Extracorporeal Membrane Oxygenation (ECMO) is a life-saving treatment used for patient suffering from acute refractory respiratory or cardio-respiratory failure, offering both circulatory and respiratory support. During the COVID-19 pandemic, ECMO has been widely used for cardio-pulmonary support. In Europe only, over 2687 patients (data from January 11) have at this point received ECMO treatment. Although associated with a relatively high survival rate, ECMO is not complication free. Cerebral bleedings and thromboembolism are the most common severe complications leading to death in 80–85% of the cases (Dalton et al., 2015; Fletcher-Sandersjöö et al., 2017). These complications are associated with blood component damage/activation with subsequent increased risks of hemolysis, thrombosis, and emboli (Chen et al., 2019b). Thrombus has been observed in tubing connectors, cannulae, in the artificial lung, as well as in the blood pumps (Fraser et al., 2012; Fujiwara et al., 2017; Hastings et al., 2017). In the complete circuit, venous blood is drained via a drainage cannula from a major vein and then pumped through a membrane lung (artificial lung, oxygenator) where carbon dioxide is cleared from the blood and the blood is re-oxygenated before it is reinfused back into the patient via a return cannula. If the return cannula is placed on the venous side of the patient's circulation veno-venous ECMO (lung support) is offered. If the return is into the arterial side of the circulation veno-arterial ECMO for cardio-pulmonary support is provided. The cannula, pump, and membrane lung are connected with 3/8 inch tubing in the ECMO circuit for the adult. In small children and the newly born the tubing size is of 3/16 or 1/4 inch inner diameter. The recirculation fraction (R_f) (proportion of extracorporeal flow not contributing to the patient's oxygenation) reflects the intrinsic problem associated with veno-venous (VV) ECMO where an immediate drainage of the extracorporeally oxygenated blood reduces the effective flow. Recirculation fraction has no direct influence on complications such as thrombosis. Instead it is indirect, resulting from when the provider increases ECMO blood flow to compensate for R_f, leading to enhanced mixing and higher shear stresses. Recirculation is affected by factors such as the cannula position, orientation, and design (van Heijst et al., 2001; Körver et al., 2012; Palmér et al., 2016). Adjusting the position of the cannulae can lower the recirculation factor from 46% returned oxygenated ECMO blood being directly drained back into the ECMO circuit to as little as 10% (Broman et al., 2015; Palmér et al., 2016). Hence, improving the understanding of mixing properties between oxygenated blood and native blood in the area of return prior to entering the volume from which blood is drained out into the ECMO circuit is of great importance. Moreover, mixing characteristics depends on the flow regime, and directly connected to the flow rates and geometrical dimensions of cannula and vessel.

Regarding thrombus composition, it varies depending on the formation process, where artificial surfaces as well as flow characteristics as regions of high- and low- shear stresses have been identified as locus for platelet activation (Casa and Ku, 2017). Also, the response of platelets to stress is likely to depend on the type of stress (Qiu et al., 2014), i.e., normal and shear stresses. Once activated, platelets may aggregate on artificial surfaces or in low-shear regions binding to fibrin more easily. Also, non-physiological high shear activation induces loss of platelet surface receptors. This may lead to weaker binding to collagen and von Willebrand factor (vWF) at sites of vascular injury where exposure time to high shear stress is a factor. The conditions faced by the platelets leading to activation have been investigated in many studies (see Mains-Balle et al., 2018; van der Meijden and Heemskrek, 2019 for recent reviews). Moreover, mathematical models based on empirical correlations between thrombus growth and shear rate have been developed (Bark et al., 2012; Mehrabadi et al., 2016; Ku et al., 2017). However, the intensity of the shear stress is not the only determinant factor for platelet activation. According to Hellums (1994) and Jesty et al. (2003), platelets can be activated, or even damaged (Giersiepen et al., 1990), through a short exposure to a high shear stress as well as through a long exposure to a moderate shear stress. Yin et al. (2011) also highlighted that a pulsatile low shear stress potentially could lead to enhanced risk for thrombosis. To take account of the shear stress amplitude and the time of exposure to this shear stress, Nobili et al. (2008) proposed Platelet Activation State (PAS) as a parameter to quantify the platelet activation probability. Using this parameter, Fuchs et al. (2018) studied the different components of the ECMO circuit, reporting that the centrifugal pump head presents the highest risk of activation followed by the return cannula. Furthermore, elevated shear stress may also cause damage to the red blood cells (RBCs) inducing hemolysis (Blackshear, 1972). Hemolysis thresholds have been determined by assessing the shear stress levels in a jet (Sallam and Hwang, 1984; Lu et al., 2001), using viscometers (Wurzinger et al., 1985), and in medical pumps (Fraser et al., 2012). As for platelet activation, the exposure time is important in hemolysis as it reduces the shear stress threshold needed to damage the RBCs (Leverett et al., 2018). To quantify the shear stress acting on blood constituents, both viscous shear stress (VSS) and Reynolds shear stress (RSS) have been used (Jones, 1995; Yen et al., 2014).

Although, confined jets have been thoroughly studied in literature (Rajaratnam, 1976; Rehab et al., 1997; Kandakure et al., 2008) and still is a highly active field of research for various applications (Wang et al., 2020; Pathikonda et al., 2021), studies on highly confined jets such as a cannula inserted in a vessel are few. Moreover, commonly the co-flow in previous studies have been larger than the jet flow (for example flame stabilization), and not focusing on a situation where the co-flow is slower than the jet flow as may occur for cannula flows. In terms of numerical simulations of cannula blood flows, early work by Bedingham and Neavin (1991) assessed probable sites for thrombus activation and deposition and De Wachter et al. (1995, 1996) studied shear stress level in dialysis needle. More recently, Grigioni et al. (2002) and Menon et al. (2013) numerically investigated the effect of the geometry of the cannula on blood flow development. However, experimental data of the detailed flow characteristics remain scarce and are challenging due to the both the liquid (blood) and the components to be studied.

In this experimental study, the focus is on return cannula flows relevant for ECMO cannulation in VV ECMO. In VV ECMO, the blood is drained and returned in the inferior and superior vena cava (IVC and SVC, respectively), or vice versa, with the right atrium being the connecting volume in which the flow to and from the cannulae are communicating. From a fluid mechanical point of view, this is a highly complex flow case including multiscale physics as well as transient phenomena. Thus, in order to decompose the problem and to separate the effect due to the different variables, we have in this comprehensive study, adopted the approach to systematically study the effect of return cannula position and flow rates in both cannula and vessel, considering a simplified geometry. In our previous study, Lemétayer et al. (2020), the flow structures and velocity field characteristics of a smaller cannula was studied. In this study, the aim was to study the impact of the flow on in particular mixing characteristics and shear stresses. Considering also a larger cannula diameter enabled larger flow rates to be assessed, including a wider range of cannula to vessel flow rate ratios and the effect of change in confinement.

Although ECMO is the clinical application forming the basis of this study, cannula flows in which blood is drained from or returned to the circulatory system is relevant in several medical situations where the geometry of the cannula, its position in the vessel and the flow rates are parameters affecting the flow characteristics and potentially increase in risk of thrombus formation and/or hemolysis.

To ensure the symmetry of the flow, measurements of the flow velocity and its fluctuations were performed for two symmetric positions of the cannula relative to the outer tube centreline. The results, presented in Lemétayer et al. (2020), showed excellent agreement. Thus, detailed experiments were only carried out tilting the cannula toward one side relative the outer tube centreline.

Capturing the flow dynamics in a cannula inserted in a vessel is a highly complex problem as there are several variables needed to be taken into account and that varies depending on the clinical situation of interest such as ECMO flows, blood viscosity, position, cannula diameter, and vessel to cannula diameter ratio.

In this work, the research approach was to minimize complexity, in terms of both return cannula design and vessel geometry of the IVC, to enable identifying the underlying fluid dynamical structures influencing cannula performance, in particular in VV ECMO. Shifting the cannula position increases the flow complexity and allows, in a controlled environment, to approach the real situation using flexible cannulae, for which preliminary experiments shows that the cannula is unlikely to be centered in the vessel. Thus, understanding the flow structures involved in and governing mixing in cannula flows and how they are affected by position and operating conditions are important in order to at a later stage provide guidance to the medical community. In terms of geometrical simplification, it especially concerns the outer tube. The diameter used for the outer tube is among the highest values referred for an IVC diameter, meaning that the confinement in a medical situation would likely be more important. However, based on previous studies carried out on confined jets, especially Henzel's work (Henzler, 1978), the flow structures observed in this study would still appear as long as the condition D/d > G + Q is valid, where G ≈ 1 at equal densities of the mixed fluid. Thus, considering that Q should be smaller than 1, the flow structures are expected to remain similar for an outer tube diameter not smaller than twice the cannula inner diameter.

Two different cannula diameters were investigated allowing flow rates from 1.3 to 5.2 L/min to be studied, and as mentioned before, chosen according to the flow rates used in VV ECMO, ranging between 3.5 and 5 L/min although flows of 2–7 L/min are used in clinic and feasible with standard ECMO circuits. Using water in the flow circuit resulted in jet Reynolds numbers of 4,510–20,060 (the incoming co-flow prior cannula tip remained laminar). In all except for the lowest Reynolds numbers, the flow was found to be fully turbulent. However, in the clinical situation, the larger viscosity of blood implies a lower Reynolds number. If estimating a Newtonian blood analog fluid to have 3.4 times the viscosity of water (hematocrit of about 40%), will result in an equivalent blood Reynolds number of about 1,350–4,500 for the considered flow rates. However, in this context, it should be noted that in ECMO, the hematocrit level used by the different ECMO centers varies between 20 and 40%. A hematocrit level of 25% corresponds (at high shear rates) to a viscosity approximately twice that of water. Consequently, the Reynolds numbers obtained scaling with a blood analog Newtonian viscosity suggest that the flow is more likely to laminar to transitional rather than fully turbulent. Estimating the friction factor for a pipe flow using pressure drop measurements of cannula flows using blood (Broman et al., 2019) indicates that the flow is possible to be in transition. Whether the flow is laminar, in transition or turbulent is of clinical importance as the flow regime will alter not only the forces but also the mixing characteristics as shown in Figure 11. As mixing is enhanced by turbulence, the reported mixing lengths required will be larger for a laminar flow. Moreover, if mimicking the flow in the IVC, the flow rate studied is lower than what would be expected. Taking the blood viscosity in account, scaling the flow rates, the higher flow rate ratios (Q = 1 and 0.5) are more relevant from a clinical perspective. It is interesting to note that for the same Q, the small and large cannula perform similarly in terms of mixing length (normalized by diameter) required although initial mixing differentiated. Moreover, comparing Cases 1_SC and 3_LC having similar velocities in the cannula jet and co-flow, the large cannula performed better in terms of lowering mixing length required, associated with the larger Reynolds number obtained for the large cannula.

The flow is characterized by strong shear layers and the results show that the shear stresses found in the flow field are of an order comparable to the threshold reported in literature inducing platelet activation or damaging RBCs. This conclusion is not restricted by the use of water. Although using water, we are in our measurements capturing the larger flow scales, also expected to appear in blood. Moreover, even if a reduction of the shear stress magnitude may be expected with blood due to higher viscosity lowering the smallest turbulent scales, it is not only the level of shear stress that the blood constituents are exposed to that is of importance, but also the exposure time (Hellums, 1994; Fraser et al., 2012) that may increase in case of a more viscous fluid. It is commonly accepted that strong shear and long residence times (stagnant regions) are the main contributors for thrombus formation (Casa and Ku, 2017; Chen et al., 2019a). According to Hellums'shear stress vs. exposure time threshold map, the time associated with the shear stresses in the current study (approximately 10⁻¹ s) indicates that shear stress levels of the order of 100 Pa are sufficient to activate platelets. In this study, we chose to use two different formulations expressing the shear stress depending on their origin (TVSS and RSS) according to literature. Thus, the formulation used to define the shear stress has an impact on its value since the two formulations result in different orders of magnitude for the shear stress. Furthermore, thresholds measured from a constant shear stress device like viscometer are hardly comparable to the turbulent flow present in the return cannula due to the flow fluctuations making the shear stress varying in space and time. However, studies of real geometries can be used to infer a practical threshold. Liu et al. (1999) found a RSS of 52 Pa studying the flow in an aortic valve and pointed out that this magnitude of RSS is enough to damage RBCs and activate the platelets even if lower than the threshold usually defined. Thus, according to our investigations, the flow created by the reinjection of blood is highly susceptible to present shear stresses activating the platelets or damaging RBCs. However, it should be mentioned that in this study, the shear stress estimation is based on only two components due to the experimental set-up. Thus, the contributions of the third component and its fluctuations are not taken into account, that may lead to an underestimation of the actual shear stresses in the flow.

Shifting the cannula, the results show that the shear stress decreases as compared to the configuration with a centered cannula. On the other hand, the greater fluctuations of the shear stress when the cannula is shifted indicates that higher value of shear stress may be reached locally in the flow. Furthermore, the size of the recirculation zone on the wide side is larger, indicating that there is an elevated risk for platelets to be exposed to an increased residence time in that region. Thus, the presence of structures that can act in order to trap platelets, such as backflow or vortices, has a negative effect on platelet activation by increasing the potential of residing in zones of high shear stress as well during prolonged time. This resembles the observation made by Fuchs et al. (2018) focusing on platelet activation in ECMO circuit components. An increase in cannula flow rate will enhance the shear stress and the backflow characteristics. These results could be considered as an inducement for further developments to improve the control over the cannula position.

From the clinical perspective, the results from this work that directly apply to the clinical situation are (1) increased understanding how distances between the return and drainage cannula relate to R_f in terms of mixing of the two blood streams, and (2) how cannula size and flow rate, or rather the difference in flow speed of a narrow vs. a wider cannula influences the shear layer (between cannula and native flow) and associated shear forces and thus platelet activation and trauma to the RBCs. With this new knowledge, a clinician can get assistance in choosing the cannula size and better positioning the return cannula in terms of recirculation and blood trauma (platelet activation) with the best interest of the patient in focus.

In this experimental study of the flow dynamics in a return cannula, the main findings are as follows:

From the clinical point of view, aiming at preventing recirculation of oxygenated ECMO blood, the results show that a minimum of 10–12 cannula diameters are needed to obtain a homogeneous mixture of oxygenated blood and native blood. As shown, this depends on the flow regime and thus quantifying cannula flows using blood or blood analogs are important, in particular for the larger flow rates and cannula size. Moreover, a fully developed flow was obtained at a much larger distance (22–30 cannula diameters), indicating that higher level of oxygenated blood may be transported over longer distances by the jet. Thus, if placing the drainage cannula too close to this region, this may increase R_f, i.e., directly draining oxygenated ECMO blood back into the ECMO circuit.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

JL performed the experiments, analyzed the results, and drafted the manuscript. LB analyzed the results and critically revised the manuscript for intellectual content. LPW designed the concept, analyzed the results, and critically revised the manuscript for intellectual content. All authors approved for submission.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.