[Paper Review] From Glucose Patterns to Health Outcomes: A Generalizable Foundation Model for Continuous Glucose Monitor Data Analysis

Guy Lutsker, Gal Sapir+12 more|arXiv (Cornell University)|Aug 20, 2024

Diabetes Management and Research10 citations

TL;DR

The paper introduces GluFormer, a generalizable foundation model for CGM data that learns glycemic patterns and predicts broader health outcomes, with strong cross-cohort generalization and risk stratification capabilities, including dietary multimodality.

ABSTRACT

Recent advances in SSL enabled novel medical AI models, known as foundation models, offer great potential for better characterizing health from diverse biomedical data. CGM provides rich, temporal data on glycemic patterns, but its full potential for predicting broader health outcomes remains underutilized. Here, we present GluFormer, a generative foundation model for CGM data that learns nuanced glycemic patterns and translates them into predictive representations of metabolic health. Trained on over 10 million CGM measurements from 10,812 adults, primarily without diabetes, GluFormer uses autoregressive token prediction to capture longitudinal glucose dynamics. We show that GluFormer generalizes to 19 external cohorts (n=6,044) spanning different ethnicities and ages, 5 countries, 8 CGM devices, and diverse pathophysiological states. GluFormers representations exceed the performance of current CGM metrics, such as the Glucose Management Indicator (GMI), for forecasting clinical measures. 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. We also show that CGM representations from pre-intervention periods in Randomized Clinical Trials outperform other methods in predicting primary and secondary outcomes. When integrating dietary data into GluFormer, we show that the multi-modal version of the model can accurately generate CGM data based on dietary intake data, simulate outcomes of dietary interventions, and predict individual responses to specific foods.

Motivation & Objective

• Develop a generalizable foundation model (GluFormer) for continuous glucose monitor (CGM) data to capture nuanced glycemic patterns.
• Translate CGM dynamics into predictive representations of metabolic health.
• Demonstrate cross-cohort generalization across diverse populations, devices, and pathophysiologies.
• Show utility in long-term risk stratification for diabetes and cardiovascular events.
• Explore multi-modal extensions integrating dietary data to simulate interventions and responses.

Proposed method

• Autoregressive token prediction to model longitudinal CGM dynamics.
• Training on >10 million CGM measurements from 10,812 adults (primarily non-diabetic).
• Evaluation on 19 external cohorts (n=6,044) across ethnicities, ages, countries, devices, and conditions.
• Comparison of GluFormer representations to traditional CGM metrics (e.g., GMI) for outcome forecasting.
• Longitudinal analysis with 580 participants and 12-year follow-up to assess diabetes risk prediction.
• Multi-modal extension integrating dietary data to generate CGM from diet and simulate dietary interventions.

Experimental results

Research questions

• RQ1 Can a foundational model trained on CGM data generalize to diverse external cohorts and predict broad metabolic health outcomes?
• RQ2 Do GluFormer representations outperform traditional CGM metrics (like GMI) in forecasting clinical measures?
• RQ3 Can CGM-based representations stratify risk for incident diabetes and cardiovascular death beyond HbA1c and standard metrics?
• RQ4 Does pre-intervention CGM data from randomized trials improve prediction of trial outcomes?
• RQ5 Can dietary information be integrated to generate CGM, simulate dietary interventions, and predict individual dietary responses?

Key findings

• GluFormer generalizes to 19 external cohorts (n=6,044) across multiple countries, devices, and populations.
• GluFormer representations exceed the performance of the Glucose Management Indicator (GMI) for forecasting clinical measures.
• In a 12-year longitudinal study (n=580), top-quartile CGM-based risk identified 66% of new diabetes cases vs 7% in bottom quartile.
• Top-quartile CGM representations captured 69% of cardiovascular-death events with none in the bottom quartile.
• Pre-intervention CGM representations in randomized trials outperform other methods in predicting primary and secondary outcomes.
• A multi-modal version integrating dietary data can generate CGM from diet, simulate dietary interventions, and predict individual food responses.