Mendelian randomization in health research: using appropriate genetic variants and avoiding biased estimates.

Area of Science:

• Epidemiology
• Genetic Epidemiology
• Biostatistics

Background:

• Mendelian randomization (MR) is increasingly used in epidemiology to infer causal relationships.
• MR utilizes genetic variants as instrumental variables to address confounding and reverse causality.
• Concerns exist regarding potential biases in MR analyses, particularly in candidate gene studies.

Purpose of the Study:

• To demonstrate how selecting genetic variants based on their association within the study dataset can bias Mendelian randomization (MR) results.
• To highlight the risk of generating spurious causal associations due to data-driven variant selection.
• To discuss bias arising from using a poor proxy for the true exposure in MR.

Main Methods:

• Theoretical demonstration of bias in Mendelian randomization (MR).
• Focus on selection bias introduced by choosing genetic variants based on their association with the exposure in the specific study.
• Examination of bias from measurement error when the exposure is poorly proxied.

Main Results:

• Selecting genetic variants based on their association with the exposure in the analysis dataset can lead to biased MR estimates.
• This data-driven selection can create the appearance of a causal effect where none truly exists.
• Using a poor proxy for the actual exposure can also distort the magnitude of estimated causal effects.

Conclusions:

• Researchers must be cautious about selecting genetic variants for Mendelian randomization (MR) based on their own data to avoid spurious findings.
• Careful consideration of instrumental variable selection and exposure measurement is crucial for valid MR studies.
• Findings are illustrated with examples from tobacco research.