B1721
Title: Bayesian methods for vaccine safety surveillance using federated data sources
Authors: Fan Bu - University of Michigan (United States) [presenting]
Abstract: A Bayesian sequential analysis framework is discussed for data sources that are distributed across a federated network, motivated by vaccine safety surveillance studies. Rapid detection of vaccine safety events is enabled by observational healthcare data that accrue over time. The framework aims at resolving three main challenges: first, control of testing errors in sequential analyses of streaming data; second, correction of bias induced by observational data; third, distributed learning of federated data sources while preserving patient-level privacy. Through the extraction of profile likelihoods that retain rich distributional information while protecting individual-level privacy and hierarchical analysis of negative control outcomes, these challenges are tackled in a unified statistical framework. As evidenced by a large-scale empirical evaluation using real-world data sources, the framework provides substantial improvements over existing approaches to safety surveillance.
