A Computational Fluid Dynamics Study of the Extracorporeal Membrane Oxygenation-Failing Heart Circulation

Abstract

Extracorporeal membrane oxygenation (ECMO) is increasingly deployed to provide percutaneous mechanical circulatory support despite incomplete understanding of its complex interactions with the failing heart and its effects on hemodynamics and perfusion. Using an idealized geometry of the aorta and its major branches and a peripherally inserted return cannula terminating in the iliac artery, computational fluid dynamic simulations were performed to (1) quantify perfusion as function of relative ECMO flow and (2) describe the watershed region produced by the collision of antegrade flow from the heart and retrograde ECMO flow. To simulate varying degrees of cardiac failure, ECMO flow as a fraction of systemic perfusion was evaluated at 100%, 90%, 75%, and 50% of total flow with the remainder supplied by the heart calculated from a patient-derived flow waveform. Dynamic boundary conditions were generated with a three-element lumped parameter model to accurately simulate distal perfusion. In profound failure (ECMO providing 90% or more of flow), the watershed region was positioned in the aortic arch with minimal pulsatility observed in the flow to the visceral organs. Modest increases in cardiac flow advanced the watershed region into the thoracic aorta with arch perfusion entirely supplied by the heart.

Extracorporeal membrane oxygenation (ECMO) is increasingly used to provide mechanical circulatory support (MCS) despite incomplete understanding of its complex interactions with the failing heart and its effects on hemodynamics and perfusion.^1–3 When used for circulatory support, ECMO shunts venous blood through an oxygenator via a mechanical pump to return oxygenated blood to the arterial system.⁴ Most commonly in adult patients in cardiogenic shock, the return cannula is inserted into the femoral artery with the distal tip terminating in the iliac artery or distal aorta depending on patient anatomy. This cannulation strategy produces continuous retrograde perfusion by the ECMO circuit of the aortic tree above the cannula which collides with pulsatile antegrade perfusion from the failing heart to create a dynamic watershed region.⁵ The watershed location, size, and effect on end-organ perfusion are poorly understood as is the impact of relative contributions of flow provided by the ECMO circuit and the failing heart on systemic hemodynamics.

Computational fluid dynamics (CFD) is an ideal method to investigate the ECMO-failing heart circulation. Computational fluid dynamics enables study of varying levels of relative support from the ECMO circuit on perfusion of the major branching vessels off the aortic tree that is difficult to replicate in vivo. While prior CFD studies of ECMO provided insight into the size and behavior of the watershed region,^6,7 these studies relied on fixed boundary conditions making them not applicable to evaluating the key question of how varying levels of ECMO support alter distribution of perfusion.

In this article, a CFD model of the ECMO-failing heart circulation is presented that applies dynamic boundary conditions through the use of a three-element lumped parameter model to simulate the effect of varying levels of extracorporeal support on end-organ perfusion. Perfusion to each of the major branches of the aortic tree is quantified in the setting of variable ECMO support to simulate differing degrees of heart failure and relate these findings to the steady-state case of normal cardiac function without extracorporeal support. The effect of ECMO support on the watershed region is evaluated and how these findings relate to clinical findings and may provide insights into the clinical management of patients supported by ECMO are discussed.

METHODS

Anatomical Geometry

A previously reported idealized geometry of the aortic tree was implemented as the basis of the simulation of the vascular space.⁸ The idealized aorta geometry was created by averaging in vivo images and postmortem aorta measurements and included major aortic outlets. The lumen of aorta extends from the sinotubular junction to the bifurcation of the iliac arteries and includes the aortic branches of brachiocephalic artery (BCA), left common carotid artery, left subclavian artery, celiac artery, superior mesenteric artery, left and right renal arteries, inferior mesenteric artery, and left common iliac (LCI) and right common iliac. Clinically-relevant cannulation was done geometrically using a standard cannula tube (17 French) modeled to represent peripheral cannulation placed 15 cm distal to aortoiliac bifurcation inside the LCI artery.

Grid Generation

A polyhedral based volumetric mesh, consisting of 555,492 elements, was generated using the Fluent Meshing package (ANSYS v. 17.2, Canonsburg, PA). Polyhedral cells were used as they provide more robust computational results that are more accurate, less sensitive to geometry tortuosity, and computationally less expensive grids compared to standard tetrahedral meshes.⁹ The computational domain was refined according to geometrical features such as curvature, outlet diameter, and branching.

Computational Method

Methods of CFD were used to simulate the three-dimensional transient flow field using ANSYS Fluent (ANSYS) (Figure 1A). For the purposes of CFD simulation, blood was modeled as an incompressible Newtonian fluid with constant density and dynamic viscosity of 1,060 kg·m⁻³ and 3.5 cP, respectively.¹⁰ In vivo, blood behaves as a non-Newtonian fluid with complex shear-thinning behavior which is an important consideration in smaller vessels such as coronary arteries. However, in large vessels such as aorta, where the shear rates are >300 s⁻¹, modeling blood as a Newtonian fluid is a reasonable approximation that provides both accuracy and computational efficiency.^11,12 A multiphase Volume of Fluid model with implicit formulation was used to distinguish between highly-oxygenated ECMO flow and blood delivered by the heart after passing through the cardiopulmonary circulation. This approach enables accurate prediction of the interface and mixing of retrograde ECMO flow and antegrade flow from the heart to differentiate the share of each perfusing outlets from the aortic tree. The shear stress transport turbulent model was applied to accurately capture turbulent flow patterns created by the high-speed jet of blood ejected from the small ECMO catheter.

Figure 1.

A: Constructed polyhedral volume mesh in different views and cross-sections. B: Coupling the three-element lumped-parameter dynamic boundary condition with the idealized 3D computational cannulated model.

Blood was defined as entering the systemic circulation both from the ascending aorta and ECMO cannula. The ventricular output was generated from a patient-derived physiologic aortic waveform and applied at the sinotubular junction.¹³ Extracorporeal membrane oxygenation flow was assumed to be fixed based on the design and typical operation of clinically used centrifugal pumps common to most circuits and was modeled as a steady jet from the cannula. To simulate a patient stably supported on ECMO, a total perfusion of five LPM was assumed, to approximate normal cardiac output of a 70 kg adult. The relative fraction of flow provided by the ECMO circuit was varied (50%, 75%, 90%, and 100% of total cardiac output) to model differing severity of heart failure. A healthy case with 100% flow provided by the heart was simulated to provide a control for comparison with the ECMO-failing heart scenarios.

Boundary Conditions

Dynamic boundary conditions at the outlets were employed to determine the effect of changing flow patterns on perfusion and were modeled using a three-element lumped parameter model of the distal vascular impedance (Figure 1B).^8,14,15 This zero-dimensional electrical analogy of the hemodynamic system ensures consideration of characteristic and distal resistance and compliance of the vasculature for each aortic outlet. The relationships between flow rate and pressure at each branch are described by the following differential equation:

$$\frac{\partial p}{\partial t}+\frac{p}{C{R}_d}=\frac{Q}{C}\left(1+\frac{R_p}{R_d}\right)+R_p\frac{\partial Q}{\partial t}$$ (1)

where Q is flow into the lumped element, p is pressure at the inlet, and R is resistance (with subscript p for proximal and d for distal component) and C the compliance of the lumped element. Parameters for each artery are summarized in Table 1.^8,13 This equation was solved and updated for each outlet branch at every time step.

Table 1.

Parameters of Three-Element Lumped Parameter Model for Aorta^12,15

Artery	Rp (107 Pa·s·m−3)	Rd (108 Pa·s·m−3)	C (10−10·m3·Pa−1)
BCA	5.192	10.608	8.697
CA	11.762	7.573	12.184
IMA	74.017	46.225	1.996
LCCA	19.152	52.213	1.767
LCI	5.915	10.174	9.069
LRA	34.138	5.395	17.102
LSA	9.882	13.018	7.087
RCI	5.915	10.174	9.069
RRA	34.138	5.395	17.102
SMA	17.435	5.510	16.745

R_p and R_d denote proximal and distal resistances, respectively and C is the compliance component of the lumped element.

BCA, brachiocephalic artery; CA, celiac artery; IMA, inferior mesenteric artery; LCCA, left common carotid artery; LCI, left common iliac; LRA, left renal artery; LSA, left subclavian artery; RCI, right common iliac; RRA, right renal artery; SMA, superior mesenteric artery.

Computational analyses were conducted at a local parallel processing cluster. Hemodynamic features, stress metrics, and end-organ perfusion were measured for each simulation.

RESULTS

A computational analysis of the ECMO-failing heart circulation using an anatomical model of the aortic tree was performed. Hemodynamics and perfusion were quantified at varying levels of relative ECMO flow used to simulate different degrees of heart failure.

Watershed Region Dynamics

Flow patterns and watershed region dynamics vary significantly for different cases of volumetric blood support provided by the ECMO circuit (Figure 2 and Video1, Supplemental Digital Content 1, http://links.lww.com/ASAIO/A519). Decreasing the proportion of volumetric ECMO flow resulted in shifting the watershed region distally within the aorta. In the setting of concomitant pulmonary disease, in which insufficiently oxygenated blood ejected from the heart, the location of the watershed region identifies the anatomical regions at risk for clinically significant hypoxia. In the case of 100% ECMO support (Figure 2A) with no forward flow through the cardiopulmonary circulation (representing profound heart failure), perfusion of the aortic arch is entirely derived from the ECMO circuit with an area of stasis directly distal to the aortic valve. As ECMO support decreases to 90% (Figure 2B), the watershed region moves distally but remains in the aortic arch. At this level of support, blood ejected from the heart provides perfusion to BCA and right-sided structures. A modest decrease in ECMO support to 75% of total perfusion (Figure 2C) results in movement of the watershed region out of the aortic arch and to the level of the superior mesenteric arteries. Decreasing ECMO support to 50% of total perfusion (Figure 2D) pushes the watershed region distally where it engages with the renal arteries in addition to the superior mesenteric arteries.

Figure 2.

Snapshot of ECMO blood volume fraction in various ECMO support levels of (A) 100%, (B) 90%, (C) 75%, and (D) 50% at the peak of systole. ECMO, extracorporeal membrane oxygenation.

A critical dynamic for the watershed region over the cardiac cycle was observed (Figure 3) across simulations in which the fraction of total blood flow provided by ECMO determined the proximal location of the watershed region. The vicinity of major aortic outlets to the watershed region may be a determining factor of vital organ malperfusion and may predict clinical thromboembolic events. Although oscillating, the location of the watershed region is limited to a confined region, computationally predicted based on the ECMO blood volumetric flow relative to the total flow. The oscillations of the watershed region at the site of major aortic outlets, as is the case for the renal arteries in ECMO support of 50% (Figure 3), demonstrate the challenge of clinically predicting renal perfusion and knowing the source, cardiac versus ECMO, of end-organ perfusion. In addition, a more diffuse and elongated pattern of mixing in systole compared to diastolic period was observed as acceleration and deceleration of blood flow imposed by the heart supply considerably stretches the watershed region due to hemodynamic forces.

Figure 3.

Dynamics of mixing zone during one cardiac cycle for ECMO support level of 50%. ECMO, extracorporeal membrane oxygenation.

Extracorporeal Membrane Oxygenation Flow and End-Organ Perfusion

Analyzing the perfusion of different major outlets—grouped by their location in the proximal, abdominal, near-renal, and distal aorta—offers hemodynamic insights in terms of flow pulsatility and vital organ malperfusion (Figure 4). In the baseline case simulating normal cardiac function with no extracorporeal support, mass flow rate peaks in systole with modest reversal of flow in early diastole following valve closure in the proximal aorta (Figure 4A). This oscillating behavior is moderated in the abdominal aorta and more pronounced near the distal outlets (Figure 4B–D). However, as the volumetric contribution of ECMO is increased, and in the most extreme case of pure ECMO supply, pulsatility of flow is significantly suppressed throughout the aortic tree. This dampening effect of ECMO is more apparent in distal perfusion patterns as ECMO support decreases to 50% (Figure 4D). Pulsatility suppression is less pronounced in the abdominal aorta compared to proximal and distal outlets due to blood dynamics imposed by the geometry and vascular impedance.

Figure 4.

Comparison of pulsatility of mass flow rate waveform for various ECMO level supports. ECMO, extracorporeal membrane oxygenation.

In the setting of pulmonary disease, the perfusion pattern generated by varying the contribution of ECMO support is a determining factor of hypoxia in vital organs. As the volumetric contribution of ECMO decreases, malperfusion in the abdominal aorta may result followed by cerebral hypoxia in severe heart failure. Computational hemodynamics more thoroughly depict this concern by illustrating mixing dynamics and localization of the watershed region near different outlets demonstrating the risk of impaired flow of oxygenated blood to vital abdominal organs.

The mass flow rate into each aortic outlet provided by the heart and the ECMO circuit was determined for each level of ECMO support (Figure 5). These in-detail charts of flow distribution also demonstrate the importance of alteration in blood flow pulsatility and the location and dynamics of the watershed region in determining the perfusion of vital organs. In the setting of minimal heart function (ECMO flow 90% of total), brain perfusion is a complex contribution of heart and ECMO-derived flow leading demonstrating the risk of cerebral hypoxia in the setting of concomitant pulmonary disease. Not until the heart provided 25% of total mass flow is pulsatility observed. At this level of cardiac function, only modest contributions from the heart to abdominal and renal perfusion were observed. In the case of equal contributions from the heart and ECMO circuit, significant perfusion from either source is produced but in different periods of the cardiac cycle. At this level of support, the watershed region is highly dynamic with the resulting intricate pattern of perfusion possibly being a source of clinical complications.

Figure 5.

Details of systemic perfusion for (A) a healthy case, ECMO support levels of (B) 100%, (C) 90%, (D) 75%, and (E) 50%. Portion of heart and ECMO blood in each artery (left), mass flow rate waveform of ECMO blood in each artery (middle) and mass flow rate from heart in each artery (right). ECMO, extracorporeal membrane oxygenation.

DISCUSSION

This study aims to advance understanding of the ECMO-failing heart circulation—specifically investigating how varying levels of support provided by ECMO alters distribution of perfusion in the aortic tree and impacts the location and characteristics of the watershed region. Clinical use of MCS devices has grown rapidly without a corresponding increase in research to provide clinicians with the necessary tools to optimize mechanical support.^2,16,17 Extracorporeal membrane oxygenation is a complex support modality due in part to altered hemodynamics created by retrograde perfusion of the aorta by the ECMO circuit colliding with the antegrade flow generated by the failing heart.^18–21 The watershed region at which these two flows interact has unknown effects on systemic perfusion with poorly defined risks of complications.

In order to quantify the effect of differing levels of ECMO support on end-organ perfusion, a constant total flow was simulated to represent a patient stably maintained on support while modulating flow provided by each component of the ECMO-failing heart circulation. Extracorporeal membrane oxygenation flow was modeled as a fixed flow rate consistent with flow provided by standard, clinically available centrifugal pumps. Inducing pulsatility in the ECMO flow is an area of active investigation but is not considered in this work given its lack of present use.^22,23 In clinical practice, total flow is not readily obtainable as current diagnostics are unable to accurately determine the cardiac contribution to total flow for a patient maintained on ECMO. For this simulation, a physiologically relevant total flow of five LPM was assumed, approximately the required amount of perfusion for homeostasis in the prototypical 70 kg patient.²⁴ The fraction of flow between the ECMO circuit and the failing heart was varied to evaluate the effect on systemic hemodynamics.

A key finding of this work is the importance of even modest contributions from the failing heart to total flow on maintaining systemic pulsatility and imposing hemodynamics with a semblance of those seen in normal health. In the setting of complete loss of cardiac function, ECMO supplies all systemic perfusion and creates an area of stasis in the aortic root. This leads to a high likelihood of thrombus formation which has previously been observed clinically with fatal consequences.^25,26 In the setting of modest cardiac flow constituting only 10% of total perfusion, flow to the BCA was almost entirely supplied by the heart. This finding confirms the clinical practice of monitoring the right side of the body for concerns of hypoxia as monitoring on the left side may fail to identify potentially significant cerebral hypoxia.¹⁸ Of note, increasing the cardiac fraction of flow to only 25% of the total resulted in displacing the watershed region distally from the aortic arch and into the thoracic aorta. The clinical significance of this is decreased risk of potential thromboembolism derived from the ECMO circuit. The large extracorporeal surface area is a constant clot risk, even in the setting of systemic anticoagulation, with distribution of blood flow derived from the ECMO circuit determining likely site of embolic injury.²⁷ Increasing the cardiac flow to 25% of the total pushes the watershed region distal to the aortic arch acting to decrease risk of stroke while on ECMO support.

The location of the watershed region itself is a risk factor for potential malperfusion. The ECMO-failing heart circulation is profoundly nonphysiologic with the watershed region formed by the collision of antegrade perfusion from the heart and retrograde perfusion from the ECMO circuit serving as the ultimate representation of the dysregulation of the new circulation. The location and size of the watershed region varies over the cardiac cycle as the pulsatile flow from heart acts to propel the mixing zone distal in the aorta during systole only to have it move proximal during diastole. The anatomical location and extent of the watershed region provide insights into potential risks for organ malperfusion and identification of outlet vessels at risk for impaired perfusion and thromboembolism. Further insight into how the relative titration of ECMO support affects the watershed region will help guide the clinician in mitigating risks and optimally managing patients supported on ECMO.

For clinicians caring for ECMO patients, numerous physiologic targets are chosen to guide therapy but lack evidence to support specific parameters. Clinicians make determinations on target systemic pressure, volume status, and ECMO pump RPMs in an attempt to achieve a goal flow rate through the circuit while also determining the indications and titration parameters for vasopressors and inotropes—all of which affect the fraction of total perfusion supplied by the failing heart.¹⁸ This work demonstrates the importance of maintaining pulsatility and forward flow through the heart to propel the watershed region distal to the aortic arch and to maintain some degree of pulsatile physiologic flow to the visceral organs.

The issue of blood flow pulsatility in clinical management becomes increasingly relevant as patients are maintained on ECMO for longer periods of time and in light of recent changes to allocation of donor heart organs in the United States which prioritizes patients maintained on temporary mechanical support.²⁸ The effect of loss of pulsatility may induce vascular changes, such as those seen in patients supported by ventricular assist devices, which may risk further complications and the potential for bleeding in the setting of highly invasive surgery.^29,30 How to mitigate these risks while maintaining end-organ perfusion will become of increasing importance as ECMO use continues to expand to new indications and therapeutic intent.

LIMITATIONS AND FUTURE WORK

Computational fluid dynamics is a powerful tool capable of providing deep insight into different physiologic states while enabling independent titration of individual parameters. Using the inherent strengths of this technique, the importance of maintaining a cardiac component to the overall perfusion was demonstrated in order to maintain modest pulsatility. Even the limited pulsatility produced at a cardiac output of 25% of total perfusion was shown to produce pulsatile flow to the abdominal organs while also advancing the watershed region out of the aortic arch and into the thoracic aorta. This is significant as it decreases the theoretical risk of ECMO-circuit derived thromboembolism to the brain. Such risk to the abdominal and renal arteries remains, however, with our results confirming for the clinician to be observant for potential evidence of abdominal malperfusion or signs of embolism.

While this work provides helpful insights into the effects of variable support on end-organ perfusion, there are limitations to this study. It used a representative model of aortic and vascular anatomy to focus on general aspects of ECMO and better visualize the outcome, independent of patient-specific features. However, patient anatomy varies significantly and the impact of this variability on ECMO perfusion and mechanical support is not understood and demands investigation. Additionally, to assign boundary conditions, characteristics of physiologic cases were employed. Limited experimental benchtop data and a lack of extensive published values reduce the ability to model the range of physiologic states observed in clinical practice. Changes in vascular impedance from baseline, such as vasodilatory states that occur commonly during initiation of extracorporeal support or increased resistance in the setting of high dose vasopressor support, were not evaluated in this study. The primary motivation of this study was to model the distribution of perfusion for a stably supported patient at steady-state. Investigation of non-baseline impedance states through modulation of boundary conditions is one potential avenue for future work. At present, a lack of validated values for the lumped parameter model to simulate extremes of vascular impedance must be addressed before studying non-baseline conditions.

More importantly, further systemic study of vascular resistance and its effect on flow patterns warrant examination of fluid-structure interactions (FSI) and compliance of arterial tree which will be a focus of future work. In the current work, the arterial walls are assumed to be rigid and without changes in volume. One major challenge to considering vascular wall motion, in addition to modeling complexity and extremely high computational cost, is the range of pathologies generally associated with ECMO patients. The significant disparity of mechanical environment and physical state between physiologic and pathophysiologic cases, patient-specificity of modeling parameters, and lack of experimental data significantly limits considering a compliant wall. The extreme burden of FSI simulation may be worthwhile if numerical simulations alone would be used in a clinical decision-making procedure for an individual case with some underlying condition that might put him/her at risk should the patient proceed with ECMO.

Features associated with cannulation, including the cannula design parameters (e.g., shape, length, and side holes) and procedural factors such as the insertion location, significantly affect ECMO flow patterns and likely impact patient outcomes.^31,32 However, to generalize the results and not distract the readers, we deferred to future work where more focused research would be devoted to procedural optimization, design enhancement, and efficacy/safety assessment in patient-specific analyses. Ultimately, as with all CFD studies, the findings presented here require validation, using either a benchtop or in vivo model, to determine the effects of variable flow and assess how the watershed region affects end-organ perfusion. Leveraging such validated CFD models would offer a competent platform to not only enhance ECMO performance and improve the clinical outcome, but also to replicate different clinical scenarios that life support systems are operating in parallel to other cardiovascular devices/systems which presently are understudied and demand further investigation.