A Comparative Analysis on Deep Learning Approaches for Patient Data Privacy Using Electronic Health Record

The widespread adoption of electronic health records (EHRs) across virtually all U.S. hospitals has facilitated the accumulation of vast amounts of unstructured medical data, which, while invaluable for advancing healthcare, also pose significant privacy risks. This paper presents a comparative analysis of various deep learning strategies aimed at bolstering the privacy of EHR data before it is made publicly accessible. We introduce and evaluate two innovative deep learning models on the MIMIC_III dataset, measuring their performance in terms of accuracy and F1 scores. Our research employs these models within a framework that emphasizes “patient privacy preservation” (P3) and explores the application of differential privacy techniques using Recurrent Neural Networks (RNNs) and Artificial Neural Networks (ANNs) to secure the sharing of training data. The results demonstrate the efficacy of these advanced methodologies in enhancing the privacy protections of EHR data, thereby offering crucial insights for medical research. Our findings underscore the critical role of privacy-preserving measures in the dissemination of EHR data and validate the potential of sophisticated deep learning approaches to ensure secure data exchange. This study contributes to the broader discourse on reconciling the needs of medical research with the imperative to protect patient confidentiality.