Clinical Report: AI Model Finds Hidden Risk Signals in CGM Data

Overview

The GluFormer AI model, trained on over 10 million CGM measurements, significantly outperforms traditional metrics like HbA1c in predicting glycemic parameters and long-term diabetes risk. Its ability to identify high-risk individuals, particularly in prediabetes, highlights the potential for enhanced precision in diabetes care.

Background

Continuous glucose monitoring (CGM) has become integral in diabetes management, yet traditional metrics often fail to capture the full spectrum of glycemic dynamics. The development of advanced AI models like GluFormer represents a significant step towards utilizing CGM data more effectively, potentially transforming risk stratification and management in diabetes and cardiometabolic diseases.

Data Highlights

Key Findings

• The GluFormer model was trained on CGM data from 10,812 adults, primarily non-diabetic.
• It outperformed HbA1c and conventional CGM metrics in predicting glycemic parameters across 19 independent cohorts.
• GluFormer effectively identified individuals at high risk for clinically significant HbA1c increases over two years.
• In a long-term follow-up, it stratified diabetes and cardiovascular mortality risk better than HbA1c.
• Integration of dietary data with GluFormer improved predictions of individualized glycemic responses.

Clinical Implications

Clinicians may consider incorporating AI models like GluFormer into routine practice to enhance risk assessment and management strategies for patients with prediabetes and diabetes. This approach could lead to more personalized treatment plans and improved patient outcomes.

Conclusion

The findings from the GluFormer study underscore the potential of AI in revolutionizing diabetes care by leveraging CGM data for better risk stratification and management. This advancement may facilitate more effective interventions and improved patient outcomes.

References

1. Guy Lutsker, et al., Nature, 2023 -- A foundation model for continuous glucose monitoring data
2. AACE Endocrine AI, 2026 -- AI tool predicts hypoglycemia risk pre-exercise
3. Conexiant, 2026 -- AI Carb Estimates From ChatGPT a Glycemic Risk?
4. AACE Endocrine AI, 2026 -- AACE 2026: AI moves from hype to reality in diabetes care
5. Diabetes Technology: Standards of Care in Diabetes—2026 - PMC
6. The Journal of Clinical Endocrinology & Metabolism — Artificial Intelligence Model for Predicting Large-for-Gestational-Age Infants in Pregnant Women with Gestational Diabetes Mellitus
7. 7. Diabetes Technology: Standards of Care in Diabetes—2026 - PMC
8. Effect of Intensive Therapy on the Microvascular Complications of Type 1 Diabetes Mellitus - PMC
9. A foundation model for continuous glucose monitoring data | Nature