This study was conducted under a protocol approved by the Institutional Review Board of Union Hospital, Affiliated with Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China (number: 2023-0429, approval date: June 6, 2023).

The study protocol complies with the ethical principles of the 1975 Declaration of Helsinki. From May 2022 to January 2023, five patients in our center underwent Extra-LVAD placement via the right jugular vein and AA. We used aorta CTA data from one of these patients.

The contrast-enhanced CTA images were obtained from a heart failure patient with an LVAD. CTA imaging was performed using a dual-source CT scanner (SOMATOM Force, Siemens Medical, Germany, Software Version VA50). Volumetric datasets were retrospectively reconstructed using a multi-segment algorithm, generating thin-section images with a slice thickness of 0.3 mm.

The segmentation region encompassed the aortic lumen and its major branches, extending to the right axillary and femoral arteries as potential LVAD implantation sites. Anatomical model reconstruction was performed through manual segmentation and subsequent smoothing using the open-source software 3D Slicer (Version 5.3.0, National Institutes of Health, USA). The resulting vessel wall model was clipped, and inlet and outlet boundaries were generated to cap the ends, thereby defining a continuous fluid domain for computational simulation.

Blood was modeled as an incompressible Newtonian fluid with a density of 1,050 kg/m³ and a dynamic viscosity of 0.0035 Pa·s. The Reynolds-averaged Navier-Stokes (RANS) equations were solved using the Shear-Stress Transport (SST) k-ω turbulence model. This hybrid model was selected for its demonstrated efficacy in predicting flow separation under adverse pressure gradients (11), which is particularly relevant for capturing the jet flow dynamics induced by LVAD insertion, and its ability to accurately resolve near-wall flow, which is critical for reliable Wall Shear Stress (WSS) calculations (12). The model integrates the robustness of the k-ε model in the far-field with the precision of the k-ω model in near-wall regions.

The governing continuity (1) and momentum (2) equations are:∇⋅u=0 (1) ∂u∂t+(u⋅∇)u=−1ρ∇p+ν∇2u (2) Where ρ is the fluid density, u is the velocity vector field, p is the pressure, ν is the kinematic viscosity.

The spatial discretization employed a second-order upwind scheme, and the pressure-velocity coupling was handled using the Semi-Implicit Method for Pressure-Linked Equations (SIMPLE) algorithm. Solutions were considered converged at each time step when the maximum normalized pressure residual fell below 10⁻⁴. The fluid domain was discretized using a hybrid mesh, comprising tetrahedral elements in the core and three layers of prismatic elements adjacent to the wall to capture the boundary layer (Supplementary Figure S1) accurately.

A grid independence study was conducted to ensure robust results (Supplementary Figure S2). Simulations were performed on coarse (∼1 million cells), medium (∼2.2 million cells), and fine (∼5 million cells) meshes for the baseline model (without LVAD). The difference in computed WSS between the medium and fine meshes was less than 1%. Consequently, the medium-resolution mesh was adopted for all subsequent simulations (approximately 2.2 million cells for the baseline, and 2.4 million for LVAD-supported models). The near-wall resolution was controlled to achieve a y⁺ value of approximately 1, consistent with the SST k-ω model's requirements.

The time step was fixed at 1 ms to ensure temporal accuracy. Simulations were run for three cardiac cycles to achieve a periodic solution, with data from the final cycle used for analysis. The aortic wall was modeled as rigid with a no-slip boundary condition. The simulations were performed using OpenFOAM (version 11).

A physiologically representative flow waveform for the ascending aorta, corresponding to a cardiac output (CO) of 5 L/min for a healthy individual (13), was prescribed at the inlet (Figure 1A). To model the impedance of the peripheral vasculature, a tightly coupled 0D-3D scheme was implemented within OpenFOAM. This scheme solves the three-element Windkessel (0D) model equations in a tightly coupled manner with the 3D Navier-Stokes equations at each time step. A three-element Windkessel (0D) model, characterized by proximal resistance (R_p), compliance (C), and distal resistance (R_d) (14), was applied at each outlet (Figure 2, Supplementary Figure S3). The Windkessel parameters were calibrated to match reference systolic/diastolic pressures and presumed flow splits under normal conditions (without LVAD support).

A typical ascending aorta blood flow waveform corresponding to the normal time-average CO, which is assumed to be 5 L/min in this work, was imposed at the inlet, as shown in Figure 1A. The Windkessel models were applied to outlets to represent the impedance of the distal vascular bed to blood flow, as shown in Figure 2 and Supplementary Figure S3.

The LVAD pump function was simplified as a graft inlet with a constant flow rate. Two LVAD support configurations were modeled by introducing a graft at either the FA or the right AA. For each configuration, two constant LVAD flow rates were simulated: 2 L/min and 3 L/min. These were chosen to represent scenarios where the native cardiac output is, respectively, higher or lower than the LVAD output. A summary of the boundary conditions for the standard and LVAD-supported configurations is provided in Supplementary Table S1.

The Windkessel model parameters, calibrated for the baseline condition (without LVAD), were held constant for all simulations with LVAD support. This assumption is based on the established practice in hemodynamic simulations (15–18) and the hypothesis that the characteristics of the distal arterial system remain essentially unchanged immediately post-implantation when total perfusion is restored to a near-normal level (approximately 5 L/min) (19–21). This approach, necessitated by the lack of patient-specific postoperative data, ensures that the observed hemodynamic differences are attributable primarily to the LVAD cannulation site (FA vs. AA).

The Extra-LVAD maintained a flow of 1.82-3.25 LPM for all patients, which accounted for 36%–65% of the total flow, with a CO of 5 LPM (Figure 1B). Hence, we studied the hemodynamic changes caused by the Extra-LVAD in our models under two conditions: the LVAD supplying 40% (Figure 1C) and 60% (Figure 1D) of the total flow. Cardiac ejection accounts for 75.3% of total perfusion at the end-systolic phase, with LVAD covering the remaining 24.7%. Tables 1, 2 detail the distribution of blood flow in the aortic branches, with LVAD accounting for 40% or 60% of total blood flow. Figure 2 shows the location of the main aortic branches studied by our research, including the subclavian artery, common carotid artery, internal iliac artery, profunda femoris artery, femoral artery, renal artery, inferior mesenteric artery (IMA), superior mesenteric artery (SMA), and celiac trunk. Figures 3, 4 display the flow patterns. Regardless of the amount of blood flow the LVAD provides, axillary perfusion consistently outperformed femoral perfusion in most of these arteries, except at the placement site and the inferior mesenteric artery.

For the abdominal aorta branches, axillary perfusion resulted in a significant decrease in blood supply to the IMA by 44.4% at 40% LVAD flow, but this reduction disappeared at 60% LVAD flow. At 60% LVAD flow, all arteries supplying the intestines and kidneys showed increased blood flow with axillary perfusion compared to femoral perfusion. Specifically, the IMA showed no decrease as opposed to an 11.1% reduction with femoral perfusion, the SMA decreased by 2.8% compared to a 12.7% decrease, the celiac trunk increased by 0.6% instead of decreasing by 9.7% with femoral perfusion, the left renal artery (LRA) increased by 7.4% instead of a 4.2% decrease, and the right renal artery (RRA) increased by 10.8% compared to no change. In the lower extremities, femoral perfusion also led to greater reductions in blood supply. Except for the right femoral artery as the perfusion site, at 60% LVAD flow, blood supply to all lower extremity arteries decreased by more than 15% with femoral perfusion. In comparison, they experienced less than a 13% reduction with axillary perfusion. The upper extremity and carotid artery blood flow remained stable at 40% LVAD, regardless of the perfusion approach. At 60% LVAD flow, axillary perfusion significantly increased blood flow to three of the arteries, including the right subclavian artery (23.3%), the left common carotid artery (14.0%), and the left subclavian artery (12.0%). Still, femoral perfusion decreased the blood flow in these arteries. Blood flow increased notably at the catheter insertion sites, especially the right femoral, with a 151.5% increase at 60% LVAD flow with femoral perfusion.

In terms of blood flow disorders, axillary perfusion may cause turbulent flow in the brachiocephalic trunk (Figures 4B, D), while femoral perfusion may interfere with cardiac ejection in the abdominal aorta, significantly reducing blood flow to it (Figures 4A, C).

The distribution of TAWSS (Figures 5, 6) showed values exceeding 5 Pa at all points where the perfusion graft is attached. With femoral perfusion, TAWSS in the thoracic and upper abdominal aorta above the kidneys decreased significantly to less than 1 Pa, particularly when LVAD contributed 60% of total perfusion. However, in the axillary perfusion group, the distribution of TAWSS values was more similar to that of the standard group, regardless of total perfusion.

The OSI distribution (Figures 7, 8) was generally lower in axillary perfusion (less than 0.2 in the most regions of the aorta) compared to femoral perfusion (more than 0.2 in the most regions), especially in the thoracic aorta and the upper abdominal aorta. This indicates less pronounced changes in WSS direction with axillary perfusion as LVAD flow increased.