Meta-analysis of gene-environment interaction exploiting gene-environment independence across multiple case-control...

Area of Science:

• Genetics
• Epidemiology
• Biostatistics

Background:

• Gene-environment (G-E) independence is used to boost power in G-E interaction testing in case-control studies.
• Limited research exists on G-E independence within meta-analysis frameworks.

Purpose of the Study:

• To extend existing empirical Bayes shrinkage estimators to a meta-analysis setting.
• To develop a method that adjusts for uncertainty in the G-E independence assumption across studies.

Main Methods:

• Utilized a retrospective likelihood framework to combine estimators from constrained (G-E independent) and unconstrained models.
• Weights for combining estimators were determined by measures of G-E association across multiple studies.
• Extended single-study empirical Bayes shrinkage estimators to the meta-analysis context.

Main Results:

• The proposed adaptive estimator demonstrated improved average performance across various simulation scenarios.
• Outperformed standard inverse variance (covariance) weighted estimators combining different types of study-specific estimators.
• The method was illustrated by meta-analyzing 6 type 2 diabetes studies on FTO gene interactions with BMI and age.

Conclusions:

• The novel meta-analysis approach effectively handles uncertainty in G-E independence.
• This method provides a more robust way to test G-E interactions in meta-analyses.
• The findings have implications for genetic epidemiology, particularly in complex diseases like type 2 diabetes.