The in-vitro test setup was performed according to [20], and the numerical in-silico methods for the CFD investigations used the DP3 (Xenios AG, Heilbronn, Germany) as an example of a blood pump in current clinical use.
Flow loop design
To perform in-vitro hemolysis testing, three identical flow loops plus one reference reservoir—for static conditions—were prepared for each test day. The test loops were assembled by three segments of PVC tubing (3/8″ × 3/32″ RAUMEDIC-ECC-noDOP®, Raumedic AG, Germany). A smaller welded medos soft blood bag (Medos Medizintechnik AG, Germany) with 204 mL ± 28 mL was used as a closed reservoir and connectors with and without Luer (FLEIMA-PLASTIC GmbH, Germany) were used in between. Spacers with 3D printed tube and connector clips were added to keep the bending radii of the tubes equal and to guarantee the laminar flow inlet distance for the flow sensors in each simultaneously running flow loop, see Fig. 1. The tests were repeated four times at 4 L/min (n = 4), and five times at 1 L/min (n = 5).
To avoid results that would correlate with the specific pump head, the serial numbers were tracked and the pump heads rotated through the three flow loops at each repetition.
Blood collection and, anticoagulation
Five one-liter blood bottles were prepared with 1.8 mL glucose (0.5 mg/mL), 1.6 mL gentamycin (10 mg/mL), and 50 ml NaCl (0.009 g/mL). While 15,000 IU/L heparin was initially used at both low and high-flow, heparin was reduced to 4500 IU/L due to the high activated clotting time (ACT) values. Each bottle was filled with one liter of porcine blood at the slaughterhouse and immediately transferred to the blood lab, filtered through a nylon stocking and pooled in a five-liter bag (Nutrimix®, B. Braun Melsungen AG, Germany). The hematocrit of the pooled blood was determined after gentle, but thorough mixing.
Flow loop preparation
The flow loops were prefilled with saline solution (< 500 mL) to displace all remaining air. As much saline solution as possible was drained from the flow loops and replaced with pooled blood. Based on the weighted remaining volume of saline solution, the necessary blood volume to achieve a total volume of 454.1 mL in the first in-vitro test and in all following ones 483.94 mL ± 1.25 mL (mean ± SD) at a hematocrit of 31.16 % ± 1.12 % was added to the flow loops. Subsequently, all air was removed from the flow loops. The pumps were operated at low speeds of 1252 rpm on average, achieving a homogeneous mixing of the blood. Subsequently, a sample was drawn in order to verify the hematocrit, determine the standard base excess and finally adjust the standard base excess to 0 mmol/L ± 5 mmol/L using 8.4 % sodium bicarbonate solution (Fresenius Kabi, Germany). A reference reservoir was filled with the same pooled blood, therefore having equal hematocrit, and placed in one heated water bath under static conditions. With the exception of the pump head and the flow sensor, each flow loop was placed in a water bath (Lauda, Germany) at 37 °C (36.78 °C ± 0.57 °C). As soon as the base excess was adjusted, the pump speed was set to the target speed and pressure difference was adjusted with the Hoffmann hose clamp as a resistance. Samples were taken from the flow loops every 60 min. Sampling volume was 4 mL with a previous discard of 2 mL. After six hours test duration, the pump heads were thoroughly cleaned with a pepsin citrate solution, rinsed with de-mineralized water, dried and stored for the next test.
Multi-center studies reveal widely varying results due to many patient-specific parameters. In order to avoid this, we focused on high repeatability in the in-vitro tests. Further details on sample and data acquisition can be found in the Appendix.
Pump operating points
In this study, an upper pressure head target of 250 mmHg was chosen to be consistent with typical CO2 removal applications [10]. Two different pump flows were chosen, as those typically used in low-flow and high-flow applications of ECMO or ECCO2R [21, 22].
The low-flow operating point was set to 1 L/min (0.96 L/min ± 0.04 L/min, mean ± SD) and the pressure head to 250 mmHg (249.77 mmHg ± 5.58 mmHg), requiring a pump speed of 6250 rpm (6259 rpm ± 26 rpm). When the operating point was set to 4 L/min (3.94 L/min ± 0.04 L/min) using the same rotational speed of 6250 rpm (6226 rpm ± 78 rpm), the pressure drop measured 206.61 mmHg ± 7.32 mmHg.
In order to achieve high comparability between in-vitro and in-silico, the CFD simulations were adjusted according to the in-vitro average values of hemoglobin and hematocrit.
Hemolysis measurement
In order to compare the pumps operating points based on a clinically relevant parameter, we evaluated the hemolysis rate, i.e. how much delta plasma free hemoglobin (ΔpfHb) in mg per min was produced in the flow loops. In the comparison of pumps, both the Normalized Index of Hemolysis (NIH) and the Modified Index of Hemolysis (MIH) are established. In this study, we refer mainly to MIH, as it is normalized to the total hemoglobin content. This is meaningful since the latter varies from in vitro test to in vitro test. The formulas are based on the ASTM F1841-97(2017) standard [20] and on the publication of Adachi et al. [23] described in the Appendix, compare equations (1), (3) and (4).
$$MIH = \frac{{\Delta fHb \cdot V \cdot \left( {\frac{100 - Ht}{100}} \right)}}{Q \cdot T \cdot Hb}$$
(1)
Compare [20]
\(\Delta fHb\) increment of plasma free hemoglobin concentration in \({\text{mg}}\;{\text{dL}}^{ - 1}\)
\(V\) whole blood volume in flow loop in \({\text{mL}}\)
\(Ht\) hematocrit in \(\%\)
\(Q\) flow rate in \({\text{L}}/{\text{min}}\)
\(T\) sampling period in \({ \text{min} }\)
\(Hb\) total hemoglobin in \({\text{g}}/{\text{dL}}\)
The free plasma hemoglobin concentration was evaluated photometrically at 540 nm and 680 nm as reference wavelengths according to the gold standard, DIN 58931:2010-08, using the cyanmethemoglobin (HiCN) method [24, 25]. For further information on double determination and other aspects of this procedure, the reader is referred to the Appendix.
Numerical CFD setup and in-silico hemolysis evaluation
The simulation setup was aligned with Gross-Hardt et al. [10] to ensure comparability of the results. Micro-CT scans were performed to determine the geometry of the DP3 pump head. The geometry mesh was generated using tetrahedral elements, which merge into 18 refining prismatic wall layers with a prism growth rate of 1.2, resulting in 6.73 million mesh-elements for the 18.1 mL internal blood volume of the pump head [26].
In order to determine hemolysis numerically, the following simulation parameters are required: The time during which blood is exposed to a scalar shear stress, called exposure time. The three-dimensional shear stress tensor is converted into a scalar shear stress and, using the present flow, transformed into a hemolysis index via the well-known power-law relationship. The numerical hemolysis prediction was analogously performed to Gross-Hardt et al. [10]. By means of the commercial element-based finite volume method (ebFVM), the solver CFX (ANSYS CFX, ANSYS, Inc., Canonsburg, PA, USA) and the sliding mesh approach, the transient Reynolds-averaged Navier–Stokes (RANS) momentum and mass equations was iteratively solved. Using scalar variable residuals and stabilized predictions of the simulation parameters, the convergence was monitored until the simulation parameters showed stable results. In order to achieve transient stability, at least five complete impeller rotations were computed before transient result were averaged over the following two rotations. In all simulations, the time step was chosen proportional to 5° of the impeller rotation, resulting in 504 iterations for each operation point, of which 144 were used for statistical result averaging. Blood density was adjusted to 1059 kg/m3, while the viscosity was modelled as shear-dependent [27]. Blood-specific and varying parameters, like hemoglobin and hematocrit were averaged over all in-vitro tests and fed back into the simulation to create consistent conditions.
CFD—recirculation
Common centrifugal blood pumps have secondary flow within the pump housing to wash out bearings and avoid regions of stagnation, which are prone to thrombus formation. However, the amount of secondary flow is influenced by the geometry and the operating conditions. Recirculation within the pump head is defined as the ratio between the sum of secondary flow in the pump casing, normalized to the flow leaving the pump’s outlet, compare equation (2).
$$Recirculation_{ratio} = \frac{{\sum \dot{V}_{secondary} }}{{\dot{V}_{{pump^{\prime}outlet}} }}$$
(2)
The slopes in Fig. 2 of the ∆pfHb regression lines were tested for significance with a two sided heteroscedastic t-test. Statistical comparison of in-vitro MIH values with in silico MIH values per flow rate were calculated by means of one sample t-test (†‡). Statistical comparison of mean MIH in-vitro values were calculated by a two-sided heteroscedastic t-test (*) (MS Excel, Microsoft). Significance level of p = 0.001 was assumed.