Robust Cardiovascular Models for Prediction of Hypertension Using Minimal Data Sets

Abstract

Cardiovascular disease is the leading cause of death worldwide. Such diseases are multifaceted, but can manifest as acute events such as cardiac infactions, strokes and aortic dissections which can lead directly to death or disability. However a major risk factor for developing such diseases is high blood pressure, and high blood pressure is the globally leading cause of years of life lost to premature death or disability. Despite the dangers associated with high blood pressure being quite well-known, the number of people with hypertension worldwide has been estimated to double from 1990 to 2019. Therefore, to prevent cardiovasculare disease fatalities, developing more knowlegde about predicting who is at risk of getting high blood pressure and effectively preventing and treating this condition is important.

Applying what we know about the cardiovascular system, hemodynamics and the mechanics of biological tissues, we can constuct physics-based mathematical models of varying complexity which mimic the function of the vasculature and heart. We for explame can build computentially intensive detailed models which simulate the action of heart valves during heart beats, and how blood and tissue interacts with each other. Or we can simplify large collections of blood vessels as a geometryless pool of blood, and analyze the behaviour of the blood as electical currents in a circuit. This thesis focuses on the application of the latter type of models for simulating the blood pressure in the systemic arteries using a small set of parameters which describes the mechanical action of the vessels and heart.

This thesis is paper based and details investigations into the feasibility of personalizing such models for tracking progression in hypertension and therapy. The first part of the investigation details attempts to personalize the model to synthetic data to assess the error of parameter estimates under such conditions. The second part details the attempts to personalize two lumped parameter models using synthetic data, while the assessing the variability of parameter estimates under various conditions to assess the feasibility to resolve parameter changes over time with the selected optimization method. The final part of the thesis attempts to determine if any patterns in the data such as cardiorespiratory fitness, individual characteristics or hemodynamics explain the observed changes and variability in parameters.