To improve health and health care, researchers and policy-makers are increasingly leveraging the vast amount of administrative health information that is routinely captured by health care systems. These sizable datasets comprised of medical claims and electronic health records are especially valuable for studying rare outcomes. To feasibly adjust for the many available database confounders that may distort a treatment-outcome association, propensity score (PS) methods are often used. However, relatively little is known about the statistical performance and practical challenges of implementing PS methods in this multi-site, rare outcome setting. This session presents recent statistical advances designed to improve our understanding of PS methods when used in this context. In this overview talk, I will set the stage by 1) describing how PS methods are currently being applied in multi-site safety data networks and 2) highlighting challenges that have been encountered.