Use of Continuous Glucose Monitoring with Machine Learning to Identify Metabolic Subphenotypes and Inform Precision Lifestyle Changes

Ahmed A. Metwally, PhD, Heyjun Park, PhD, Yue Wu, PhD, Tracey McLaughlin, MD, Michael P. Snyder, PhD · 2025

Read Use of Continuous Glucose Monitoring with Machine Learning to Identify Metabolic Subphenotypes and Inform Precision Lifestyle Changes by Ahmed A. Metwally, PhD, Heyjun Park, PhD, Yue Wu, PhD, Tracey McLaughlin, MD, Michael P. Snyder, PhD (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-03986.txt.

Introduction

Ahmed A Metwally, Heyjun Park, Yue Wu, Tracey McLaughlin, and Michael P Snyder review how steady glucose monitoring joined with machine learning can split diabetes and its early stage into metabolic subtypes. Fixed glucose thresholds hide differing causes, chiefly insulin resistance, beta-cell dysfunction, and incretin deficiency, and those causes call for different responses.

Home glucose-tolerance tests with wearable sensors feed models that predict standard measures of muscle insulin resistance and beta-cell function. How each person responds to foods, sleep, activity, and meal timing link to specific faults, and dietary aids work differently by body type. The review argues that rich glucose traces can guide tailored lifestyle changes rather than one-size advice.

Catalog record

CREATORAhmed A. Metwally, PhD, Heyjun Park, PhD, Yue Wu, PhD, Tracey McLaughlin, MD, Michael P. Snyder, PhD
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-03986

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Use of Continuous Glucose Monitoring with Machine Learning to Identify Metabolic Subphenotypes and Inform Precision Lifestyle Changes Ahmed A. Metwally, PhD, Heyjun Park, PhD, Yue Wu, PhD, Tracey McLaughlin, MD, Michael P. Snyder, PhD Abstract Abstract The classification of diabetes and prediabetes by static glucose thresholds obscures the pathophysiological dysglycemia heterogeneity, primarily driven by insulin resistance (IR), β \beta -cell dysfunction, and incretin deficiency. This review demonstrates that continuous glucose monitoring and wearable technologies enable a paradigm shift towards non-invasive, dynamic metabolic phenotyping. We show evidence that machine learning models can leverage high-resolution glucose data from at-home, CGM-enabled oral glucose tolerance tests to accurately predict gold-standard measures of muscle IR and β \beta -cell function. This personalized characterization extends to real-world nutrition, where an individual’s unique postprandial glycemic response (PPGR) to standardized meals, such as the relative glucose spike to potatoes versus grapes, could serve as a biomarker for their metabolic subtype. Moreover, integrating wearable data reveals that habitual diet, sleep, and physical activity patterns, particularly their timing, are uniquely associated with specific metabolic dysfunctions, informing precision lifestyle interventions. The efficacy of dietary mitigators in attenuating PPGR is also shown to be phenotype-dependent. Collectively, this evidence demonstrates that CGM can deconstruct the complexity of early dysglycemia into distinct, actionable subphenotypes. This approach moves beyond simple glycemic control, paving the way for targeted nutritional, behavioral, and pharmacological strategies tailored to an individual’s core metabolic defects, thereby paving the way for a new era of precision diabetes prevention. \reportnumber 1 Introduction 1.1 Significance of Diabetes and Need to Define Heterogeneity Diabetes affects 38.1 million, or 14.7%, of all U.S. adults aged 18 years or older, and another 38 million, or 38%, have prediabetes, of whom less than 20% are aware ( CDC 2024 ) . Diabetes and prediabetes are classified based on the degree of glucose elevation, but this does not reflect the multiple physiological pathways that can lead to elevations in blood glucose. Specifically, insulin resistance (IR), beta-cell failure, incretin deficiency, and unrestrained hepatic glucose production (hepatic IR) all contribute. If it were possible to determine which process/processes were the predominant contributors to dysglycemia in a given individual, targeted therapy could be employed, potentially yielding more successful diabetes prevention and treatment than following a one-size-fits-all approach. Historically, both lifestyle and medical interventions have been recommended based on the degree of overweight and glucose elevation without regard for 1) interindividual variability in pathophysiology; 2) risk trajectory, or 3) anticipated individualized treatment response. With new technologies such as continuous glucose monitoring, it may be possible to characterize and subtype early stages of dysglycemia that reflect/predict these important characteristics, thus significantly impacting human health. In the following review, we outline several novel aspects of CGM and other wearable devices that make inroads into better characterizing metabolic heterogeneity of adults with early dysglycemia, and highlight the interactions between metabolic physiology and response to specific lifestyle interventions. 1.2 Limitations of Current Diabetes Classification Diabetes and prediabetes are diagnosed based on glucose cutoffs ( ADA 2025 ; Qu et al. 2022 ; Nôga et al. 2024 ; Shen et al. 2025 ) . Thus, diagnosis is based on an elevated plasma glucose without regard for the mechanism that led to the elevation. We now know that many pathophysiological pathways contribute to glucose elevation – and yet these are lumped together under the umbrella of diabetes. Current diabetes classification of non-pregnant adults is limited to type 1, type 2, and other specific forms ( ADA 2025 ; Qu et al. 2022 ; Nôga et al. 2024 ; Shen et al. 2025 ) . Of all prevalent cases of diabetes, type 2 represents > 95%. It is becoming clear that not all type 2 diabetes patients follow the same clinical trajectory, with some more prone than others to developing deterioration of glucose control, cardiovascular events, nephropathy, retinopathy, or fatty liver disease ( Hulman et al. 2017 ; Kobayashi et al. 2025 ; Karpati et al. 2018 ; Anjana et al. 2020 ) . Similarly, it is also known that not all patients respond equally well to the same diabetes medications ( Nair et al. 2022 ) . Thus, there is a movement towards subclassifying type 2 diabetes to better reflect the heterogeneity and inform targeted treatments. Methods for subtyping diabetes/prediabetes range from genetic ( Udler et al. 2019 ) to clinical/laboratory biomarkers ( Ahlqvist et al. 2018 ) . We recently proposed that CGM represents a novel and accurate method that can be used to subtype the early dysglycemic state according to underlying metabolic physiology ( Metwally et al. 2024 ) . This method has the advantage over genetic subclassification of reflecting the current clinical state which can change as a result of contributory environmental factors (e.g., body weight, physical activity), and over both genetic and clinical/demographic factors in degree of accuracy ( Metwally et al. 2024 ) . It does not require laboratory tests or measurement of waist circumference, which pose potential obstacles. Furthermore, because only a single CGM sensor is needed, the cost and burden are minimal, and because it does not require access to a health care system, it bypasses socioeconomic and geographic barriers that might preclude diagnosis/subphenotyping in a standard health care system that is relatively inaccessible to low income and/or geographically remote populations. Lastly, this method is amenable to tracking changes or stability in metabolic health over time, as demonstrated in the relative stability of a five-point OGTT glucose trajectory over three years ( Hulman et al. 2018 ) . 1.3 Overview of why postprandial glucose pattern would be good for metabolic phenotyping based on dynamic features of the glucose curve Glucose elevations following an oral nutrient challenge have been used in clinical practice since the development of the first oral glucose tolerance test (OGTT) in 1917 ( Jagannathan et al. 2020 ) . While the use of this test has changed over time, early in its evolution, post-glucola metrics including the 1-hour glucose, the integrated area-under-the-curve glucose, whether the shape of the glucose curve from a five point test (0, 30, 60, 90, 120 min) was mono- or bi-phasic or unclassified was used to identify both underlying metabolic features of IR and beta-cell dysfunction, as well as individuals at highest risk of progressing to diabetes ( Cheng 2015 ; Abdul-Ghani et al. 2008 ; Alyass et al. 2015 ; Kim et al. 2016 ; Tänczer et al. 2020 ; Kim et al. 2012 ; Kanauchi et al. 2005 ; Bervoets et al. 2015 ; Chung et al. 2017 ; Kasturi et al. 2019 ; Cheng et al. 2019 ; Kaga et al. 2020 ; Manco et al. 2017 ; Ismail et al. 2018 ; Abdul-Ghani et al. 2010 ) . Later, mathematical approaches such as functional component analysis ( Frøslie et al. 2013 ) , and latent trajectory analysis ( Hulman et al. 2018 ) were explored as biomarkers of physiologic subtypes of diabetes. Due to the burden of the five-point OGTT, however, this was soon abandoned in favor of the two-point (0 and 120 min) test with use of values for both fasting and 120 min as diagnostic thresholds for diabetes based on the correlation with development of retinopathy ( Report of

Read the complete text