Prediction of metabolic subphenotypes of type 2 diabetes via continuous glucose monitoring and machine learning

Ahmed A Metwally; Dalia Perelman; Heyjun Park; Yue Wu; Alokkumar Jha; Seth Sharp; Alessandra Celli; Ekrem Ayhan; Fahim Abbasi; Anna L Gloyn; Tracey McLaughlin; Michael P Snyder

doi:10.1038/s41551-024-01311-6

Prediction of metabolic subphenotypes of type 2 diabetes via continuous glucose monitoring and machine learning

Nat Biomed Eng. 2024 Dec 23. doi: 10.1038/s41551-024-01311-6. Online ahead of print.

Authors

Ahmed A Metwally^{1

2

3}, Dalia Perelman^{1

4}, Heyjun Park^{1

5}, Yue Wu¹, Alokkumar Jha⁶, Seth Sharp⁶, Alessandra Celli¹, Ekrem Ayhan⁴, Fahim Abbasi⁴, Anna L Gloyn^{6

7}, Tracey McLaughlin^#⁸, Michael P Snyder^#⁹

Affiliations

¹ Department of Genetics, Stanford University, Stanford, CA, USA.
² Systems and Biomedical Engineering Department, Cairo University, Giza, Egypt.
³ Google LLC, Mountain View, CA, USA.
⁴ Department of Medicine, Stanford University, Stanford, CA, USA.
⁵ Department of International Health, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA.
⁶ Department of Pediatrics, Stanford University, Stanford, CA, USA.
⁷ Stanford Diabetes Research Centre, Stanford University, Stanford, CA, USA.
⁸ Department of Medicine, Stanford University, Stanford, CA, USA. tmclaugh@stanford.edu.
⁹ Department of Genetics, Stanford University, Stanford, CA, USA. mpsnyder@stanford.edu.

^# Contributed equally.

PMID: 39715896
DOI: 10.1038/s41551-024-01311-6

Abstract

The classification of type 2 diabetes and prediabetes does not consider heterogeneity in the pathophysiology of glucose dysregulation. Here we show that prediabetes is characterized by metabolic heterogeneity, and that metabolic subphenotypes can be predicted by the shape of the glucose curve measured via a continuous glucose monitor (CGM) during standardized oral glucose-tolerance tests (OGTTs) performed in at-home settings. Gold-standard metabolic tests in 32 individuals with early glucose dysregulation revealed dominant or co-dominant subphenotypes (muscle or hepatic insulin-resistance phenotypes in 34% of the individuals, and β-cell-dysfunction or impaired-incretin-action phenotypes in 40% of them). Machine-learning models trained with glucose time series from OGTTs from the 32 individuals predicted the subphenotypes with areas under the curve (AUCs) of 95% for muscle insulin resistance, 89% for β-cell deficiency and 88% for impaired incretin action. With CGM-generated glucose curves obtained during at-home OGTTs, the models predicted the muscle-insulin-resistance and β-cell-deficiency subphenotypes of 29 individuals with AUCs of 88% and 84%, respectively. At-home identification of metabolic subphenotypes via a CGM may aid the risk stratification of individuals with early glucose dysregulation.