Second Use Case Workshop “Help shape the future of AI-driven health data” in Portugal | 28 February 2025

On February 28, 2025, a group of 13 AI developers and health tech professionals gathered to discuss the future of privacy-compliant health data solutions. The workshop focused on shaping a Privacy-Compliant Health Data-as-a-Service Platform designed to meet the evolving needs of the healthcare industry.

Key Takeaways:

• Question 1: System Requirements
  Essential functionalities include access to anonymized medical images, federated learning infrastructure, data interoperability, and standardization. Challenges include GDPR restrictions, lack of clinical site support, and insufficient datasets. A centralized data space could facilitate AI training.
• Question 2: Anticipated Improvements
  The system is expected to enhance data accessibility, improve AI model generation, ensure data quality, and optimize research outcomes. It could also enable better connections between medical conditions and facilitate cost reductions in data handling.
• Question 3: Current Challenges
  Major issues include poor data quality, difficulties accessing clinical data, lack of standardization, scalability concerns, and complex data governance. Hospitals are reluctant to share data due to privacy concerns, and visualization tools are needed to connect datasets effectively.
• Question 4: Compliance and Security
  The system should align with EHDS and GDPR, ensuring data governance compliance. Participants emphasize that anonymization, rather than synthetization, is sufficient for regulatory compliance. Ethical approvals and privacy concerns remain barriers to data sharing.
• Question 5: Integration and Technical Feasibility
  Integration challenges include poor hospital infrastructure, limited computational resources, and lack of interoperability between systems. Suggestions include federated learning and homomorphic encryption to enable secure off-site computing.
• Question 6: Success Metrics and Impact
  Improvements in data quality, AI-driven discoveries, and real-world healthcare benefits should measure success. Metrics should include model accuracy, system adoption, and impact on medical research and practitioners’ efficiency.
• Question 7: Use Case Monetization
  Monetization strategies involve compensating data providers and establishing a sustainable funding model. Research institutes could charge for system usage, while pharmaceutical and insurance companies could pay for access. Ensuring financial sustainability is a key priority.
• Question 8: Synthetization and Anonymization
  Data anonymization should maintain diversity (e.g., race, gender) and high quality. Synthetic data can be helpful but must not compromise privacy or misrepresent rare conditions. There’s concern about losing individual identifiers across hospitals, which could affect causality research.