Early detection of heart attacks and atrial fibrillations
(Jorge Goncalves, with external partner Dominik Linz)
This project aims to apply the tools developed in the theoretical part to predict CTs from normal heart rhythm to atrial fibrillation (AF). AF affects about 2% to 3% of the population, and this percentage increases with age (Kirchhof 2009). The switch to AF typically happens without warning. In some cases it may switch back to normal rhythm on its own, or it may require treatment by drugs or by electrical shocks, also known as cardioversion (Fuster 2006). Ideally, there would be a simple monitoring device, such as a watch or a basic ECG that would record signals and provide advance warning of a CT to AF. A patient would then be able to take specific medication to prevent this transition.
This project will consider two forms of CTs at different time scales: 1) a slow time-varying evolution of the disease, which can take years (development of an arrhythmic remodelling) and 2) a relatively fast time-scale where hearts transition between normal rhythm and AF.
Regarding point 1), AF affects mostly older populations and tends to degrade with age (Stewart 2001). In this case, healthy patients can start developing the disease by changes in the electric activity and structure in the heart years before seeing any symptoms (Schotten 2011). Detecting the disease early could provide a chance for patients to change their life style and habits (Kirchhof 2013). Healthy patients are unlikely to get AF. Hence, the system can be seen with a single (normal) rhythm. With age, AF can appear from changes in the system dynamics, which eventually will lead to the establishment of two possible trajectories (healthy and AF). The CT here is the change in dynamical parameters that will lead to establishment of the disease. Eventually, the disease progresses from paroxysmal to persistent AF. The project will try to capture which information can pinpoint the changes in dynamics.
Regarding point 2), a totally different type of CT occurs when the disease is already established and the heart can switch between two stable trajectories: normal rhythm and AF. This occurs in a much faster time-scale and the CT that we try to predict is the switch between normal rhythm and AF. With enough warning time, patients could take medicines that could prevent the switch to AF. These ideas will also be explored with respect to heart attacks.
The datasets at the basis of this study are provided by Dr. Dominik Linz, clinician-scientist and lecturer at The University of Adelaide (Australia), furthermore interested in investigating the potential relation between sleep disordered breathing and heart arrhythmia.
We aim to build models from time-series data, such as ECG. For part 1) above, we will use data from regular patient visits to build dynamic models. We would then compare, not the data, but the models to see how they are changing over time, and how those changes are related from the changes of healthy to paroxysmal AF and then eventually to persistent AF. For point 2) above, of particular use are data that capture the transitions from normal rhythm to AF and back to normal. Finally, we aim to determine the best signals to capture CTs and to develop new personalised and proactive therapies to avoid developing or switching to AF. For both parts 1) and 2), we will calculate differences in dynamic models instead of just differences in expression of the data.