Mendelian randomisation with coarsened exposures.

Area of Science:

• Epidemiology
• Genetic Epidemiology
• Biostatistics

Background:

• Mendelian randomization (MR) relies on the exclusion restriction assumption, where genetic variants influence outcomes solely through the exposure.
• In practice, exposures are often coarsened approximations of latent continuous traits, potentially violating this assumption.
• Genetically driven outcome variations within exposure categories can lead to biased MR estimates.

Purpose of the Study:

• To propose a framework for understanding and quantifying bias caused by coarsened exposure measurements in MR.
• To develop a method for estimating the effect of standard deviation increases in latent exposures.
• To provide a sensitivity analysis for MR studies with potential violation of the exclusion restriction assumption.

Main Methods:

• Derivation of a bias expression for MR with coarsened exposure data.
• Development of a method using a sensitivity parameter, interpretable as latent exposure's genetic variance.
• Application of the method in both one-sample and two-sample MR settings.

Main Results:

• The proposed framework clarifies the violation of the exclusion restriction assumption due to coarsened exposures.
• Bias derived may inflate or deflate effect estimates but does not reverse their direction.
• The novel method allows for estimation of effects on a continuous latent exposure scale.

Conclusions:

• The developed method provides a robust approach to address bias from coarsened exposure in MR.
• This framework enhances the meaningful interpretation of effect sizes in genetic epidemiology.
• Applied examples and reanalyses demonstrate the practical utility and validity of the method.