Contextualizing selection bias in Mendelian randomization: how bad is it likely to be?

Area of Science:

• Epidemiology
• Genetic Epidemiology
• Biostatistics

Background:

• Selection bias is a concern in Mendelian randomization (MR) when study sample selection depends on a collider.
• The relative impact of selection bias in MR compared to other biases is not well understood.

Purpose of the Study:

• To assess the impact of selection bias on Mendelian randomization investigations through simulations.
• To evaluate inverse probability weighting as a method to mitigate selection bias.
• To explore if selection bias could explain a reported null finding for lipoprotein(a) and cardiovascular mortality.

Main Methods:

• Extensive simulation study to model selection bias in MR.
• Application of inverse probability weighting (IPW) to address selection bias.
• Case study investigating lipoprotein(a) and cardiovascular mortality.

Main Results:

• Selection bias significantly impacted bias and Type 1 error rates in simulations, particularly with large selection effects.
• Moderate selection effects resulted in small bias and minimal Type 1 error inflation.
• IPW reduced bias when the selection model was correct but increased bias if misspecified.
• Selection bias plausibly explains the null finding for lipoprotein(a) due to strong genetic associations and confounder effects on selection.

Conclusions:

• Selection bias can negatively affect Mendelian randomization studies, though its impact may be less severe than other biases.
• The magnitude of selection bias is substantial only when risk factor and confounder effects on selection are very large.