Course contentsShow
← Security
Lesson 2930 of 3,05254. Privacy EngineeringFree lesson

Privacy-Preserving Record Linkage

Matching records across datasets while protecting sensitive identifiers and attributes from disclosure.

Privacy-Preserving Record Linkage

What you'll learn: How to match records belonging to the same entity across separate datasets without revealing the underlying identities or sensitive attributes.

The Challenge

Organizations frequently need to determine which records in different databases refer to the same person, patient, or customer—without actually sharing the raw identifiable data. Imagine two hospitals wanting to identify shared patients for research, but neither can legally share patient names or addresses with the other. Traditional record linkage would require pooling identifiable data, creating massive privacy risks.

How It Works

Privacy-Preserving Record Linkage (PPRL) applies cryptographic and privacy techniques you've already learned to enable matching without disclosure:

Core Techniques

Hashing and Bloom Filters: Instead of sharing plaintext identifiers, parties hash names, birthdates, or addresses into fixed-length Bloom filter bit arrays. Similar records produce similar bit patterns, enabling approximate matching without revealing the original values—a form of pseudonymization combined with similarity-preserving encoding.

Secure Multi-Party Computation: Organizations can jointly compute which records match using SMPC protocols, where each party keeps their data encrypted and only learns the match outcome, not the other party's raw identifiers.

Differential Privacy Integration: Adding calibrated noise to matching statistics prevents membership inference while still identifying genuine links.

Trusted Third Parties or TEEs: A neutral entity or Trusted Execution Environment performs the linkage inside a secure enclave, ensuring neither party sees the other's data in plaintext.

The Privacy-Utility Balance

Strong privacy protections (heavy noise, coarse hashing) reduce linkage accuracy. Weaker protections improve matches but increase re-identification risk—requiring careful sensitivity analysis and privacy budget management, just like differential privacy applications.

Key Takeaway:
Privacy-preserving record linkage uses cryptographic hashing, secure computation, and noise addition to match records across datasets while protecting identities—enabling collaborative analytics without compromising individual privacy.