Diversity-Aware Anonymization for Structured Health Data.

Area of Science:

• Health Informatics
• Data Privacy
• Machine Learning

Background:

• Hospital patient data holds significant analytical potential but is restricted due to privacy concerns.
• Sharing health data across institutions is crucial for valuable insights but faces legal and privacy barriers.
• Existing anonymization techniques struggle to address sophisticated linkage attacks.

Purpose of the Study:

• To propose a novel anonymization method for secure sharing of patient health data.
• To address both record-linkage and attribute-linkage attack models.
• To balance data privacy with data utility for analysis.

Main Methods:

• Formulated data anonymization as a constrained optimization problem.
• Integrated k-anonymity, l-diversity, and t-closeness privacy models.
• Developed a novel technique for anonymizing and sharing sensitive health records.

Main Results:

• The proposed method effectively anonymizes patient data.
• Evaluated the trade-off between data utility and privacy preservation.
• Demonstrated improved privacy against linkage attacks compared to existing methods.

Conclusions:

• The novel constrained optimization approach offers a robust solution for health data anonymization.
• This method facilitates secure data sharing for enhanced healthcare analytics.
• The technique effectively preserves data utility while ensuring strong privacy guarantees.