Predicting Complications of Percutaneous Coronary Intervention Using a Novel Support Vector Method

Clinical tools to identify patients at risk of complications during percutaneous coronary intervention (PCI) are important to determine care at the bedside and to assess quality and outcomes. We address the growing need for such tools by proposing a novel support vector machine (SVM) approach to stratify PCI patients. Our approach simultaneously leverages properties of both one-class and two-class SVM classification to address the diminished prevalence of many important PCI complications. When studied on the Blue Cross Blue Shield of Michigan Cardiovascular Consortium (BMC2) multi-center cardiology registry data, our SVM method provided moderate to high levels of discrimination for different PCI endpoints, and improved model performance in many cases relative to both traditional one-class and two-class SVMs.