JMIR Publications

ABSTRACT

Background:

Diabetes is a chronic metabolic condition, characterized by impaired blood glucose regulation. It is often linked to serious health complications and comorbidities that significantly affect quality of life, requiring effective management, continuous monitoring, and advanced data analytics. Notably, tailored diabetes management can be enhanced by digital twins (DTs), which serve as an adaptive digital replica of patients using clinical, physiological, and lifestyle data.

Objective:

This review explores diabetes-related DTs by examining their patient representation levels. We aim to synthetize the current state of the art and outline the foundations of a holistic multi-level, multi-functional personalized digital twin for diabetes management.

Methods:

We investigate requirements for a patient-centred holistic DT and classify existing approaches into three representation levels: 1) Data representation, involving structured, context-aware, and AI-ready data architectures, that support data analysis, enable semantic interoperability, relationship extraction, and real-time bidirectional data exchange between patient and virtual replica. 2) Process representation, primarily based on mechanistic models simulating glucose-insulin-meal, and exercise-glucose dynamics. 3) Data-driven representation, focusing on individualization through predictive modelling of disease onset, adverse events, and generation of explainable, personalized recommendations. The literature is synthetized to provide a perspective of a holistic multi-level DT and to identify research gaps.

Results:

Digital twins accompany patients throughout their lifecycle and span a wide range of use cases from long-term disease prediction to timely prediction of severe events. However, patient-centered DTs remain at an early developmental stage. Most existing systems function primarily as simulation tools and lack comprehensive integration of data, process, and data-driven representations. Key gaps include limited use of standardized semantic data models and ontologies, insufficient real-time bidirectional architectures, and fragmented integration of mechanistic and machine learning models, which are often treated as independent rather than complementary components.

Conclusions:

Although digital twins show substantial potential for advancing personalized diabetes care, current implementations remain fragmented and incomplete. Future research should prioritize the development of holistic, multi-level digital twins that integrate interoperable data infrastructures, mechanistic simulations, and data-driven models into cohesive, patient-centered systems capable of supporting lifelong disease management.