Computational Cardiology: Improving Markers and Models to Stratify Patients with Heart Disease

Heart disease is the leading cause of death around the world, claiming over 17 million lives each year (30% of all global deaths). The burden of heart disease can be attributed, in part, to the lack of clinically useful tools that can accurately stratify patients and match them to appropriate therapies. In this thesis, we explore the use of computation as a solution to this problem. Specifically, the goal of our work is to develop novel approaches that can be applied to cardiovascular datasets to discover diagnostic markers and to improve models for predicting adverse cardiovascular outcomes. Our research focuses on the following opportunities: (1) improving the computational efficiency of existing ECG markers while maintaining clinically useful discrimination; (2) developing new ECG markers based on short-term heart rate structure that are complementary to existing markers; (3) building more accurate models in the presence of small training cohorts with class-imbalance; and (4) proposing approaches to decompose ECG signals into atrial and ventricular components to predict arrhythmias arising from specific anatomical regions. When evaluated on multiple cohorts comprising patients with coronary artery disease and patients undergoing cardiothoracic surgery, our work substantially improves the ability to deliver cardiac care.