SAFEVID-TREs

The new TREvolution programme introduces simplified federated learning capabilities within TREs, addressing both the technical and governance hurdles. Leveraging this innovation, our project aims to explore the feasibility of improving our existing IES detection AI model by developing and evaluating different federated learning options within the DARE ecosystem (e.g., RO-Creates, FS-TES, SATRE, K8TRE).

This project explores how federated learning can safely and effectively connect multiple Trusted Research Environments (TREs) to unlock new insights from health data — without moving or sharing sensitive patient information.

Federated learning allows machine learning models to be trained across different secure data environments while keeping data where it is stored. We are investigating how this approach can work across TREs hosted on different cloud platforms and containing data from diverse patient groups.

In partner with NHS West of Scotland Innovation Hub (WoS-IH) and Health Informatics Centre (HIC) at University of Dundee, our work focuses on:

• Assessing whether federated learning can operate reliably between secure research environments built on different technologies.
• Understanding how models can learn from distinct patient cohorts while maintaining privacy and data security.
• Identifying the most practical and secure ways to implement federated learning within the DARE ecosystem.
• Measuring how this approach improves the performance and reliability of our IES detection machine learning model, particularly by improving confidence and uncertainty estimates in its predictions.

By the end of the project, we will provide clear evidence on how federated learning across multiple TREs can enhance research while maintaining the highest standards of security and governance. Our findings will support researchers and developers across the community in adopting this secure, collaborative, and data-diverse approach to innovation.