Preserving information while respecting privacy through an information theoretic framework for synthetic health data...

Area of Science:

• Medical Informatics
• Data Science
• Biostatistics

Background:

• Generating synthetic medical data is challenging due to privacy concerns.
• Existing methods often fail to jointly evaluate data quality and privacy.
• Assessing multivariate associations is crucial for accurate synthetic data.

Purpose of the Study:

• Introduce a novel algorithm (MIIC-SDG) for synthetic data generation from electronic health records.
• Propose a new metric (QPS) to quantify the quality-privacy trade-off.
• Demonstrate the performance of MIIC-SDG against state-of-the-art methods.

Main Methods:

• Developed MIIC-SDG based on multivariate information theory and Bayesian networks.
• Proposed a Quality and Privacy Score (QPS) metric for evaluation.
• Validated MIIC-SDG on diverse clinical datasets.

Main Results:

• MIIC-SDG effectively generates synthetic data preserving multivariate associations.
• The QPS metric provides a quantitative measure of the quality-privacy balance.
• MIIC-SDG demonstrates superior performance compared to existing methods.

Conclusions:

• MIIC-SDG offers a robust approach for generating high-fidelity, privacy-preserving synthetic medical data.
• The QPS metric facilitates objective comparison of synthetic data generation techniques.
• This work advances the field of secure and reliable health data synthesis.