A new study proposes representation learning to advance multi-institution electronic health record (EHR) studies across the United States and France, led by Zhou, Tong, and Wang. The approach is designed to address how to use EHR data from different institutions while improving learnability and reducing the dependence on strict local harmonization. EHR-based research is increasingly central to observational comparative effectiveness, phenotyping, and safety surveillance. Representation learning can help models generalize across sites by capturing underlying clinical structure rather than site-specific artifacts. The work emphasizes cross-border feasibility, a key requirement for scaling real-world evidence generation in drug development and post-market monitoring.