Enabling the Informed Patient Paradigm with Secure and Personalized Medical Question Answering

Abstract

Quality patient care is a complex and multifaceted problem requiring the integration of data from multiple sources. We propose Medicient, a knowledge-graph-based question answering system that processes heterogeneous data sources, including patient health records, drug databases, and medical literature, into a unified knowledge graph with zero training. The knowledge graph is then utilized to provide personalized recommendations for treatment or medication. The system leverages the power of large language models for question understanding and natural language response generation, while hiding sensitive patient information. We compare our system to a large language model (ChatGPT), which does not have access to patient health records, and show that our system provides better recommendations. This study contributes to a growing body of research on knowledge graphs and their applications in healthcare.