Addressing Instrument-Outcome Confounding in Mendelian Randomization through Representation Learning
This addresses a key limitation in epidemiological research for causal inference, offering a method to handle violations of core assumptions like instrument independence from unobserved confounders, though it appears incremental as it builds on existing representation learning techniques.
The paper tackles the problem of instrument-outcome confounding in Mendelian Randomization by proposing a representation learning framework that uses cross-environment invariance to recover latent exogenous genetic instruments, demonstrating effectiveness through simulations and semi-synthetic experiments with data from the All of Us Research Hub.
Mendelian Randomization (MR) is a prominent observational epidemiological research method designed to address unobserved confounding when estimating causal effects. However, core assumptions -- particularly the independence between instruments and unobserved confounders -- are often violated due to population stratification or assortative mating. Leveraging the increasing availability of multi-environment data, we propose a representation learning framework that exploits cross-environment invariance to recover latent exogenous components of genetic instruments. We provide theoretical guarantees for identifying these latent instruments under various mixing mechanisms and demonstrate the effectiveness of our approach through simulations and semi-synthetic experiments using data from the All of Us Research Hub.