Data Leakage Prevention in Health Insurance: A Comparative Analysis of Differential Privacy Techniques

Abstract

Abstract
Health insurance providers increasingly rely on data-driven decision-making, leveraging large-scale electronic health records (EHRs), claims data, and patient-reported outcomes to enhance underwriting, claims management, and predictive analytics. However, the sensitive nature of patient information makes these datasets prime targets for unauthorized access, inadvertent disclosure, and data leakage, posing significant ethical, regulatory, and operational challenges. Differential Privacy (DP) has emerged as a leading approach to mitigate these risks, offering provable privacy guarantees while preserving data utility for analytical purposes. This paper presents a comprehensive comparative analysis of differential privacy techniques applied to health insurance datasets, evaluating trade-offs between privacy, utility, scalability, and ease of integration. By benchmarking contemporary DP mechanisms including Laplace, Gaussian, and randomized response techniques across real-world health insurance scenarios, we provide actionable insights for insurers seeking to implement robust data leakage prevention frameworks. Our findings indicate that while no single DP mechanism universally dominates across all performance metrics, context-specific hybrid approaches and adaptive privacy budgets can achieve optimal privacy-utility balances. The paper concludes with recommendations for deploying DP in operational insurance systems and outlines future research directions integrating federated learning, synthetic data generation, and explainable AI to strengthen privacy-preserving analytics.