Bayesian data integration with applications in health sciences

Dr. Guangquan Li della School of Engineering, Physics and Mathematics della Northumbria University (U.K.) 

Abstract: Data collection in the health sciences, spanning both individual and population levels, has grown substantively over recent years. This abundance of information presents a unique opportunity to uncover the underlying structure of complex health phenomena at unprecedented resolution, enabling, for example, near-real time disease surveillance at fine spatial scales and the monitoring of future health trajectories at the patient level. However, integrating these multidimensional and disparate data sources poses considerable challenges, including differences in scale and granularity, heterogeneity across different analysis units, errors in measurement, and the need to capture complex dependency structures inherent in the data – these obstacles are common across a wide range of applications.

In this talk, I present a unified data integration framework in which Bayesian hierarchical models are employed to address these challenges simultaneously. I illustrate the approach through two health applications: the first on integrating novel data sources for national COVID-19 surveillance, and the second on the dynamic prediction of personalised health risks. Together, these examples demonstrate the applicability and flexibility of the framework in turning complex, heterogeneous health data into actionable insights.