods yielded the narrowest confidence intervals in the empirical example.
• Most methods performed poorly when more than 50% of variants were invalid instruments.

Conclusions:

• No single robust method is universally superior; performance varies by metric and scenario.
• Using a variety of robust methods with different assumptions is recommended for reliable MR analysis.
• Investigators should assess the reliability of MR findings by employing diverse analytical approaches.