Abstract 1003: Blood Pressure Metrics In Predicting Successful Thoracic Endovascular Aortic Repair

Abstract

Blood pressure (BP) is a well-known risk factor for propagation of type B aortic dissections; however, there is limited data on its impact following surgical repair. While metrics such as BP variability have been associated with higher risk of cardiovascular events, such metrics has not been explored post TEVAR where management has largely focused on reducing systolic BP. By pairing unique BP metrics with preoperative aortic geometry, which has been shown to predict post-TEVAR outcomes, we can better understand what BP metrics are most useful. BP data collected in the first 24 hours after TEVAR for a cohort of 31 patients was used to calculate 24 unique BP metrics, including measures of centrality, variability, and stationarity. Preoperative CTA imaging data were used to create surface models for each aorta, which were then used to define aortic geometry as a unique point in the validated shape-size feature space. Logistic regression models with the BP and aortic geometry metrics were created to predict successful TEVAR outcomes using standard feature selection procedures and were validated using cross-validation (CV). Training and cross-validation accuracy were then compared between the models to determine the most generalizable model. A simple model using only preoperative aortic geometry in the two-dimensional shape-size feature space yielded a training accuracy of 77.4% and CV accuracy of 74.8%. An overfit model with all 24 BP metrics calculated from all available BP data (18.1 ± 6.0 BPs, range 5 - 29) had a 100% training accuracy but CV accuracy of 77.6%. Using the coefficients from this overfit model, we selected the most important features to train the next model. This resulted in a model using aortic geometry in addition to two systolic BP stationarity metrics and diastolic BP standard deviation. This model yielded a 96.8% training accuracy and 93.8% CV accuracy. Adding BP metrics creates a novel multi-dimensional aortic geometric-physiologic feature space and improves our ability to predict post-TEVAR outcomes. The BP variables that best contributed to this model were all measures of variability and stationarity, indicating that trends in BP must be evaluated to better identify patients at risk of poor outcomes.