Avoiding bias from weak instruments in Mendelian randomization studies.

Area of Science:

• Biostatistics
• Genetic Epidemiology
• Causal Inference

Background:

• Mendelian randomization (MR) uses genetic variants as instrumental variables (IVs) to estimate causal effects.
• Estimates in MR are susceptible to bias influenced by the strength of the IV-phenotype relationship (F-statistic).

Purpose of the Study:

• To develop guidelines for designing and analyzing MR studies to minimize bias.
• To investigate the impact of study size, instrument selection, and analysis methods on bias.

Main Methods:

• Instrumental variable (IV) analysis was applied to simulated and real-world data.
• The study examined the influence of various factors on bias, including F-statistic, model parsimony, and covariate adjustment.

Main Results:

• Bias increases as the F-statistic decreases; parsimonious models and covariate adjustment can reduce bias.
• Injudicious instrument choice significantly altered causal estimates, demonstrating substantial bias.
• A correlation between causal estimates and standard errors in meta-analyses introduces further bias, which can be mitigated by using individual-level data.

Conclusions:

• Weak instrument bias is a critical concern in MR study design and analysis.
• Selecting instruments or models post hoc based on F-statistics can worsen bias.
• The common F-statistic > 10 guideline is insufficient to prevent bias in instrumental variable analysis.