Avoiding collider bias in Mendelian randomization when performing stratified analyses.

Area of Science:

• Epidemiology
• Statistical Genetics
• Causal Inference

Background:

• Mendelian randomization (MR) uses genetic variants as instrumental variables to infer causal relationships.
• Stratifying MR analyses by population characteristics can reveal effect heterogeneity but risks collider bias.
• Collider bias arises when a variable is a common effect of both the exposure and the outcome, or their common causes.

Purpose of the Study:

• To propose a novel statistical approach to perform Mendelian randomization in population strata while avoiding collider bias.
• To develop a method for valid causal inference in subgroup analyses within MR studies.
• To investigate the causal effect of smoking on bladder cancer, accounting for bodyweight as a potential collider.

Main Methods:

• Introduced a 'residual collider' variable derived from regressing the collider on the genetic instrument.
• Calculated Mendelian randomization estimates stratified by quantiles of the residual collider.
• Validated the approach through simulation studies with varying collider and instrument characteristics.
• Applied the method to real-world data examining smoking and bladder cancer risk across bodyweight strata.

Main Results:

• The proposed residual collider method yielded unbiased Mendelian randomization estimates in all simulated scenarios.
• Stratum-specific estimates in the smoking-bladder cancer example showed a trend suggesting stronger effects at lower bodyweight.
• The approach effectively mitigates collider bias in stratified MR analyses.

Conclusions:

• The residual collider method provides a robust framework for conducting Mendelian randomization subgroup analyses without collider bias.
• This approach enhances the reliability of causal inference for risk factors across diverse population subgroups.
• The findings highlight the potential utility of this method for exploring effect modification in epidemiological research.