Federated Learning Based Model for Recommendation System Based in Health Care

Abstract

Precision Medicine is an emerging healthcare approach that focuses on tailoring treatments for individual patients. The implementation of patient-centred Decision Support Systems, including Health Recommender Systems, is a key component of this initiative, aimed at augmenting the accuracy and individualization of healthcare delivery. However, a significant challenge in developing these systems is the confidential nature of the medical data, as these systems require large volumes of data to function effectively. Unfortunately, medical data are distributed across multiple institutions and cannot be centralized due to privacy concerns. To overcome this challenge, this position paper presents an architecture that uses Federated Learning to build a HRS (Health Recommender System). Federated Learning enables the use of data from different institutions without requiring direct data sharing. To demonstrate the feasibility of this approach, we developed a Federated Drug Recommendation System designed to assist physicians in prescribing medications by utilizing historical data on disease-drug interactions and pharmaceutical information. As this is a position paper, we focus on presenting a proof-of-concept utilizing publicly available, non-sensitive datasets.