A New Hemodynamic Profile Signaling Early Death on Left Ventricular Assist Device

Purpose To assess for a post left ventricular assist device (LVAD) hemodynamic profile predictive of early mortality. Methods A single center first-time, continuous-flow LVAD dataset was utilized for this analysis (n=505). Vasoactive drips and hemodynamics were collected at the following time points in all subjects with available data: pre-op, immediate post-op, and on post-operative days 1, 2, and 3 (n=325). Vasoactive intropes scores (VIS) were calculated at each time point. Trends before and after LVAD were computed and plotted using Locally Estimated Scatterplot Smoothing (LOESS), stratfied by death at 60 days post LVAD implantation. Mutivariate logistic regression and reciever operator curve analyses were performed to test the association between candidate variables and death at 60 days. Results The distinctive hemodynamic feature of patients who experienced death by 60 days was a failure to increase mean arterial pressure (MAP) from the time of leaving the operating room to post-operative day two accompanied by persistently elevated VIS (Figure). A model incorporating post-operative day two VIS, mean arterial pressure and right atrial pressure had an AUC of 78% to predict 60-day death. The group experiencing early mortality had higher measured cardiac indices, lower systemic vascular resistance, and higher right atrial pressures than patients who survived past 60 days, suggesting overlap between classic right ventricular failure and vasoplegia physiology. In this dataset, half of patients who met the criteria for vasoplegia also met criteria for right ventricular failure (21/41, 51%). Conclusion The failure to increase mean arterial pressure accompanied by high doses of inotropic and vasopressor support early after LVAD is highly associated with early mortality. This hemodynamic profile may be more unifying than existing right heart failure and vasoplegia definitions, which appear to have significant overlap. These findings require validation in an external dataset.