Using Differential Privacy and Federated Learning to Enhance Health Data Protection
Open Access DepositedDownloadable Content
The Health Insurance Portability and Accountability Act (HIPAA) Privacy Rule has been a standard for determining the protections for health patient privacy for decades. The protections defined by the experts consulted for the HIPAA Privacy Rule state that no protective measures are absolute. This is understandable, as the cost of health data breaches has surpassed that of data breaches in any other industry in the United States. In 2024 the health data breach average cost is 9.77 million dollars per breach (Alder, S., 2024). According to cyber security experts, the threat of recurring breaches can be mitigated using privacy enhancement techniques like federated learning and differential privacy. The feasibility and affordability of using these techniques remain subjects of ongoing debate (Ziller, A., Mueller, T. T., et, al., 2024). The choice of whether or not to use machine learning and data science techniques to enhance health data privacy has been considered an expert determination by HIPAA; however, to mitigate the breaches caused by using only de-identifying methods to protect patient information, there is a need for a non-expert to be able to understand the impact of these methods and consider the use of these enhancements as well. This praxis explores the comparison between federated learning and federated learning with differential privacy as a means of health data protection and contributes a decision support tool: The Federated Learning Algorithm Recommendation & Evaluation with Differential Privacy Decision Tool, or FLAREDP, to assist not only data scientists and machine learning experts but also healthcare professionals with the choice of protection techniques for their research or practice.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.