Predicting Bleeding and Thrombosis Complications in Patients with Continuous Flow Left Ventricular Assist Devices

Background: Left ventricular assist device (LVAD) therapy has been proven to relieve heart failure symptoms and improve survival, but is not devoid of bleeding and/or thrombotic complications. Risk stratification tools have been utilized in other cardiovascular disease populations to estimate the risk of bleeding and thrombosis with and without anticoagulation, including the HAS-BLED, HEMORR2HAGES, CHADS2 and CHA2DS2-VASc models. The study objective was to evaluate the predictive value of available risk models for bleeding and thrombotic complications in patients with an LVAD within one year of implantation. Methods: This was a retrospective, single-center analysis of patients implanted with the HeartMate II continuous-flow LVAD from July 2011 to June 2016. All patients who received an LVAD within the study period were eligible for inclusion. The primary endpoint was the first occurrence of bleeding or thrombosis within one year from implantation. Baseline risk model scores were calculated at the time of LVAD implantation. Chi-square and student’s t-test were used to measure baseline differences and compare mean risk model scores between patients who had an event. A receiver operator characteristic (ROC) curve analysis was performed to evaluate the accuracy of the risk models to predict an event. Results: A total of 129 patients underwent LVAD implantation within the study time period. Mean CHADS2, CHA2DS2-VASc, and HAS-BLED scores were not significantly different in patients with and without an event. The mean HEMORR2HAGES score was 3.09 and 2.51 in those with and without a bleeding event, respectively (p = 0.008). The ROC curve area for the HEMORR2HAGES model was the highest at 0.620. Conclusion: The HAS-BLED, HEMORR2HAGES, CHADS2and CHA2DS2-VASc risk stratification models did not accurately predict bleeding or thrombosis events in our population. The mean HEMORR2HAGES model score was higher in patients who experienced a bleeding event. However, this model did not have strong positive predictive value. Better risk models are needed to predict bleeding and thrombotic events in this patient population