Privacy-preserving data quality assessment for federated health data networks

Area of Science:

• Health Informatics
• Data Privacy
• Distributed Systems

Background:

• Federated health data systems face challenges in data quality assessment due to privacy regulations preventing raw data access.
• Centralized data quality assessment methods conflict with data sovereignty and patient confidentiality principles.

Purpose of the Study:

• To evaluate the utility of federated data quality assessment techniques incorporating differential privacy.
• To develop and demonstrate a proof-of-concept for privacy-preserving data quality assessment in health data.

Main Methods:

• Utilized differential privacy techniques within a federated learning framework.
• Developed a proof-of-concept implementation using synthetic observational medical data.
• Enabled local computation of quality metrics and sharing of aggregated, privacy-protected results.

Main Results:

• Presented a privacy-preserving framework for federated data quality evaluation.
• Demonstrated a proof-of-concept supporting cross-data model quality checks.
• Showcased the ability to gain data quality insights without compromising sensitive health information.

Conclusions:

• Differential privacy effectively enables federated quality assessment in health data networks while preserving individual privacy.
• A proof-of-concept system over synthetic data confirmed the feasibility of obtaining meaningful quality metrics in a decentralized setting.