Leveraging longitudinal data from electronic health records to boost statistical power for gene-environment interaction analysis

Gene-environment interactions (G E) play a crucial role in advancing genetic discovery, adressing missing heritability, and facilitating precision medicine. However, most existing G E methods are designed for cross-sectional data, limiting their utility in analyzing the rich longitudinal data in electronic health records. Here, we propose SAGELD, a scalable and accurate genome-wide G E method of longitudinal traits while controlling for sample relatedness in large-scale datasets. SAGELD employs a matrix projection strategy to construct test statistics and adopts the SPAGRM framework for efficient control of sample relatedness, achieving speedups of 10- to 10,000-fold over existing methods while maintaining greater statistical power. We evaluated SAGELD through extensive simulations and real data analyses in the UK Biobank. Using age and body mass index as environmental exposures, we identified 74 genetic loci with significant G age interactions and five with G adiposity interactions, many of which are novel findings. These results highlight the advantages of leveraging longitudinal data in G E analyses.