Dissecting cardiometabolic disease heterogeneity using machine learning and multi-scale modelling & simulation

Thesis

Metabolic disease is one of the most pressing health and societal challenges of the 21st century. Over half a billion adults are affected by type 2 diabetes, a burden that is expected to grow alongside increasing rates of obesity and older age. Diabetes and obesity are major risk factors for cardiovascular disease, the leading global cause of death, particularly for heart failure with preserved ejection fraction (HFpEF). HFpEF is a complex, multifactorial syndrome affecting over 30 million people worldwide and represents a common “cardiac end-stage” of chronic metabolic dysfunction. Yet, its pathophysiology remains poorly understood and challenging to treat. Understanding these phenotypes and the early cardiac changes preceding overt HFpEF in key drivers such as type 2 diabetes is crucial to improve early identification, personalise care, and prevent disease progression.

Harnessing complementary data-driven and mechanistic computational modelling approaches, this thesis aims to characterise heterogeneity in patient clinical profiles, cardiac structure, function, and electrophysiology, as well as underlying disease mechanisms, to advance clinical identification and mechanistic understanding of adverse cardiac remodelling in HFpEF and type 2 diabetes. I leveraged statistics and machine learning applied to large-scale multi-modal data from the UK Biobank to identify three distinct clinical phenogroups of HFpEF, which differed in cardiometabolic comorbidity burden. Each group exhibited distinct cardiac function and clinical characteristics, highlighting the importance of sex- and comorbidity-specific context in HFpEF characterisation. Focusing on type 2 diabetes as an early stage of the HFpEF spectrum, I demonstrated that despite the absence of diagnosed cardiovascular disease, patients exhibited concurrent changes in ECG and imaging biomarkers suggesting abnormalities in heart rate, repolarisation and impaired left ventricular function; possibly indicative of early-stage HFpEF. Finally, I developed a multi-scale, human-based computational modelling and simulation framework to investigate mechanisms driving these electrophysiological abnormalities in type 2 diabetes. Progressive ionic remodelling characterised by reduced repolarisation currents and increased CaMKII activity underpinned disease severity-dependent changes in action potential duration and calcium handling. The resulting QT prolongation and modest QRS shortening mirror subclinical ECG abnormalities identified at cohort-level.

Overall, this thesis supports the vision of a more personalised and preventive medicine approach in cardiometabolic disease management by offering insights into subclinical cardiac abnormalities in type 2 diabetes and high-resolution phenomapping in overt cardiometabolic HFpEF. This work opens new avenues for in silico investigations to better understand and ultimately prevent cardiac decline in patients with metabolic diseases.

Authors: Bertrand, A

• DOI: 10.5287/ora-mv56ebyg8
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: type 2 diabetes; computational modelling and simulation; heart failure; machine learning; UK Biobank
• Subjects: Medical sciences; Computational biology; Biomedical engineering