Transfer learning-based two-sample Mendelian randomization method for heterogeneous population.

Area of Science:

• Epidemiology
• Statistical Genetics
• Causal Inference

Background:

• Population heterogeneity, where covariate distributions differ across populations, poses a significant challenge for two-sample Mendelian randomization (MR).
• This heterogeneity can lead to biased causal effect estimates, particularly when covariates act as both confounders and effect modifiers.
• Existing MR methods often struggle to account for these complex population differences.

Purpose of the Study:

• To develop a novel method, transfer learning-based Mendelian randomization (TLMR), to address population heterogeneity in MR.
• To enable accurate causal effect estimation in a target population by leveraging data from a source population.
• To provide a flexible and robust MR approach applicable to various outcome types and settings.

Main Methods:

• TLMR utilizes observable effect modifiers to transfer predicted exposures from a source to a target population.
• The method employs minimal modeling assumptions, allowing for flexible exposure modeling.
• TLMR supports both continuous and binary outcomes and includes an extension for reverse transfer in the outcome model.

Main Results:

• Simulations demonstrate that TLMR provides robust and consistent estimates in heterogeneous populations.
• TLMR outperforms eight widely used MR methods that exhibit substantial estimation bias.
• In homogeneous populations or without effect modification, TLMR performs comparably to existing approaches.
• The study systematically evaluated the causal relationship between body mass index and pulmonary function using TLMR.

Conclusions:

• TLMR effectively addresses population heterogeneity in two-sample MR, yielding accurate causal effect estimates.
• The method offers improved accuracy and practical utility compared to existing MR approaches.
• TLMR is a valuable tool for causal inference in diverse populations, as evidenced by its application to body mass index and pulmonary function.