Combining Different Privacy-Preserving Record Linkage Methods for Hospital Admission Data

Area of Science:

• Health Informatics
• Data Science
• Biostatistics

Background:

• Accurate patient identification is crucial for healthcare and research.
• Existing record linkage (RL) methods face challenges in privacy preservation.
• Hospital admission data requires robust and secure linkage techniques.

Purpose of the Study:

• To evaluate the effectiveness of privacy-preserving record linkage methods on hospital admission data.
• To assess the performance of deterministic RL (DRL), probabilistic RL (PRL), and Bloom filters in identifying unique patients.
• To determine the suitability of combined RL methods for population-based research.

Main Methods:

• Applied deterministic RL (DRL), probabilistic RL (PRL), and Bloom filters to hospital admission data under privacy-preserving conditions.
• Utilized one-way encryption for patient characteristics (names) in DRL and PRL.
• Transformed patient characteristics into cryptographic long-term keys for Bloom filters.
• Split a dataset of one year's hospital admissions into new and known patient cohorts (30,000 new, 1.5 million known).

Main Results:

• A combination of the three RL methods achieved a positive predictive value of 83% (95% CI: 65%-94%).
• Privacy-preserving techniques were successfully applied to patient data.
• The linkage methods demonstrated effectiveness in a large-scale hospital admission dataset.

Conclusions:

• The presented combination of privacy-preserving record linkage methods is effective for identifying unique patients in hospital data.
• This approach shows promise for enhancing population-based research requiring accurate data linkage.
• The study validates the utility of integrating DRL, PRL, and Bloom filters for secure and efficient record linkage.