GlucoFM: A Dual-Stream Foundation Model for Continuous Glucose Monitoring

2026-05-29

AI Summary

This work proposes GlucoFM, a lightweight dual-stream foundation model for continuous glucose monitoring (CGM) data that addresses the limitation of existing approaches treating glucose trajectories as a single time series by explicitly modeling their inherent dual physiological dynamics. GlucoFM aligns irregular CGM readings to a 24-hour grid and decomposes them into a slow-varying physiological state stream capturing baseline glycemia and a transient event stream representing short-term fluctuations, incorporating this physiology-aware decomposition as an inductive bias. During self-supervised pretraining, it jointly employs masked context prediction and dual-stream dynamic prediction. Evaluated across four cohorts and seven clinical tasks, GlucoFM achieves state-of-the-art linear probing performance, improving average PR-AUC by 4.1 points over the best existing CGM model, demonstrates superior accuracy in predicting diabetes risk and β-cell dysfunction, and exhibits strong cross-dataset transferability and few-shot adaptation capabilities.

Abstract

Continuous glucose monitoring (CGM) provides a dense view of daily metabolic physiology, yet existing generic time-series and CGM-specific foundation models often encode glucose traces as entangled single-stream sequences, leaving the distinct temporal structure of glycemic dynamics only implicitly modeled. We present GlucoFM, a lightweight CGM foundation model that aligns irregular recordings to a 24-hour chronological grid, preserves observation masks, and decomposes glucose dynamics into slow physiological state and transient event streams, capturing low-frequency glycemic baselines and short-term deviations that may reflect acute physiological responses or sensor artifacts. GlucoFM is pretrained on 109,066 hours of unlabeled CGM recordings from 477 subjects with two complementary objectives: masked contextual latent prediction over fused daily representations and temporal dynamics prediction over state and event streams. Across four diverse cohorts and seven clinical prediction tasks, GlucoFM achieves the strongest subject-disjoint linear-probing performance among evaluated baselines, improving average PR-AUC by 4.1 points over the best CGM-specific foundation model. Its gains are most pronounced on core metabolic outcomes, leading PR-AUC on all diabetes-risk and $β$-cell dysfunction tasks and on 3 of 4 insulin-resistance tasks. GlucoFM also achieves the best overall cross-dataset transfer performance and strong few-shot adaptation among evaluated methods, and consistent gains when aggregating multiple days for subject-level prediction, highlighting physiology-aware decomposition as an effective inductive bias for transferable CGM representation learning.

Problem

• continuous glucose monitoring
• glycemic dynamics
• time-series modeling
• physiological state
• event decomposition

Innovation

• dual-stream architecture
• physiology-aware decomposition
• continuous glucose monitoring
• foundation model
• temporal dynamics modeling