CoChat

Personalized nutrition in type 2 diabetes remission: application of digital twin technology for predictive glycemic control

Open access

The DOI matches its Crossref record; 31 of the 32 references it cites were found in a registry or an index; 48 of the 73 sources its claims cite were checked for support.

Abstract

Background: Type 2 Diabetes (T2D) is a complex condition marked by insulin resistance and beta-cell dysfunction. Traditional dietary interventions, such as low-calorie or low-carbohydrate diets, typically overlook individual variability in postprandial glycemic responses (PPGRs), which can lead to suboptimal management of the disease. Recent advancements suggest that personalized nutrition, tailored to individual metabolic profiles, may enhance the effectiveness of T2D management. Objective: This study aims to present the development and application of a Digital Twin (DT) technology-a machine learning (ML)-powered platform designed to predict and modulate PPGRs in T2D patients. By integrating continuous glucose monitoring (CGM), dietary data, and other physiological inputs, the DT provides individualized dietary recommendations to improve insulin sensitivity, reduce hyperinsulinemia, and support the remission of T2D. Methods: We developed a sophisticated DT platform that synthesizes real-time data from CGM, dietary logs, and other biometric inputs to create personalized metabolic models for T2D patients. The intervention is delivered via a mobile application, which dynamically adjusts dietary recommendations based on predicted PPGRs. This methodology is validated through a randomized controlled trial (RCT) assessing its impact on various metabolic markers, including HbA1c, metabolic-associated fatty liver disease (MAFLD), blood pressure, body weight, ASCVD risk, albuminuria, and diabetic retinopathy. Results: Preliminary data from the ongoing RCT and real-world study demonstrate the DT's capacity to generate significant improvements in glycemic control and metabolic health. The DT-driven personalized nutrition plan has been associated with reductions in HbA1c, enhanced beta-cell function, and normalization of hyperinsulinemia, supporting sustained T2D remission. Additionally, the DT's predictions have contributed to improvements in MAFLD markers, blood pressure, and cardiovascular risk factors, highlighting its potential as a comprehensive management tool. Conclusion: The DT technology represents a novel and scalable approach to personalized nutrition in T2D management. By addressing individual variability in PPGRs, this method offers a promising alternative to conventional dietary interventions, with the potential to improve long-term outcomes and reduce the global burden of T2D.

Overview

CoChat Summary

TL;DR

This methods article describes a machine-learning Digital Twin (DT) platform that integrates CGM, dietary, activity, sleep, medication, and laboratory data to predict personalized postprandial glycemic responses and guide nutrition in type 2 diabetes. It reports model performance and summarizes ongoing RCT and real-world outcomes showing improved HbA1c, remission, liver, cardiovascular, and weight measures.

Key findings

  • The DT PPGR prediction model achieved an RMSE of 24.96 mg/dL, MAE of 17.21 mg/dL, R² >0.85, and AUC-ROC >0.85; MSE was between 45 and 55, with 91% of predictions falling within 40 points of actual values.
  • In the ongoing open-label RCT at 1 year, the DT group’s mean HbA1c improved from 9.0 ± 1.9 to 6.1 ± 0.7 (P<0.001), while standard care changed from 8.5 ± 1.9 to 8.2 ± 1.6 (P=0.051); 94% of the DT group discontinued all T2D medications and 72.7% achieved remission, compared with none in standard care.
  • Liver and metabolic outcomes improved in the DT group: normal NAFLD-LFS rose from 11.8% to 67.4% versus 16% to 9.9% in standard care (P<.00001); normal FLI rose from 13.6% to 54.3% versus 15.9% to 13.4% (P=0.0003); MRI-PDFF liver fat was 5.5 ± 4.7% versus 10.9 ± 6.8% (P<.001), and 83.2% versus 34% achieved liver fat <6%.
  • Weight and cardiovascular measures also improved: 73.8% of DT patients lost ≥5% weight and 41.6% lost ≥10%; mean weight reduction was 7.4 ± 5.9 kg versus 0.4 ± 3.6 kg (P<0.001), SBP fell from 140.2 ± 15.8 to 127.1 ± 8.3 mmHg (P=0.02), DBP from 94.7 ± 11.7 to 86.8 ± 6.4 mmHg (P=0.014), and 60% achieved hypertension remission. At 18 months, 64.3% of DT participants who achieved remission at 12 months maintained it.

Methods

  • This is a methods article describing the DT platform: it uses CatBoostRegressor and Random Forest, and later a hybrid Gradient Boosting Machine plus LSTM architecture, to predict PPGRs from 37 input variables including meal composition, CGM trends, medications, sleep, activity, and laboratory data.
  • The intervention is delivered through the Whole Body Digital Twin mobile app, with real-time dietary recommendations, food classification into Green/Orange/Red For You, telehealth support, and continuous CGM/wearable monitoring.
  • Validation is summarized from an ongoing open-label RCT: participants aged 18–70 with T2D ≤8 years were randomized 3:1 to DT (n=250) or standard care (n=86); 17 DT participants withdrew before intervention, leaving 233 DT and 86 SC for intent-to-treat analysis. Primary outcomes were HbA1c and T2D medication dosage over two years; remission was defined as HbA1c <6.5% without glucose-lowering medications for at least 3 months. Model evaluation used k-fold cross-validation (k=5), an 80-20 split, grid search, and metrics including MSE, RMSE, MAE, R², and AUC-ROC.

Limitations

  • Stated: the DT technology relies on accurate and complete data input, and may be affected by language barriers, user proficiency, patient compliance with food logging and device usage, incomplete tracking, and technical issues such as calibration errors in wearables or CGMs.
  • Stated: generalizability to diverse populations with varying dietary habits, socioeconomic conditions, and technology access is a concern; regulatory compliance with evolving data privacy laws such as HIPAA and GDPR may require continuous adaptation; self-reported adherence may introduce bias.
  • Evident from the text: the article is a methods/overview summary of an ongoing open-label RCT and real-world studies, with many outcomes cited from separate publications rather than fully reported here; the open-label design and reliance on preliminary/ongoing data limit causal certainty, and the remission sample size was noted as smaller than that for HbA1c reduction.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3389/fendo.2024.1485464

Cites

31 of 32 references verified0 citation errors · 1 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Digital Twin Platform: Member’s Whole Body Digital Twin for Chronic Disease Reversal. This figure illustrates the Digital Twin Platform’s use of real-time sensor data to create a Whole Body Digital Twin, enabling continuous, personalized health interventions across various domains for optimal well-being.
Figure 1 Digital Twin Platform: Member’s Whole Body Digital Twin for Chronic Disease Reversal. This figure illustrates the Digital Twin Platform’s use of real-time sensor data to create a Whole Body Digital Twin, enabling continuous, personalized health interventions across various domains for optimal well-being.CC BY · PubMed Central
Figure 2 Whole Body Digital Twin® (WBDT) mobile app User Interface. The WBDT Member App interface centers on the daily Action Score, which aggregates data from health modules like Sleep, Breathing, Activity, Nutrition, and Medicine. Users receive actionable steps, such as measuring weight, to stay aligned with their personalized health plan. The app promotes seamless interaction with health metrics, encouraging consistent engagement.
Figure 2 Whole Body Digital Twin® (WBDT) mobile app User Interface. The WBDT Member App interface centers on the daily Action Score, which aggregates data from health modules like Sleep, Breathing, Activity, Nutrition, and Medicine. Users receive actionable steps, such as measuring weight, to stay aligned with their personalized health plan. The app promotes seamless interaction with health metrics, encouraging consistent engagement.CC BY · PubMed Central
Figure 3 Flowchart illustrating the process for training a machine-learned model to output a representation of a patient’s metabolic health. This figure outlines the process used by the digital twin module to train a machine-learning model for predicting a patient’s metabolic state. The process begins with retrieving historical biological and patient data (710) to train a baseline model (720) reflecting population-level trends. Next, a patient-specific dataset (730) is generated, and a personalized model (740) is trained to predict individual metabolic responses. Finally, current biological and patient data are inputted into the trained model (750) to output real-time metabolic states. Data sources include lab tests, sensor data, and patient-recorded measurements, enabling precise metabolic monitoring and health predictions. This figure is taken from the patent US 2021/0196195 A1.
Figure 3 Flowchart illustrating the process for training a machine-learned model to output a representation of a patient’s metabolic health. This figure outlines the process used by the digital twin module to train a machine-learning model for predicting a patient’s metabolic state. The process begins with retrieving historical biological and patient data (710) to train a baseline model (720) reflecting population-level trends. Next, a patient-specific dataset (730) is generated, and a personalized model (740) is trained to predict individual metabolic responses. Finally, current biological and patient data are inputted into the trained model (750) to output real-time metabolic states. Data sources include lab tests, sensor data, and patient-recorded measurements, enabling precise metabolic monitoring and health predictions. This figure is taken from the patent US 2021/0196195 A1.CC BY · PubMed Central
Figure 4 Process of Predicting and Verifying Patient-Specific Metabolic Responses Using a Machine-Learned Model. (A) This figure depicts the process of implementing a machine-learned model to predict a patient’s specific metabolic response. The model uses wearable sensor data (805), lab test data (810), and symptom data (815) to establish an initial metabolic state (825). Once sufficient training data for a patient exists, the digital twin module (450) can predict the patient’s metabolic response (850) based on input biosignals (830), such as nutrition (835), medication (840), and lifestyle data (845). The predicted metabolic response reflects changes in the patient’s health, corresponding to the input biosignals. (B) This figure illustrates the comparison process between a patient’s predicted metabolic state and true metabolic state. During a given time period, wearable sensor data (805), lab test data (810), and symptom data (815) are used to generate a patient’s true metabolic state (860) via the digital twin module (450). Simultaneously, input biosignals (830), such as nutrition, medication, and lifestyle data, are processed to predict the patient’s metabolic state (870). The Response Review Module (880) compares the true and predicted metabolic states to identify discrepancies, helping detect any errors in the biosignal inputs that might have contributed to the differences. This figure is taken from the patent US 2021/0196195 A1.
Figure 4 Process of Predicting and Verifying Patient-Specific Metabolic Responses Using a Machine-Learned Model. (A) This figure depicts the process of implementing a machine-learned model to predict a patient’s specific metabolic response. The model uses wearable sensor data (805), lab test data (810), and symptom data (815) to establish an initial metabolic state (825). Once sufficient training data for a patient exists, the digital twin module (450) can predict the patient’s metabolic response (850) based on input biosignals (830), such as nutrition (835), medication (840), and lifestyle data (845). The predicted metabolic response reflects changes in the patient’s health, corresponding to the input biosignals. (B) This figure illustrates the comparison process between a patient’s predicted metabolic state and true metabolic state. During a given time period, wearable sensor data (805), lab test data (810), and symptom data (815) are used to generate a patient’s true metabolic state (860) via the digital twin module (450). Simultaneously, input biosignals (830), such as nutrition, medication, and lifestyle data, are processed to predict the patient’s metabolic state (870). The Response Review Module (880) compares the true and predicted metabolic states to identify discrepancies, helping detect any errors in the biosignal inputs that might have contributed to the differences. This figure is taken from the patent US 2021/0196195 A1.CC BY · PubMed Central
Figure 5 Predictor Features for Digital Twin Model Development. The predictor model includes demographic info (age, sex, BMI, gender), activity level (steps, active/sedentary minutes), sleep data, blood test results (HbA1c, HOMA), food nutrition data, recent glucose features, medications, and time of day.
Figure 5 Predictor Features for Digital Twin Model Development. The predictor model includes demographic info (age, sex, BMI, gender), activity level (steps, active/sedentary minutes), sleep data, blood test results (HbA1c, HOMA), food nutrition data, recent glucose features, medications, and time of day.CC BY · PubMed Central

Verification

Claims the paper makes

45 Valid · 0 Not valid · 22 Not checked · 32 Own results and interpretations

29 more claims · 16 with their cited sources checked

Create a free account to see all 99 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 55
  • Probably not a substantial claim. 25

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

1 of 1

Citations

Cites

Show all 32 references

Cited by

See all 56

Metadata

Authors
Shamanna, P., Joshi, S., Thajudeen, M., Shah, L., Poon, T., Mohamed, M., & Mohammed, J.
Affiliations
  • Bangalore Diabetes Centre (IN)
  • Lilavati Hospital & Research Centre (IN)
Corresponding author
Shamanna, Paramesh
Venue
Frontiers in Endocrinology
Published
2024-11-20
Volume
15
Pages
1485464-1485464
ISSN
1664-2392 (ISSN-L)
OpenAlex
W4404538085
Cited by
56
References
30
Citations by year
2026: 37 · 2025: 19
FWCI
18.32
Type
journal article
Language
en
Open access
Locations
  • Frontiers in Endocrinology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Health Sciences › Medicine › Endocrinology, Diabetes and Metabolism › Diabetes Management and Research
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Genetics › Nutrition, Genetics, and Disease
  • Health Sciences › Health Professions › General Health Professions › Mobile Health and mHealth Applications
Keywords
type 2 diabetes remission, personalized nutrition, digital twin, postprandial glycemic response, continuous glucose monitoring, dietary recommendations, glycemic control, hyperinsulinemia, beta-cell function, insulin sensitivity, metabolic-associated fatty liver disease, cardiovascular risk, mobile health apps, HbA1c
MeSH terms
Machine Learning, Digital Technology, Glycemic Control, Blood Glucose, Diabetes Mellitus, Type 2, Female, Humans, Male, Middle Aged, Remission Induction, Blood Glucose Self-Monitoring, Postprandial Period, Precision Medicine

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Personalized nutrition in type 2 diabetes remission: application of digital twin technology for predictive glycemic control. https://app.cochat.ai/papers/10.3389/fendo.2024.1485464
The paper
Shamanna, P., Joshi, S., Thajudeen, M., Shah, L., Poon, T., Mohamed, M., & Mohammed, J. (2024). Personalized nutrition in type 2 diabetes remission: application of digital twin technology for predictive glycemic control. Frontiers in Endocrinology. https://doi.org/10.3389/fendo.2024.1485464

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.