T1D Resources

Type-One Digital Twin (T1DT)

Patient-specific simulation of glucose dynamics: models, tools for fitting them to real data, and ways to evaluate them.

Overview

High-fidelity mechanistic models such as the UVA/Padova simulator describe the complex glucose-insulin dynamics of T1D as a system of differential equations. Their parameters encode an individual’s physiology: how carbohydrate (CHO) intake and insulin action drive glucose levels. Identifying those parameters from a person’s observed data gives a digital twin (DT), which can be used for retrospective analysis of treatment decisions.

Building an accurate DT is hard. Hidden states relate nonlinearly to what we observe, and physiology varies widely between individuals.

From MCMC to simulation-based inference

Earlier T1D digital twin methods such as ReplayBG use Markov chain Monte Carlo (MCMC) to infer a posterior distribution over patient-specific parameters from observed glucose and insulin dosing data. The fitted model can then be replayed under alternative treatments, such as different insulin doses. MCMC is computationally expensive, though: every new window of observed data needs a fresh sampling run.

Simulation-based inference (SBI) scales better and has shown success in T1D digital twinning. In neural posterior estimation (NPE), a neural network is trained on simulated parameter-CGM pairs to learn a conditional approximation of the posterior. Once trained, it infers a patient’s parameters almost instantly. This removes MCMC’s per-window cost while improving digital twin accuracy on held-out evaluation windows.

Code & getting started

  • ReplayBG MATLAB

    Fits a patient-specific glucose model to recorded data and replays it under altered therapy ("what if") scenarios.

Datasets

For public continuous glucose monitoring (CGM) data, start with Glucose-ML, which collects public CGM datasets for research.

Related work

  1. The UVA/PADOVA Type 1 Diabetes Simulator: New Features Chiara Dalla Man, Francesco Micheletto, Dayu Lv, Marc Breton, Boris Kovatchev, Claudio Cobelli. Journal of Diabetes Science and Technology, 2014.