Efficient and effective pruning strategies for health data de-identification.

Area of Science:

• Biomedical Informatics
• Data Privacy
• Computer Science

Background:

• Protecting sensitive biomedical data is crucial for research.
• Data de-identification balances privacy risks and data utility.
• Existing methods struggle with scalability for high-dimensional datasets.

Purpose of the Study:

• Develop a novel, scalable method for de-identifying biomedical data.
• Address limitations of current de-identification techniques for complex datasets.
• Improve the efficiency of privacy-preserving data transformations.

Main Methods:

• Combined antichains with prefix tree-inspired data structures.
• Integrated the novel method into existing de-identification algorithms.
• Implemented a best-first branch and bound search (BFS) algorithm.

Main Results:

• Reduced memory by up to 10x and execution time by 25% for low-dimensional data.
• Increased processed solution space size by nearly 10x.
• Outperformed state-of-the-art algorithms by 12% in output data quality for high-dimensional data.

Conclusions:

• The novel approach enhances scalability for de-identifying real-world biomedical datasets.
• Successfully addresses the challenge of de-identifying large, complex datasets.
• The method is implemented in ARX, an open-source de-identification software.