Linkage Attacks Expose Identity Risks in Public ECG Data Sharing

Area of Science:

• Biomedical Informatics
• Data Privacy
• Cardiovascular Health

Background:

• Publicly available electrocardiogram (ECG) data presents significant privacy challenges.
• Biometric properties of ECGs make individuals susceptible to re-identification and linkage attacks.
• Existing privacy assessments often assume unrealistic adversarial capabilities.

Purpose of the Study:

• To evaluate ECG privacy risks under realistic conditions with partial adversarial knowledge.
• To assess the effectiveness of current anonymization techniques against sophisticated re-identification.
• To determine the feasibility of identity linkage using real-world ECG datasets.

Main Methods:

• Utilized diverse, real-world ECG datasets from 109 participants.
• Simulated realistic adversarial scenarios with partial knowledge.
• Developed and applied an approach to re-identify individuals in public datasets.

Main Results:

• Achieved 85% accuracy in re-identifying individuals in public ECG datasets.
• Reported a 14.2% overall misclassification rate at an optimal confidence threshold.
• Observed misclassification rates of 15.6% (unknown to known) and 12.8% (known to unknown).

Conclusions:

• Simple anonymization methods are insufficient for protecting ECG data privacy.
• Partial adversarial knowledge significantly enhances the risk of identity linkage.
• Urgent implementation of advanced privacy-preserving strategies like differential privacy is required for shared biosignal data.