Evaluating privacy-preserving record linkage using cryptographic long-term keys and multibit trees on large medical...

Area of Science:

• Health Informatics
• Data Security
• Medical Research

Background:

• Record linkage integrates diverse medical data but faces challenges with unique personal identifiers.
• Privacy regulations necessitate encryption of sensitive attributes for record linkage.
• Privacy-Preserving Record Linkage (PPRL) techniques, including Bloom filters and composite Bloom filters (CLKS), are gaining traction.

Purpose of the Study:

• To evaluate the real-world performance of an innovative PPRL technique using CLKS with multibit trees.
• To compare the linkage time and quality of CLKS-based PPRL against traditional clear-text probabilistic record linkage.

Main Methods:

• Utilized a large subset of Australian hospital admission data.
• Implemented and tested an innovative PPRL technique employing CLKS with multibit trees.
• Evaluated linkage performance based on linkage time, recall, precision, and F-measure.

Main Results:

• Clear-text probabilistic linkage achieved marginally higher precision and recall than CLKS-based PPRL.
• PPRL required increased computing time, though 5 million records were de-duplicated within a day.
• The PPRL approach necessitated fine-tuning of parameters for optimal performance.

Conclusions:

• Enhanced privacy in PPRL is associated with slight reductions in precision/recall and a significant increase in computational burden and setup time.
• The trade-offs of PPRL appear acceptable for most applications, but require careful consideration.
• Further research is needed to develop optimal automatic parameter selection for PPRL.