A simple heuristic for blindfolded record linkage.

Area of Science:

• Health Informatics
• Biostatistics
• Data Privacy

Background:

• Cross-site patient registries are crucial for research but face privacy challenges.
• Matching patient data across sites requires balancing accuracy with privacy regulations like HIPAA.
• Existing linkage algorithms can be complex and cumbersome.

Purpose of the Study:

• To develop and evaluate a privacy-preserving method for generating anonymous patient identifiers for cross-site data linkage.
• To maximize accurate identification of shared patients while minimizing false positives.
• To assess the efficacy of a novel matching strategy against traditional identifiers.

Main Methods:

• A hashed string variable was created using the first two letters of patients' first and last names, combined with their date of birth.
• The performance of this derived match variable was compared against Social Security Numbers (SSN) and full names.
• Match algorithm characteristics, specifically the trade-off between false positive and false negative rates, were analyzed.

Main Results:

• The derived match variable achieved a sensitivity of 97% in a dataset of 19,000 records.
• In contrast, anonymized identifiers based on full names and date of birth had 87% sensitivity, and SSN had 86% sensitivity.
• The proposed method demonstrated superior accuracy in identifying shared patients.

Conclusions:

• The described approach is highly effective for situations where strict privacy policies limit data exchange.
• This method is easier to explain to institutional review boards and complies with the HIPAA privacy rule's minimum necessary standard.
• It offers a faster and less cumbersome alternative to complex probabilistic linkage algorithms for high-quality datasets.