Privacy-preserving record linkage on large real world datasets.

Area of Science:

• Health Informatics
• Data Privacy
• Population Health Research

Background:

• Traditional record linkage relies on personally identifying information (PII), posing privacy risks despite data custodian safeguards.
• Residual disclosure risks of sensitive health information remain a concern for data custodians.
• Existing methods separate PII from clinical data but do not eliminate all privacy concerns.

Purpose of the Study:

• To trial a novel record linkage method that further reduces privacy risks on large, real-world administrative health data.
• To evaluate the effectiveness of using encrypted personal identifying information (bloom filters) within a probabilistic linkage framework.
• To compare the linkage quality of the privacy-preserving method against traditional probabilistic methods.

Main Methods:

• A probability-based record linkage framework was employed.
• Encrypted personal identifying information, specifically bloom filters, were used to represent PII.
• The method was tested on over 26 million hospital admission records from New South Wales and Western Australia spanning ten years.

Main Results:

• The privacy-preserving record linkage method demonstrated no difference in linkage quality compared to traditional methods using unencrypted PII.
• The study successfully applied the method to a large-scale, real-world administrative dataset.
• The bloom filter approach effectively reduced privacy risks without compromising linkage accuracy.

Conclusions:

• The tested privacy-preserving record linkage method offers a viable approach to mitigate disclosure risks in population-level research.
• This technique can help alleviate concerns of data custodians regarding sensitive information exposure.
• Further development and adoption of such methods can enhance the realization of benefits from linked health research.