Abstract

Introduction: Left ventricular assist devices are used as a possible bridge to transplant or destination therapy for advanced heart failure patients. After LVAD implantation, patients require long-term follow-up care, including carefully balanced anticoagulation, dressing changes to limit infection, and monitoring of end organ function. Due to the mechanical device, patients are also at increased risk for bleeding, hemolysis, thrombosis, and stroke, among other adverse events. This study aims to model the post-operative care pathway of patients after LVAD implementation using INTERMACS data. Methods: Sequential clinical data of 107,440 patients from the time of LVAD implantation to termination of support were abstracted from the INTERMACS data between 2010 and 2015. In the first step, the sequential clinical data for each patient was transferred into the sequences of labeled nodes such that each node represents a visit's information at a specific point of the time after LVAD implant. After forming a sequence of visits for each patient, patients are clustered into the three distinct subgroups of patients whose sequences of visits demonstrated the most similarity using hierarchical clustering. The similarity between sequences of visits is measured based on the longest common subsequence. Results: The results from clustering have shown that patients' care pathways after LVAD implant mostly differ based on occurrence of post-implant adverse events such as bleeding. In addition, patients with similar outcomes such as death are categorized in the same groups, which means similar care pathways ended in the same outcomes. Future work is to conduct sequence mining to determine the most common care pathway in each subgroup. The result of this study will be used to improve care management and medical decision making after LVAD implant.