Objective
The objective of this project is to support more personalized type 1 diabetes management by proposing a digital twin that represents the individual patient, their clinical history, real-time sensor data, and their changing context over time. The project focuses on combining patient-specific data, process-based modeling, and data-driven personalization to support individualized predictions and decision-making and enable a bidirectional flow between the virtual model and the patient.
Description
In chronic disease management, treatment decisions often depend on patient-specific context. While general clinical guidelines provide an important foundation, they may not fully capture the individual variability that influences disease onset, treatment response, and daily decision-making. This has increased interest in personalized thresholds, treatment titration, and patient-specific models rather than one-size-fits-all care.
Digital twins can support this development by representing an individual patient and their context over time. This is particularly relevant for type 1 diabetes, where physiological parameters are affected by well-being, meal intake, lifestyle, and daily routines.
In this project, we classify patient representations across three levels: data, process, and data-driven representation. The first level is the patient’s data representation. It integrates semantic, contextual, historical, and real-time sensor data in a structured, machine-readable format to support efficient inference and model training. The second level is the process representation, which models mathematical, mechanistic, and biochemical processes related to glucose metabolism. This representation requires individualization for type 1 diabetes and different age ranges. The third level is the data-driven representation, consisting of personalization modules that individualize decisions and predictions based on the patient’s data and environment from the data representation layer.
Current Research
For the data representation, this study enhances DiaData into a graph-based semantic version supported by a clinically accurate, standardized ontology for type 1 diabetes management that adheres to FAIR principles. The data representation supports both process- and data-driven approaches by providing AI-ready data and enabling continuous adaptation to the patient’s current state and environment. The process representation is modeled using mathematical and biochemical functions. For the data-driven modules, the project focuses on short-term prediction of hypoglycemia onset as the primary use case.
Future Work
These representations can be used as redundant or complementary modules to improve stability, performance, and reliability while enabling continuous adaptation to the patient’s current state and environment. This project aims to explore neurosymbolic AI to investigate the intersection of data, process, and data-driven representations [1].
References
[1]Cinar B, van den Boom L, Maleshkova M. Enhanced diabetes management with digital twins: A scoping review (preprint). JMIR Preprints. Published online 26 February 2026. doi:10.2196/preprints.91729
Letzte Änderung: 6. August 2026