A Federated Record Linkage Algorithm for Secure Medical Data Sharing.

Area of Science:

• Health Informatics
• Data Privacy
• Computer Science

Background:

• Consolidating medical records from multiple institutions necessitates robust privacy-preserving record linkage (PPRL).
• Existing PPRL methods often limit linkage to two institutions or discard non-matching records, resulting in incomplete datasets.
• There is a need for advanced PPRL techniques that support multi-party linkage while ensuring data privacy and completeness.

Purpose of the Study:

• To propose a novel algorithm for federated record linkage among multiple parties.
• To enhance patient data privacy during medical record consolidation using record-level bloom filters.
• To develop an end-to-end pseudonymization framework for secure medical data sharing.

Main Methods:

• Development of a new federated record linkage algorithm facilitated by a trusted third party.
• Utilization of record-level bloom filters to maintain patient data privacy.
• Optimization of linkage-relevant data fields through a dedicated study to determine optimal weights.

Main Results:

• Achieved 99.5% linkage accuracy in tests conducted on the Febrl record linkage dataset.
• Demonstrated the effectiveness of the proposed algorithm in a multi-party federated setting.
• Successfully integrated the algorithm into a comprehensive pseudonymization framework.

Conclusions:

• The proposed federated record linkage algorithm effectively addresses limitations of existing PPRL methods.
• The use of bloom filters ensures strong patient data privacy in multi-institutional record linkage.
• This approach facilitates more complete and accurate medical data sharing through pseudonymization.