Deep phenotyping and digital twin of a person living with diabetes

Digital twins have the potential to transform diabetes care by integrating heterogeneous health data into dynamic, patient-specific models. However, current approaches often remain fragmented, focusing on isolated risk factors or outcomes rather than capturing the complex, multidimensional profiles that shape disease progression and complication risk.

This project, funded by DataSpace4Health, aims to develop a digital twin framework for people living with diabetes to support phenotype discovery, risk stratification, and personalised clinical decision-making. Using data from Hôpitaux Robert Schuman (HRS) and the French cohort of people living with type 1 diabetes (SFDT1), the project will combine clinical, anthropometric, biochemical, glycaemic control, lifestyle, patient-reported, and complication-related variables into interpretable machine learning models. The framework will rely on dimensionality reduction and graph-based approaches to map patients into clinically meaningful trajectories and phenotypes. The resulting digital twin will allow the exploration of how individual patients relate to reference populations, how their risk profiles evolve over time, and how specific variables contribute to cardiovascular, renal, neurological, and retinal complications.

Ultimately, the project aims to support personalised monitoring, improve understanding of diabetes heterogeneity, and inform targeted interventions to prevent or delay diabetes-related complications