How Two NHS Trusts Are Training AI Without Sharing a Single Patient Record

The challenge of training healthcare AI without centralizing sensitive patient data may have found a practical solution in the NHS. FLock.io, a federated learning infrastructure company, has been recognized by the World Economic Forum's MINDS programme for deploying privacy-preserving AI with Moorfields Eye Hospital and University College London Hospitals (UCLH).

Unlike conventional AI training that aggregates patient records in centralized environments, FLock.io's platform allows algorithms to learn locally at each hospital site. Only encrypted model updates—not raw clinical data—are shared and combined, addressing GDPR compliance concerns that have stalled numerous European health AI initiatives.

Targeting Diabetes and Eye Disease

The NHS partnerships focus on two high-burden clinical areas. At UCLH, researchers are developing AI-powered glucose prediction alerts trained on data from more than 400 patients with diabetes. The system currently engages approximately 14,000 end users across the UK, Southeast Asia, and East Asia. A multi-continental real-world trial involving 100 patients is scheduled to launch this summer.

FLock.io estimates that AI-driven diabetes prevention could save the NHS over £100 million annually, based on achieving a 1% reduction in the health service's £10+ billion diabetes management expenditure. At Moorfields Eye Hospital, the company has completed an initial research phase and is now training AI models using hospital imaging data for eye disease detection.

Performance and Efficiency Claims

FLock.io reports operational benefits from its federated approach: 37% improvement in model accuracy, 44% reduction in total cost of ownership, 63% shorter deployment time, and 80% less training energy per model update. These metrics incorporate blockchain-based verification mechanisms designed to ensure model integrity across distributed training environments.

If validated at scale, such performance gains could reshape health system AI procurement criteria, particularly as energy costs and sustainability concerns intensify scrutiny of computationally intensive centralized training. The company's inclusion in the WEF MINDS programme alongside industrial giants like Lenovo and KUKA signals growing enterprise confidence in federated architectures for regulated industries.

International Expansion Underway

FLock.io is extending its approach beyond the NHS. The Government of Sarawak, Malaysia has completed a sovereign AI pilot, with subsequent deployments planned for hospital partners in the United States, Europe, and China. This international expansion will test whether federated learning can navigate divergent data governance regimes while maintaining model performance—a critical question as nations increasingly mandate local data residency for healthcare AI.

The technology faces competition from established privacy-preserving AI vendors like Owkin and Rhino Health, but FLock.io's WEF recognition and demonstrated NHS deployments position federated learning as a credible alternative to centralized cloud training. For healthcare systems struggling to balance AI innovation with privacy compliance, the NHS use cases offer a potential roadmap where algorithms travel to the data, rather than the reverse.

Sources

1. NHS AI use cases spotlighted by World Economic Forum for work with FLock.io — Healthcare-in-Europe.com