From Glucose Patterns to Health Outcomes: A Generalizable Foundation Model for Continuous Glucose Monitor Data Analysis
Lutsker G, Sapir G, Shilo S, Merino J, Godneva A, Greenfield J, Samocha-Bonet D, Dhir R, Gude F, Mannor S, Meirom E, Chechik G, Rossman H, Segal E, arxiv
Paper summary
The paper describes the GluFormer model that was developed based on CGM data from 10,812 non-diabetic individuals from the HPP dataset. The AI model is from the class of transformer-based generative models, similar to ChatGPT (GPT = Generative, Pretrained, Transformer). For each glucose measurement, it provides a prediction for the next point in time. The model offers insights into metabolic health, predicting health outcomes 4 years in advance, outperforming state-of-the-art CGM analysis tools. It can be used for different downstream tasks, for example to forecast outcomes of clinical trials or to simulate glucose response to diet. Most important numbers:
In a longitudinal study of 580 adults with CGM data and 12-year follow-up, GluFormer identifies individuals at elevated risk of developing diabetes more effectively than blood HbA1C%, capturing 66% of all new-onset diabetes diagnoses in the top quartile versus 7% in the bottom quartile.
Similarly, 69% of cardiovascular-death events occurred in the top quartile with none in the bottom quartile, demonstrating powerful risk stratification beyond traditional glycemic metrics.
