iMpact Seminar – Linying Zhang

Real-world evidence (RWE) plays a vital role in complementing clinical trials to inform drug safety and effectiveness, especially in the context of rare adverse events. However, generating reliable RWE from real-world data (RWD), such as electronic health records, presents significant challenges. These data are often very large, inaccurate, incomplete, and siloed behind institutional firewalls. In this talk, Dr. Linying Zhang will present novel methodologies and scalable workflows for leveraging federated research data networks to generate RWE while preserving patient privacy. She will highlight a recent case study evaluating the comparative safety of semaglutide, a GLP-1 receptor agonist used for diabetes, versus other anti-glycemic medications on the risk of non-arteritic anterior ischemic optic neuropathy (NAION), a rare vision-threatening condition. This study uses 14 observational health databases from multiple institutions within the OHDSI network. To improve the robustness of causal inference from RWD, Dr. Zhang will introduce the use of causal representation learning to extract low-dimensional patient embeddings from high-dimensional health data for treatment effect estimation. She will also discuss the integration of geospatial data to capture neighborhood-level social determinants of health (SDoH), which further reduces confounding bias and enables large-scale analyses of health disparities.