Network Physiology of Diabetes and Hypertension: Modeling Disease Progression from Multiparametric Biosignals

Abstract

Type II diabetes mellitus (T2DM) is a heterogeneously presenting metabolic disorder that poses a serious health concern worldwide due to its rising prevalence. T2DM is preceded by prediabetes, which is a reversible condition given early detection and appropriate management. Hypertension (HT) is a frequent comorbidity of T2DM, with the cooccurrence of both conditions increasing the risk of diabetes-associated complications, particularly cardiovascular disease (CVD). Identifying individuals at risk of T2DM development and progression to comorbid disease with HT is critical in ameliorating debilitating complications.

The thesis is organized in four studies: the first conducts a univariate analysis of inflammation, oxidative stress (OS) mitochondrial dysfunction (MD) and hemostasis in prediabetes, T2DM and comorbid T2DM and HT to identify discriminatory biomarkers between stages of disease progression.

The second study examines the link between mitochondrial OS and cardiac autonomic neuropathy (CAN) by investigating the relationship between Humanin, a mitochondrial-derived peptide (MDP), and short-term heart rate variability (HRV) indices. This is carried out in the context of examining a potential therapeutic target for CAN, a diabetes-associated complication with no specific treatment.

Utilizing the same biomarkers in the first study, the third and fourth studies conduct a multivariate analysis of T2DM progression. The third study aims to construct accurate and interpretable models of T2DM incidence risk. The models address two frequently encountered problems in T2DM prediction: class imbalance and poor interpretability. Class imbalance is addressed by upsampling the minority class using variational autoencoders (VAE). Interpretability of the models is ensured by utilizing white-box logistic regression models.

The fourth and final study predicts the risk of HT development in T2DM patients through an extreme gradient boosting (XGBoost) model. To ensure explainability of the model, Shapley additive values (SHAP) are utilized for the identification of influential features.

Date of Award	Dec 2022
Original language	American English
Supervisor	Herbert Jelinek (Supervisor)