A Multi-Scale Computational Framework for Human-Based Modelling and Simulation of Adverse Cardiac Remodelling in Type 2 Diabetes

Ambre Bertrand1, Lucas Arantes Berg1, Ruben Doste2, Jakub Tomek1, Albert Dasí1, Abhirup Banerjee1, Julia Camps3, Vicente Grau4, Blanca Rodriguez1
1University of Oxford, 2Department of Computer Science, University of Oxford, 3Universitat Pompeu Fabra, 4Institute of Biomedical Engineering, Department of Engineering Science, University of Oxford


Abstract

Introduction: Type 2 diabetes affects (T2D) over half a billion people worldwide and is a major risk factor for cardiac arrhythmias and heart failure. Early detection of cardiac abnormalities and understanding the underlying mechanisms are essential to prevent further cardiac deterioration in T2D.

Aim: To quantify subclinical cardiac remodelling in patients with T2D and develop an in-silico framework to investigate the mechanisms driving these changes at multiple scales.

Methods: Using UK Biobank data, we conducted a matched cross-sectional study comparing ECG and cardiac magnetic resonance imaging biomarkers between 1781 patients with T2D and healthy controls. We integrated experimental and clinical data into adapted state-of-the-art human-based modelling and simulation tools. These included the ToR-ORd ventricular cardiomyocyte model, a deep learning-based 3D anatomical reconstruction pipeline, and inference-based personalisation of cardiac activation times, to replicate the diabetic cardiac phenotype.

Results: Despite no cardiovascular disease, patients with T2D showed significant cardiac differences, notably a higher resting heart rate (66 vs. 61 bpm, p < 0.001), longer QTc interval (424 vs. 420ms, p < 0.001), lower stroke volume (72 vs. 78ml, p < 0.001) and thicker left ventricular wall (6.1 vs. 5.9mm, p < 0.001). Diabetes-specific ionic changes implemented in the ToR-ORd model resulted in prolonged action potentials, aligning with QTc prolongation, and reduced active tension, consistent with lower stroke volume. The reconstructed patient-specific cardiac anatomy and torso captured key features such as chest size and ventricular wall thickness. Personalised cardiac activation times were successfully computed, enabling whole-heart electrophysiology simulations and ECG reconstruction.

Conclusion: By integrating population-level Big Data analysis and patient-specific modelling and simulation, we identified early cardiac abnormalities in T2D and developed a multi-scale framework for in-silico studies of the diabetic heart. This approach may help improve our understanding and management of cardiac health in this high-risk population.