New Method for Privacy-Preserving Record Linkage Introduced
/ 4 min read
Quick take - Researchers from the University of Tübingen have developed a new method for Privacy-Preserving Record Linkage (PPRL) that enhances medical data privacy and significantly improves the efficiency and accuracy of linking healthcare records across multiple sources while adhering to privacy regulations.
Fast Facts
- Researchers from the University of Tübingen developed a new Privacy-Preserving Record Linkage (PPRL) method to enhance medical data privacy, addressing challenges posed by GDPR and HIPAA regulations.
- The proposed method utilizes a secure three-party computation framework, allowing efficient record linkage without exposing sensitive data, achieving speeds up to 14 times faster than existing solutions.
- The method eliminates vulnerabilities associated with traditional techniques like Bloom filters by employing a bigram mapping approach for string similarity, enhancing both security and performance.
- Evaluations showed the new method significantly reduces errors in linkage quality, achieving an optimal threshold score of 0.7852 and completing large dataset tasks in a fraction of the time compared to current methods.
- The study, supported by the German Ministry of Research and Education, provides accessible datasets and code in the PPRL repository on GitHub for further research and application.
Novel Method for Privacy-Preserving Record Linkage in Healthcare
A recent study by researchers from the Department of Computer Science at the University of Tübingen, Germany, introduces a novel method for Privacy-Preserving Record Linkage (PPRL) aimed at enhancing medical data privacy. The study, authored by Şeyma Selcan Mağara, Noah Dietrich, Ali Burak Ünal, and Mete Akgün, highlights the importance of record linkage in integrating healthcare data from multiple sources. This is particularly crucial in situations where datasets lack exact identifiers.
Addressing Privacy Regulations
The paper addresses the need for PPRL in healthcare, where privacy regulations like the General Data Protection Regulation (GDPR) and the Health Insurance Portability and Accountability Act (HIPAA) restrict the sharing of personally identifiable information without explicit consent. These regulations present challenges in maintaining privacy and efficiency during large-scale record linkage processes.
To address these challenges, the authors propose a new PPRL method based on a secure three-party computation (MPC) framework. This approach allows multiple parties to compute linkage results without exposing sensitive private inputs. The proposed method significantly improves the speed of the linkage process, achieving performance enhancements of up to 14 times faster than existing solutions. For example, linking a record against a database of 10,000 records can be completed in 8.74 seconds on a network with 700 Mbps bandwidth and 60 ms latency. On a slower connection of 100 Mbps bandwidth, the process takes 28 seconds.
Enhancing Security and Performance
The study explains how record linkage can identify records referring to the same entity across different datasets, even without common identifiers. Practical applications include linking cancer registry data with treatment records to understand disease progression across hospitals. However, the presence of sensitive personal information complicates the record linkage process and raises privacy concerns.
The authors critique existing PPRL methods, which often rely on encryption and algorithms to conduct record linkage without revealing identities. They highlight vulnerabilities in commonly used techniques, such as Bloom filters, which can be susceptible to cryptanalysis and frequency attacks. The proposed method eliminates the need for Bloom filters by using a bigram mapping approach for string similarity, enhancing security and performance.
The three-party MPC framework involves two proxies holding secret shares of data and a helper party assisting in computations. This enhances efficiency and reduces overhead compared to traditional two-party systems. The method also supports linkages among multiple data owners, providing greater flexibility for practical applications.
Evaluation and Results
The researchers conducted thorough evaluations of the proposed method, assessing timing and linkage quality against established state-of-the-art methods. Experiments were performed on Google Cloud machines under varying network conditions, using synthetic datasets with various data corruption techniques. Linkage quality was evaluated based on false positive and false negative rates against different threshold values.
The proposed method significantly outperformed existing techniques in minimizing total errors, achieving an optimal threshold score of 0.7852 compared to 0.7468 for the state-of-the-art method. It also demonstrated a higher ROC AUC value, indicating improved accuracy. The runtime analysis highlighted considerable efficiency gains, especially for larger datasets. For instance, a 10,000-by-10,000 record matching task was completed in 24.2 hours using the proposed method, a substantial reduction from the 256.44 hours required by the current leading method.
The study was supported by the German Ministry of Research and Education (BMBF). Both the datasets and the code related to the research are accessible in the PPRL repository on GitHub.
Original Source: Read the Full Article Here
