No More Free Lunch: Challenges to Mendelian Randomization Due to Sample Selection and Complex Methods.

Area of Science:

• Epidemiology
• Genetic Epidemiology
• Statistical Genetics

Background:

• Mendelian randomization (MR) is a powerful tool for inferring causality in epidemiological research.
• MR relies on instrumental variable assumptions: relevance, independence, and exclusion restriction.
• Random genetic variant allocation is presumed to mitigate confounding bias.

Purpose of the Study:

• To discuss potential sources of bias in Mendelian randomization analyses.
• To explore scenarios leading to bias, including collider bias and indirect genetic effects.
• To provide practical strategies for mitigating these biases in MR studies.

Main Methods:

• Utilized causal directed acyclic graphs (DAGs) to model potential biases.
• Examined biases arising from nonrandom participant selection in genome-wide association studies (GWAS).
• Investigated indirect genetic effects in population-based versus within-family studies and nonlinear MR analyses with gene-environment interactions.

Main Results:

• Identified collider bias as a potential issue due to nonrandom selection into GWAS populations.
• Highlighted indirect genetic effects as a source of bias in population-based GWAS.
• Discussed collider bias in nonlinear MR analyses involving gene-environment interactions.

Conclusions:

• Mendelian randomization analyses are susceptible to biases not always accounted for.
• Careful consideration of study design, data selection, and analytical methods is crucial for valid causal inference.
• Practical approaches are needed to detect and reduce biases in MR studies for more reliable results.