MACHINE LEARNING AND COMPUTATIONAL CARDIAC MODELING FOR CLINICAL DECISION SUPPORT IN PATIENTS WITH CARDIOVASCULAR DISEASE

Cardiovascular diseases are the leading cause of mortality around the world. Many of these deaths result from abnormal heart rhythms, or arrhythmias, including ventricular tachycardia (VT) and atrial fibrillation (AF). Effective treatments and preventative measures for VT and AF include prophylactic implantable cardioverter defibrillator (ICD) therapy and pulmonary vein isolation (PVI) ablation, respectively. However, ICD therapy comes with a significant risk of adverse events, and PVI is not effective for all patients. For many patient populations, it is unknown how to identify the patients at highest risk for arrhythmia and other adverse cardiac events, and it is difficult to predict how each patient will respond to therapy. In recent years, multi-scale (i.e. from cellular to the whole heart) computational cardiac modeling has emerged as a way to predict arrhythmia risk by investigating the susceptibility of a patient’s fibrotic substrate to sustaining reentrant activity. These imaging-based models may have the potential to assess VT risk in previously unexplored patient populations, such as patients with structurally abnormal hearts and patients with severe localized cardiac inflammation. Here, cardiac models that represent these pathophysiologies are developed and validated, thus expanding the population of patients for whom cardiac modeling is an applicable risk prediction tool. Machine learning (ML) classifiers are ideal for identifying predictive patterns in high-dimensional data, so I have developed an arrhythmia risk prediction methodology that combines ML and mechanistic modeling and applied it to cardiac conditions for which the previously developed modeling-based VT risk prediction methodology is not appropriate. Finally, since little is known about the pathophysiology of cardiovascular complications in COVID-19, I have developed a continuously updating ML classifier to predict which hospitalized COVID-19 patients are at risk of impending adverse cardiovascular events. In conclusion, I have engineered computational risk prediction tools that use mechanistic cardiac modeling, ML, or a combination of the two, and demonstrated their utility in predicting the course of a variety of cardiac diseases to guide clinical decision making. The merging of computational modeling and data science with medicine, as exemplified here, is poised to lead to major advances in the improvement of patient care