Personalized Diabetes Management Using Large Language Models and CGM Data

Abstract

Objectives: To develop a novel framework that leverages Large Language Models (LLMs) for generating personalized, context-aware diabetes management recommendations using Continuous Glucose Monitoring (CGM) data combined with patient lifestyle logs.

Methods: The proposed system transforms structured CGM data into natural language text and generates synthetic contextual log data, including meal descriptions and activity levels, based on health metrics. These textual inputs are then integrated and processed by an LLM to produce individualized, actionable recommendations tailored to each patient's unique glucose patterns and lifestyle context.

Results: The framework was evaluated through expert review, which assessed the clinical relevance, clarity, and practicality of the generated recommendations. The findings suggest that the model can provide coherent, understandable, and personalized guidance that may support improved self-management of diabetes.

Conclusion: This study demonstrates the potential of LLMs to enhance personalized diabetes care by converting complex medical data into accessible, patient-centric recommendations. The integration of CGM data with LLMs represents a promising direction for advancing intelligent, user-friendly digital health interventions.