Transforming Personal Health AI: Integrating Knowledge and Causal Graphs with Large Language Models

Abstract

Large Language Models (LLMs) hold considerable promise for healthcare applications, leveraging vast, diverse datasets to deliver insights across a broad range of tasks. However, their effectiveness in personal health settings is currently limited by their dependence on unstructured information, leading to issues concerning accuracy, trustworthiness, and personalization. From this perspective, we propose a transformative framework that integrates knowledge graphs (KGs) and causal graphs (CGs) with LLMs to tackle these challenges. KGs contribute structured, verifiable knowledge that grounds LLM outputs in validated information, while CGs delineate causal relationships that are crucial for precise health assessments and intervention planning. We illustrate this framework with a practical example in diabetes management to show its real-world application. We indicate that integrating KGs and CGs into LLMs is a pivotal advancement for addressing key challenges in personal healthcare. This integration directly tackles issues of trustworthiness, truthfulness, and personalization by anchoring LLM in such structured knowledge. Practical solutions can be deployed using this integrated approach with the support of personal health data.