Architecting Federated Learning Frameworks for Real- Time Multi-Hospital Chronic Disease Trajectory Prediction without Compromising HIPAA Privacy

• Sunday Sunday

  Ladoke Akintola University of Technology

  Author

The increasing prevalence of chronic diseases necessitates predictive models that leverage distributed healthcare data across multiple institutions. However, centralized machine learning approaches face significant barriers due to patient privacy concerns, regulatory constraints under the Health Insurance Portability and Accountability Act (HIPAA), and institutional data silos. This research addresses the critical gap between the need for robust multi-institutional chronic disease prediction and the imperative of stringent privacy preservation. We propose and validate a federated learning framework that enables real-time trajectory prediction for chronic conditions including diabetes and hypertension across decentralized hospital networks. Our methodology integrates differential privacy with Paillier homomorphic encryption and an adaptive node weighting mechanism to ensure secure aggregation of model updates while maintaining compliance with HIPAA and GDPR standards. Evaluated on the MIMIC-III clinical database across simulated multi-hospital environments, the proposed framework achieved an overall prediction accuracy of 92.0% with an AUC-ROC of 0.94 for chronic disease trajectory prediction, statistically validated through paired t-tests (p < 0.01). Privacy assessments demonstrated a reduction in membership inference risk from 20% to 5%, confirming robust privacy preservation. The framework achieved a 41.6% reduction in communication overhead compared to baseline federated approaches. These findings establish that federated learning architectures can effectively balance predictive performance with strict privacy guarantees, offering a scalable and replicable solution for collaborative healthcare analytics that does not compromise patient confidentiality. The framework provides actionable pathways for healthcare administrators seeking to implement privacy-preserving AI systems across institutional boundaries.

• pdf

Copyright (c) 2026 Sunday Sunday (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.