Childhood obesity is the main driver of early metabolic risk, predisposing to cardiovascular disease (CVD) and type 2 diabetes (T2D), which cause millions of deaths worldwide. Their progression is influenced by biological, behavioral, and environmental factors. Digital Twin Systems (DTS) offer innovative ways to monitor and predict cardiometabolic risk. This work presents a prototype digital twin platform called PODiaCarD designed for managing pediatric obesity and related cardiometabolic complications. The system integrates clinical, anthropometric, and lifestyle data with machine learning to estimate outcomes in youth. Built on a three-layer architecture (frontend, backend, predictive engine), PODiaCarD ensures scalability, observability, and reproducibility while enabling continuous model improvement. Models, trained on the PODiaCar project dataset (n = 552, 12.2 ± 2.9 years) with cross-validation and target-specific algorithms, predict eight key metabolic outcomes. The infrastructure follows privacy-by-design and GDPR standards, ensuring security, auditability, and clinical compliance. PODiaCarD achieved excellent performance for TyG index (F1 = 0.975 ± 0.014, random forest) and solid results for HbA1C (F1 = 0.844 ± 0.028, random forest). Moderate accuracy was observed for HOMA (F1 = 0.670 ± 0.070, Gradient Boosting). In contrast, models for blood pressure (R2 = 0.05–0.21; F1 = 0.446 ± 0.045) and glycemia (F1 = 0.113 ± 0.113) showed poor predictive capacity, while insulin regression (R2 = 0.211) remained limited, highlighting the need for richer datasets. Conclusions: PODiaCarD is a promising tool for managing pediatric obesity and complications. It integrates clinical, anthropometric, and behavioral data with ML-based models to support pediatricians in early risk detection, dynamic monitoring, and personalized prevention. Its federated design allows continuous dataset growth and improved predictive performance, strengthening its role in pediatric cardiometabolic care.<Table Float="No" ID="Taba"> <tgroup cols="2"> <colspec align="left" colname="c1" colnum="1" /> <colspec align="left" colname="c2" colnum="2" /> <tbody> <row> <entry nameend="c2" namest="c1"> What is Known: • Pediatric obesity is a major early driver of cardiometabolic risk; body mass index, waist circumference, and lipid profile are key indicators of insulin resistance, type 2 diabetes, and cardiovascular diseases. Existing pediatric predictive models are often static and limited in longitudinal integration. • Digital Twin Systems enable dynamic monitoring and “what-if” simulations in healthcare, but cardiometabolic applications in pediatric populations remain scarce and insufficiently validated. </entry> </row> <row> <entry nameend="c2" namest="c1"> What is New: • PODiaCarD introduces a federated pediatric digital twin that integrates clinical, anthropometric, and lifestyle data with machine learning to dynamically update individual cardiometabolic risk profiles over time. • The platform achieves strong performance for insulin resistance surrogates and HbA1c prediction, provides explainable AI outputs, ensures privacy-by-design, and supports scalable, multi-centre personalized prevention strategies. </entry> </row> </tbody> </tgroup> </Table>

PODiaCarD: a prototype of a digital twin platform for the management of pediatric obesity and related cardiometabolic complications

期刊论文
发表日期：

2026年1月14日

查看全文

Valeria Calcaterra,
Umberto Ciriello,
Samuele Medici,
Valter Pagani,
Cristina Campoy,
Lucia Labati,
Virginia Rossi,
Mireia Escudero-Marin,
Matteo Vandoni,
Camilo Corbellini,
Elvira Verduci,
Luca Marin,
Rocio Bonillo-Leon,
Khatija Bahdur,
Alessandro Gatti,
Giulia Fiore,
Vittoria Carnevale Pellino,
Savina Mannarino,
Gianvincenzo Zuccotti

摘要

Childhood obesity is the main driver of early metabolic risk, predisposing to cardiovascular disease (CVD) and type 2 diabetes (T2D), which cause millions of deaths worldwide. Their progression is influenced by biological, behavioral, and environmental factors. Digital Twin Systems (DTS) offer innovative ways to monitor and predict cardiometabolic risk. This work presents a prototype digital twin platform called PODiaCarD designed for managing pediatric obesity and related cardiometabolic complications. The system integrates clinical, anthropometric, and lifestyle data with machine learning to estimate outcomes in youth. Built on a three-layer architecture (frontend, backend, predictive engine), PODiaCarD ensures scalability, observability, and reproducibility while enabling continuous model improvement. Models, trained on the PODiaCar project dataset (n = 552, 12.2 ± 2.9 years) with cross-validation and target-specific algorithms, predict eight key metabolic outcomes. The infrastructure follows privacy-by-design and GDPR standards, ensuring security, auditability, and clinical compliance. PODiaCarD achieved excellent performance for TyG index (F1 = 0.975 ± 0.014, random forest) and solid results for HbA1C (F1 = 0.844 ± 0.028, random forest). Moderate accuracy was observed for HOMA (F1 = 0.670 ± 0.070, Gradient Boosting). In contrast, models for blood pressure (R² = 0.05–0.21; F1 = 0.446 ± 0.045) and glycemia (F1 = 0.113 ± 0.113) showed poor predictive capacity, while insulin regression (R² = 0.211) remained limited, highlighting the need for richer datasets. Conclusions: PODiaCarD is a promising tool for managing pediatric obesity and complications. It integrates clinical, anthropometric, and behavioral data with ML-based models to support pediatricians in early risk detection, dynamic monitoring, and personalized prevention. Its federated design allows continuous dataset growth and improved predictive performance, strengthening its role in pediatric cardiometabolic care.

What is Known:

• Pediatric obesity is a major early driver of cardiometabolic risk; body mass index, waist circumference, and lipid profile are key indicators of insulin resistance, type 2 diabetes, and cardiovascular diseases. Existing pediatric predictive models are often static and limited in longitudinal integration.

• Digital Twin Systems enable dynamic monitoring and “what-if” simulations in healthcare, but cardiometabolic applications in pediatric populations remain scarce and insufficiently validated.

What is New:

• PODiaCarD introduces a federated pediatric digital twin that integrates clinical, anthropometric, and lifestyle data with machine learning to dynamically update individual cardiometabolic risk profiles over time.

• The platform achieves strong performance for insulin resistance surrogates and HbA1c prediction, provides explainable AI outputs, ensures privacy-by-design, and supports scalable, multi-centre personalized prevention strategies.