XMR: A cross-population Mendelian randomization method for causal inference using genome-wide summary statistics

Abstract

Mendelian randomization (MR) is an important tool for inferring causal relationships between exposures (like lifestyle factors or biomarkers) and health outcomes using genome-wide association study (GWAS) summary data, yet the small sample sizes of non-European populations often result in insufficient instrumental variables (IVs) and unreliable causal effect estimates. In this paper, we consider causal inference in underrepresented populations to improve global health equity. We propose a statistical method for cross-population MR, XMR, to enhance causal inference in these target populations by using auxiliary GWAS summary statistics from global biobanks. By leveraging the shared genetic basis of exposure traits in the target and auxiliary populations, XMR increases the number of IVs while maintaining robust estimates via rigorous evaluation of IV validity and accounting for confounding factors. Through extensive simulations and real-data analyses, we demonstrate that XMR can achieve greater statistical power, better control of false positive rates and more replicable results compared to existing methods. Notably, XMR successfully identifies novel causal relationships in our studies of the East Asian (including Japanese and Taiwanese), Central/South Asian, and African populations. These findings reveal potential heterogeneity in causal patterns across populations, highlighting the importance of causal inference in underrepresented populations.