T1D is clinically heterogeneous, yet its key parameters are still largely interpreted in isolation. As a first step towards developing a digital twin for T1D, we aimed to build a phenotypic mapping and virtual representation of a person living with T1D (PwT1D) usable at the individual level.

We analysed baseline data from 2,187 adults with T1D in the SFDT1 cohort using Discriminative Dimensionality Reduction via learning a Tree (DDRTree). Twenty-five sociodemographic, clinical, biochemical, and continuous glucose monitoring (CGM) variables were used to construct a low-dimensional structure. Structural reproducibility and individual projection were assessed in an unseen set of 235 participants. Cardiovascular disease (CVD) and end-stage kidney disease (ESKD) risks were computed with the Steno T1 Risk Engine.

We identified seven interpretable sub-phenotypes, ranging from well-controlled profiles to profiles characterized by poor glycaemic control, metabolic abnormalities, social vulnerability and a high burden of micro- and macrovascular complications. An independently fitted map for the additional participants showed a similar overall branching structure and comparable clinical patterns, and individual projection was accurate (R² 0.85-0.98). A multilingual application enables exploration of the phenotypic map, individual projection, and estimation of Steno risk scores.

This phenotypic mapping and virtual representation capture clinically meaningful heterogeneity and let care teams locate and explore individual profiles. It provides the foundation for an AI-enabled digital twin an interpretable virtual representation that integrates a person's clinical, biological, glycaemic, and social profile and supports individualized decisions - and a practical clinical decision-support tool for precision diabetes care.