What does this research mean for the field?

Topological Data Analysis of Continuous Glucose Monitoring data identifies distinct glycemic phenotypes—such as saturated versus erratic hyperglycemia—that are conflated by standard metrics like HbA1c. Novelty: ClaimNovelty.NOVEL_FINDING. Consensus alignment: ConsensusAlignment.ESTABLISHES_NEW_DIRECTION.

What question did this study set out to answer?

The aim was to identify distinct glycemic phenotypes from continuous glucose monitoring data using topological methods.

June 7, 2026

1919-P: Topological Phenotyping of Continuous Glucose Monitoring Data Identifies Saturated and Erratic Glycemic Phenotypes Missed by Standard Metrics

Key Points

The aim was to identify distinct glycemic phenotypes from continuous glucose monitoring data using topological methods.
Analyzed CGM data from 1,049 participants in the AI-READI dataset.
Computed pairwise Wasserstein distances between glucose distribution probability across participants.
Applied hierarchical clustering to identify glycemic subgroups and compared clinical characteristics.
Identified four phenotypes with significant clinical differences (HbA1c p<0.0001).
Cluster 1 showed highest mean glucose (342.4 mg/dL) and HbA1c (10.0%), indicating sustained hyperglycemia.
Cluster 2 had highest variability (CV: 23.6%) with notable erratic glucose patterns.

Abstract

Introduction and Objective: Continuous Glucose Monitoring (CGM) generates complex time-series data often reduced to summaries (e.g., Time-in-Range). These metrics fail to capture the distributional shape of glycemic dynamics. This study applied Topological Data Analysis (TDA) and Wasserstein distances to identify intrinsic phenotypes of glycemic magnitude and evaluated their clinical relevance. Methods: We analyzed Dexcom G6 CGM data from 1,049 participants in the AI-READI dataset. We computed the pairwise Wasserstein Distance (mg/dL) between the glucose probability distributions of all participants. Hierarchical clustering was applied to the distance matrix to identify distinct subgroups. Clinical characteristics (HbA1c) and standard variability metrics (Coefficient of Variation) were compared across clusters using Kruskal-Wallis tests. Results: Topological clustering identified four distinct phenotypes with highly significant differences in clinical control (HbA1c p0.0001). Cluster 1 (Mean Glucose: 342.4 ± 20.3 mg/dL) represented a saturated phenotype with the highest HbA1c (10.0% ± 3.4%) but low variability (CV: 16.4% ± 5.0%), reflecting sustained hyperglycemia near the sensor limit. In contrast, Cluster 2 (Mean: 260.6 ± 16.6 mg/dL; HbA1c: 9.3% ± 1.4%) represented an erratic phenotype with the cohort's highest variability (CV: 23.6% ± 4.6%), distinct from the saturated group. Cluster 3 identified a moderate-variable phenotype (HbA1c: 6.7% ± 1.1%; CV: 22.8% ± 6.0%), while Cluster 4 represented tight control (HbA1c: 5.6% ± 0.7%; TIR: 96.0% ± 7.4%). Conclusion: Topological phenotyping effectively disentangles glucose level severity from instability. Specifically, it distinguishes patients with saturated, unvarying hyperglycemia from those with chaotic, high-amplitude fluctuations; distinct clinical states that standard HbA1c metrics often conflate. This approach offers a precise framework for stratifying type 2 diabetes phenotypes beyond averages. Disclosure B. Panny: None. Funding National Institutes of Health (1OT2OD032644)

Ask AI

Helpful

Bookmark