SecureSim: A Transparent Framework for Privacy-Preserving Relational Data Synthesis

Access to sensitive medical data is critical for research and innovation, yet stringent privacy regulations and security concerns impose significant barriers to collaboration. Traditional de-identification and anonymization techniques remain vulnerable to re-identification risks, while Controlled Data Access Platforms (CDAPs) introduce operational friction. To address these challenges, we present SecureSim, a privacy-preserving synthetic data generator designed specifically for relational databases. Unlike black-box generative models, SecureSim adopts a white-box approach, ensuring transparency, auditability, and differential privacy compliance. It enables configurable fidelity levels, allowing users to generate synthetic datasets that retain key statistical properties of real patient data while minimizing disclosure risks. We demonstrate the application of SecureSim using publicly available datasets, highlighting its effectiveness in maintaining data utility while safeguarding privacy. Our findings underscore the need for explainable, scalable, and domain-specific synthetic data generation methods, paving the way for enhanced data accessibility in healthcare research and beyond.