On the use of kernel machines for Mendelian randomization

Area of Science:

• Biostatistics
• Genetic Epidemiology
• Causal Inference

Background:

• Accurate estimation of causal effects in health studies requires adjustment for unmeasured confounders.
• Instrumental variable (IV) methods, including Mendelian randomization (MR), are used to address confounding.
• Two-stage least squares (TSLS) is a common IV/MR estimator, but its validity depends on precise first-stage exposure prediction.

Purpose of the Study:

• To propose a novel method for modeling the first stage of TSLS in IV and MR studies.
• To evaluate the performance of the proposed method using simulation studies.

Main Methods:

• Modeling the association between genetic IVs and exposure using the least-squares kernel machine (LSKM).
• Integrating LSKM into the TSLS framework for causal effect estimation.
• Simulation studies to assess LSKM's feasibility and performance in TSLS.

Main Results:

• LSKM effectively models the relationship between genetic variants and exposure within the TSLS framework.
• Simulation results demonstrate the feasibility of using LSKM in TSLS settings.
• LSKM may enhance statistical power when the exposure-IV relationship is nonlinear.

Conclusions:

• LSKM, utilizing either genotype score or genotype data, is a viable approach for the first stage of TSLS.
• The LSKM-based TSLS estimator offers potential advantages in power for nonlinear exposure-IV associations.
• This method provides a robust tool for causal inference in genetic epidemiology.