Maximal Local Privacy Loss-A New Method for Privacy Evaluation of Synthetic Datasets.

Area of Science:

• Medical Informatics
• Data Privacy
• Biostatistics

Background:

• Synthetic patient data offers privacy-preserving research opportunities.
• Current privacy assessment methods are often inadequate, either missing key privacy aspects or being overly cautious.
• Accurate privacy evaluation is crucial for compliance and ethical research.

Purpose of the Study:

• To introduce a novel approach for evaluating the privacy of synthetic datasets.
• To measure maximal local privacy loss to detect potential record reconstruction.
• To ensure acceptable privacy risks for all individuals in synthetic datasets.

Main Methods:

• Developed a new privacy evaluation strategy based on maximal local privacy loss.
• Measured individual contributions to synthetic dataset generation likelihood.
• Applied the method to synthetic time-to-event data derived from real cancer registry data using sequential regressions and a flexible parametric survival model.

Main Results:

• The proposed method effectively measures information leakage at an individual level.
• Demonstrated the ability to detect possibilities of reconstructing original data records.
• The approach provides a quantifiable measure of privacy risk for each patient.

Conclusions:

• The novel maximal local privacy loss method offers a more accurate and less overly cautious approach to synthetic data privacy assessment.
• This method enhances the trustworthiness of synthetic patient data for medical research.
• Ensures robust privacy protection for individuals contributing to original datasets.