Reducing patient re-identification risk for laboratory results within research datasets

Area of Science:

• Biomedical Informatics
• Health Data Privacy
• Clinical Data Management

Background:

• Biomedical research increasingly relies on shared clinical data.
• Protecting patient privacy in de-identified databases is a significant challenge.
• Existing de-identification methods may impact data utility.

Purpose of the Study:

• To develop and evaluate methods for reducing patient re-identification risks in laboratory test result databases.
• To minimize alterations to clinical data interpretation during the de-identification process.

Main Methods:

• A threat model simulating an attacker using laboratory results as a search key was employed.
• Two data perturbation models were tested: simple random offsets and an expert-derived clinical meaning-preserving model.
• Re-identification risk and clinical meaning retention were assessed using a rank-based re-identification algorithm on a large de-identified dataset.

Main Results:

• The expert-derived algorithm preserved clinical meaning better than simple random offsets, affecting only 4% of test results compared to 26%.
• Differences in re-identification rates between the algorithms were minimal.
• The study demonstrated a trade-off between data perturbation levels and clinical meaning retention.

Conclusions:

• Expert-derived data perturbation offers a practical approach to mitigate re-identification risks in biomedical databases.
• This method balances the need for data privacy with the retention of clinical data utility for research.
• The findings support the development of adaptable data protection schemes for shared health information.