Protecting Record Linkage Identifiers Using a Language Model for Patient Names.

Area of Science:

• Medical Informatics
• Cryptography
• Data Security

Background:

• Linking information across databases is crucial for medical research.
• European privacy regulations mandate encryption of personal identifiers for data linkage.
• Bloom filter methods are popular for record linkage but vulnerable to attacks.

Purpose of the Study:

• To present a novel hardening method for Privacy-Preserving Record Linkage (PPRL).
• To enhance the security of Bloom filter-based record linkage against cryptographic attacks.
• To evaluate the proposed method's effectiveness using real-world mortality data.

Main Methods:

• Developed a Privacy-Preserving Record Linkage (PPRL) technique.
• Implemented a Markov chain-based language model for encrypting identifier bigrams.
• Compared the proposed method against unencrypted and existing state-of-the-art PPRL techniques.

Main Results:

• The proposed Markov chain-based hardening method significantly increases protection against cryptographic attacks.
• Empirical evaluation on mortality data demonstrates superior security compared to basic Bloom filters.
• The method maintains effective record linkage while enhancing data privacy.

Conclusions:

• The novel hardening method offers a robust solution for secure Privacy-Preserving Record Linkage.
• This advancement is vital for complying with privacy regulations and enabling secure medical data sharing.
• The technique provides a strong defense against sophisticated cryptographic attacks in record linkage.