Development of a Novel Intravascular Oxygenator Catheter: Oxygen Mass Transfer Properties Across Non-Porous Hollow Fiber Membranes

Abstract

Despite hypoxic respiratory failure representing a large portion of total hospitalizations and healthcare spending worldwide, therapeutic options beyond mechanical ventilation are limited. We demonstrate the technical feasibility of providing oxygen to a bulk medium, such as blood, via diffusion across non-porous hollow fiber membranes (HFM) using hyperbaric oxygen. The oxygen transfer across Teflon® membranes was characterized at oxygen pressures up to 2 bars in both a stirred tank vessel (CSTR) and a tubular device mimicking intravenous application. Fluxes over 550 mL min⁻¹ m⁻² were observed in well mixed systems, and just over 350 mL min⁻¹ m⁻² in flow through tubular systems. Oxygen flux was proportional to the oxygen partial pressure inside the HFM over the tested range and increased with mixing of the bulk liquid. Some bubbles were observed at the higher pressures (1.9 bar) and when bulk liquid dissolved oxygen concentrations were high. High frequency ultrasound was applied to detect and count individual bubbles, but no increase from background levels was detected during lower pressure operation. A conceptual model of the oxygen transport was developed and validated. Model parametric sensitivity studies demonstrated that diffusion through the thin fiber walls was a significant resistance to mass transfer. Promoting convection around the fibers should enable physiologically relevant oxygen supply. This work indicates that a device is within reach that is capable of delivering greater than 10% of a patient’s basal oxygen needs in a configuration that readily fits intravascularly.

Graphical Abstract

1. Introduction

Hypoxic respiratory failure is a common reason for hospitalization and intensive care unit (ICU) admission for children and adults. A subset of hypoxic respiratory failure resulting from acute lung injury, acute respiratory distress syndrome (ARDS), is present in 7% of patients admitted to ICUs worldwide with a mortality ranging from 34–44% (Phua et al., 2009; Schouten et al., 2016; Sweeney & McAuley, 2016). Despite the prevalence and morbidity associated with hypoxic respiratory failure, treatment options remain relatively few beyond supplemental oxygen and/or traditional mechanical ventilation, which can carry risks. When mechanical ventilation is insufficient, clinicians turn to veno-venous extracorporeal membrane oxygenation (VV-ECMO). ECMO is the only treatment option that delivers oxygen directly to the bloodstream independent of lung function, but is associated with adverse events including intracranial hemorrhage, ischemic stroke, acute kidney injury, and infection (ECLS Registry Report, International Summary, 2020). ECMO also carries a large economic burden and is only available in specialized centers in developed nations, which greatly limits its use (Harvey, Gaies, & Prosser, 2015; Wallace et al., 2014). The need for alternative methods of support for severe hypoxic respiratory failure that function independent of the lungs is becoming evident. This has become particularly apparent at the time of writing this manuscript as the COVID-19 pandemic progresses worldwide causing hypoxic respiratory failure in those most severely affected. COVID-19 is straining healthcare systems with their limited ability to provide mechanical ventilation and even scarcer capability to offer ECMO.

In this light, novel therapies that can safely deliver oxygen directly into the blood are an attractive option as they may reduce the need for mechanical ventilation and/or ECMO. Such devices ideally would be no more invasive than a large central venous catheter and, more importantly, be equally available to physicians and medics worldwide. Previous groups have attempted direct intravascular oxygenation, though no device has succeeded due to their reliance on a large surface area for oxygen delivery that resulted in bulky catheters that impaired cardiac return (Anthony J. Makarewicz, Mockros, & Anderson, 1994; Nodelman, Baskaran, & Ultman, 1998). The first and only such intravascular device to reach human clinical trials was the IVOX catheter, which was composed of over 1000 microporous polypropylene hollow fibers coated in siloxane furled together to fit in the vena cava with sub-atmospheric oxygen flowing within. While this device delivered a measurable oxygen amount, its large size (12.6 to 16 mm diameter in furled insertion configuration, containing 0.21 to 0.51 m² of diffusing surface area) was cumbersome and led to a 24.5% clinically recognized complication rate, including bleeding during insertion, thrombophlebitis, and hemodynamic instability (Cockroft et al., 1992; Conrad, Bagley, Bagley, & Schaap, 1994). IVOX did not gain FDA approval and device development was halted (Federspiel & Henchir, 2004). Hattler et al. improved upon this concept with the addition of a balloon pulsating mixing mechanism that allowed a 0.17 m² device to deliver 23.8 mL min⁻¹ of oxygen in water. However, this device still had a 10.6 mm outer diameter in the insertion configuration, and they predicted needing a total of 0.4 m² of HFM diffusing surface area to accomplish their CO₂ clearance and oxygenation goals (Hattler et al., 2002). These previous attempts at intravascular gas exchange were limited in their oxygen delivery due to their reliance on a large surface area for gas exchange (especially as they were trying to exchange both oxygen and carbon dioxide) which made for large impractical catheters.

In contrast to the previous attempts at intravascular gas exchange, our team is developing an intravascular oxygenator catheter that relies on hyperbaric oxygen to generate a large driving gradient across a non-porous diffusing surface, greatly increasing O₂ transfer, rather than relying on a large surface area. Our device is intended only to deliver oxygen, the primary deficit in most forms of acute lung injury. This approach requires less total HFM surface area and allows for a more compact device amenable to intravascular use, overcoming the challenges faced by previous groups. It is the first intravascular device to use such a method to deliver oxygen. This approach to optimizing oxygen transport from non-porous HFMs into a bulk medium may have many other non-medical applications, such as in biotechnology and wastewater treatment (e.g., bubbleless aeration and membrane aerated bioreactors (MABRs)).

Previous work by Hattler et al. showed that providing approximately 20% of an adult’s basal oxygen needs (about 50 mL O₂ min⁻¹⁾ would achieve a life-sustaining arterial oxygen saturation of 90% (P_aO₂ of 60 mmHg) in patients with only 50% of their native lung function intact (Hattler et al., 2002). Our goal, therefore, is for our device to deliver 25 −100 mL O₂ min⁻¹ (a minimum of 10% of an adult’s basal oxygen needs). At this oxygen delivery rate, our device would allow for decreased reliance on mechanical ventilation to support the patient. Conrad et al, demonstrated that delivering 40–70 mL O₂ min⁻¹ with the intravascular respiratory assist catheter, IVOX, allowed for reduction of mechanical ventilation by 25% or more of pre-IVOX insertion levels of intensity in about 50% of their patients (Conrad et al., 1994).

In this work, we report on foundational mass transfer experiments used in the development of our intravascular oxygenation device. Experiments were conducted in-vitro with single and multiple hollow fiber configurations in both a stirred tank vessel (CSTR) and a tubular device mimicking intravenous application to quantify and characterize the rate of oxygen delivery as a function of the operating conditions. The purpose of this study was to develop an analytical model of the oxygen mass transfer via HFMs and validate it through in-vitro experimental testing. Our work indicates that a device is within reach that is capable of delivering greater than 10% of a patient’s basal oxygen needs in a configuration that readily fits intravascularly.

2. Materials and Methods

2.1. Experimental systems

A schematic of the experimental setup is shown Figure 1. Continuously stirred tank reactor (CSTR) batch systems were constructed using gas tight glass vessels fitted with a single 20 cm fiber. The fibers in this study were made of polytetrafluoroethylene (PTFE); specifically, Teflon AF2400 polymer, chosen for its high oxygen gas permeability (even higher than PDMS, another polymer commonly used in medical applications, and recently deployed in artificial lung prototypes) of 9.9 ×10⁻⁸ cm² s⁻¹ cmHg⁻¹, and sufficiently high burst strength (theoretically > 130 bar) (The Chemours Company, 2016). The membrane used had an outer diameter of 406.4 µm and a wall thickness of 88.9 µm (Biogeneral, Inc., San Diego, CA). The oxygen pressure at the inlet of the fiber(s) was controlled using a mass flow controller (MFC) (Alicat Scientific, Tucson, AZ), while a manual backpressure regulator (Airtrol RV-5300–90, New Berlin, WI) regulated the outlet pressure. The fiber(s) effluent was connected to a gas flow meter (Dwyer Instruments, Michigan City, IN) to ensure a minimum outlet oxygen flow (generally about 20 mL min⁻¹), thus avoiding any effects of back diffusion of water vapor and water condensation inside the fibers. The circuit was composed of Tygon tubing with a Rotaflow centrifugal console pump (Getinge AB, Goteborg, Sweden). The CSTR was mixed using a magnetic stirrer at 400 (low speed) and 1200 rpm (high speed). A medical grade gas bubble trap (Terumo Cardiovascular Group, Ann Arbor, MI) was installed after the oxygenator to collect any bubbles that formed during operation prior to the water being recirculated back through the pump. Dissolved oxygen was recorded with an optical DO probe (Vernier, Beaverton, OR) housed in a custom chamber and connected to a data logger. A flush outlet and inline pressure relief reservoir were included in the closed loop to expose the system to atmospheric pressure.

Figure 1.

Process schematic of bench top system and instruments. A) System layout showing interchangeable chamber location, B) CSTR chamber detail, C) tubular system detail.

A second series of experiments was conducted in a custom-made tubular chamber with water flowing parallel to the long axis of the fiber(s) to create a counter-current oxygen transfer system (Figure 1) mimicking intravascular flow. The acrylic tube was 30 cm in length with a 1.0 cm internal diameter. Different tubular chambers were fitted with 1, 4, and 16 fibers. The closed loop system described above was used for tubular systems by replacing the batch reactor with the appropriate tubular assembly.

2.2. Experimental procedure and analytical equipment

All experiments were conducted at room temperature (23–25° C) and atmospheric pressure. The closed loop circuit was flushed with at least 1.5 L of nitrogen sparged water such that the starting dissolved oxygen concentration was 1.5 mg L⁻¹ or less. At this point, oxygen (100% O₂, Airgas, Radnor, PA) was introduced into the fiber(s) at the prescribed pressure (0.24–1.9 bar) and the increase in dissolved oxygen (DO) was monitored using the DO probe connected to a data logger.

2.3. Data analysis

For all experiments, the oxygen transfer flux was calculated as the total mass of the oxygen transferred to the liquid between two data points, while the dissolved oxygen concentration at which this flux occurred was taken as the average between the two points. A moving average was used to smooth each dissolved oxygen time series; 5 sample points (50 to 75 seconds span) were used for experiments conducted with the tubular systems and 9 sample points (90 to 135 seconds span) for those with the CSTRs. Each pressure and flow combination tested was conducted in triplicate for tubular systems and duplicate for CSTRs. A linear regression with confidence intervals of 95% for the true mean was used to calculate the average flux at a given bulk water gas partial pressure for oxygen over the range of 30 to 100 mmHg.

2.4. Ultrasound bubble detection

As a complement to measurements of dissolved oxygen, bubble formation was studied downstream of the prototype oxygenators using high frequency ultrasound imaging. A Vantage 256 research scanner (Verasonics Inc., Kirkland, WA) was used with the L22–14 linear array (Vermon S.A., Tours, France) to perform high frame rate imaging through a CDI blood parameter monitoring acrylic 3/8 inch cuvette (Terumo Cardiovascular Group, Ann Arbor, MI) to directly detect the presence of gas bubbles. Plane wave transmissions (Montaldo, Tanter, Bercoff, Benech, & Fink, 2009) with a center frequency of 18 MHz were performed in a tubular system at 0.24 bar average fiber pressure and 2 L min⁻¹ flow rate. First, the imaging plane was aligned along the direction of flow, and high-resolution imaging was performed (five plane waves steered at −8 to 8 degrees in 4-degree increments) at 4 kHz frame rate for 100 milliseconds. Second, the imaging plane was placed in cross-section of the flow and high-speed imaging was performed (a single plane wave) at 20 kHz for 100 milliseconds. In both cases, channel data sets were acquired for offline processing at approximately 5-minute intervals for the duration of the experiments.

Gas bubbles were detected in the image frames using automated bubble counting after suppression of background noise using spatial coherence beamforming. We performed mid-lag spatial coherence processing (Hsi et al., 2018) to remove specular reflection and to isolate air bubbles from background and noise. Bubbles were identified in individual frames using thresholding and blob analysis. Identified bubbles from each frame were paired through time to fit a motion model and provide an estimate of unique bubbles observed during each observation period.

2.5. Model development

A conceptually accurate dynamic model describing oxygen mass transfer in our experimental systems was developed. The model assumes that Fick’s first law applies for oxygen flux across the membrane wall via diffusion, while convective mass transport occurs from the outer surface of the membrane into the bulk liquid (Figure 2). Partial differential equations were written describing the transport of oxygen and were then solved numerically.

Figure 2.

Conceptual model of the membrane interfaces with high pressure oxygen inside the fiber, diffusion through the non-porous membrane, liquid boundary layer, and bulk liquid. See SI for detailed model equations.

For device modeling purposes, the liquid in the stirred batch systems was assumed to be ideally mixed (which was verified experimentally), whereas spatial discretization for the liquid was applied along the axial direction for the tubular systems to account for the axial gradient of DO and its effect on the transfer flux. Tubular system models were discretized in the direction of flow only; no gradient was assumed in the liquid along the radial direction (except for the boundary layer). In both CSTR and tubular models, the fiber was discretized along the axial direction with the partial pressure inside the fiber (P_fiber) assumed to drop in a linear fashion from the inlet, as set by the MFC, to the outlet, as set by a back-pressure regulator. The partial pressure of oxygen (P_wall) at the membrane-liquid interface was calculated using the Henry’s constant for oxygen and solving a mass balance at the interface (Equations 1 – 4 in SI). Because the fibers are continuously purged using low humidity 100% oxygen, the impacts of any back diffusion (e.g., of CO₂ or water vapor leading to condensation) into the HFMs is estimated to be negligible. Additionally, previous authors have shown that the effects of back diffusion on closed end fibers used for bubbleless aeration were negated with relatively short venting periods (20 s every 30 min) (Perez-Calleja et al., 2017). Detailed model equations can be found in SI.

The model also incorporated intrusion of small amounts of oxygen from the atmosphere into the system by diffusion occurring through the Tygon circuit tubing (determined separately in independent experiments) which was relevant, especially for experiments at low fluxes, i.e., those with low membrane surface area and low pressures. This oxygen intrusion into the system was removed for all sensitivity analyses.

The partial differential equations were solved numerically using a Runge-Kutta 4^th order algorithm. Model parameters (Table 1) included both physical constants taken from the literature and system specific parameters. The latter were determined experimentally. In particular, the model was fit to the DO time series (see Results section) using the mass transfer coefficient of the liquid boundary layer (k_boundary) as the fitting parameter. Sensitivity analyses were conducted with each model on the CSTR and tubular systems while maintaining a constant pressure along the fiber length (i.e., no pressure drop inside fiber).

Table 1.

Model parameters used in this study.

Parameter	Value	Unit	Source
Fiber length	CSTR: 20	cm	Measured
	Tubular: 30*	cm	Measured
Fiber inner diameter	228.6	µm	Manufacturer
Fiber wall thickness	88.9	µm	Manufacturer
Fiber O2 permeability	9.9×10−8	cm2 s−1 cmHg−1	Manufacturer(The Chemours Company, 2016)
Convective mass transfer coefficient	CSTR#: 9.54 ×10−3, 1.45 ×10−2	cm s−1	Fitted to experiment
	Tubular¥: 3.65×10−3, 6.42 ×10−3	cm s−1	Fitted to experiment
Number of discrete fiber elements	CSTR: 11	-	Assumed
	Tubular: 30	-	Assumed
Circuit flow rate	CSTR: 0.5	L min−1	Measured
	Tubular: 0.5 – 2.0	L min−1	Measured
Oxygen pressure range	0.24 – 1.9	bar	Measured, independent parameter

^*Tubular devices length varies slightly (31–33 cm) between the different units.

^#400 and 1200 rpm,

^¥0.5 and 2.0 L min⁻¹, respectively.

3. Results and Discussion

3.1. Oxygen transfer: time series and fluxes

Selected time series for oxygen transfer are reported in Figure 3. They show a typical saturation behavior (Figure 3 inset) caused by the decrease of the DO gradient over time as DO in the liquid increases. Comparison of the different curves in Figure 3 reveals that oxygen transfer is limited by both the gradient across the fiber (DO increase is faster at 1.9 bar vs. 0.24 bar) as well mixing in the system (DO increase is faster at 1200 rpm vs. 400 rpm mixing). Note that the tubular system shown in Figure 3 cannot be directly compared to the CSTR since the devices have different fiber lengths (20 cm for the CSTR vs. 32.4 cm for the tubular device shown). To conduct a direct comparison, calculation of the oxygen transfer flux is needed. These results are shown in Figure 4 where the instantaneous oxygen transfer flux is reported as a function of the partial pressure of oxygen in the liquid during selected experiments.

Figure 3.

Time series of dissolved oxygen for single fiber tubular device (30 cm fiber) and CSTRs (20 cm fiber) at two different average oxygen pressures. The inset shows an extended time scale run for a select tubular single fiber illustrating decreased transfer as oxygen saturation increases.

Figure 4.

Subset of instantaneous oxygen flux vs. liquid partial pressure of oxygen for the selected single fiber CSTR and tubular experiments shown in Figure 3.

Overall, high oxygen fluxes were observed compared to earlier reports of related devices, even in the tubular device at lower water flows (Eash, Mihelc, Frankowski, Hattler, & Federspiel, 2007; Hattler et al., 2002). The general trend downwards in Figure 4 was expected, caused by the decrease in the driving force for oxygen transfer as the liquid DO increases. The tubular device was less performant than the CSTR, since the parallel flow fiber orientation with no additional mixing mechanism resulted in greater resistance to oxygen transfer. However, the results show that increasing the system pressure within the fiber could greatly improve oxygen transfer, consistent with our hypothesis that hyperbaric conditions are a possible means to achieve physiologically relevant oxygenation rates in a compact device.

Summaries of all fluxes at physiologically relevant liquid partial pressures of oxygen are shown in Figures 5 and 6 for the CSTR and tubular systems, respectively. CSTR results in Figure 5 show the potential for very high fluxes using the non-porous fibers when external mass transfer is not a limiting factor (1200 rpm mixing). In the low mixing CSTR (400 rpm), results show small decrease in flux at times as the pressure increases. This is likely caused by fiber touching the glass vessel wall during low mixing as the pressure within the fiber increased. Another possibility is the supersaturation of water with oxygen at the outer surface of the HFM during low mixing leading to formation of small oxygen bubbles on the fibers and a decrease in the overall transfer. The effect of bubbles on oxygen transfer remains to be defined and is further discussed below. Regardless, the fluxes in the CSTR remained higher than in the tubular systems. For the tubular systems, examination of Figure 6 reveals that increasing the fiber density from one fiber to four, and from four fibers to sixteen resulted in a slight decrease in the flux. This is likely due to a combination of fiber to fiber touching (thus reducing area for transfer), as well as local oxygen gradients being lower within the bundle of fibers (gradients overlapping). Well mixed CSTRs (Figure 5) had higher fluxes than all tubular devices at the same system pressures due to higher mixing rates increasing the convective mass transfer coefficient.

Figure 5.

Experimentally determined fluxes at select liquid partial pressures for CSTR systems. Dashed lines show highest flux efficiency achieved by previous authors with extreme mixing (10,000 rpm) and more modest pulsing methods (300 BPM).

Figure 6.

Experimentally determined flux at select liquid partial pressures for tubular devices (a and b are replicate experiments). Low flow high pressure (0.5 L min⁻¹ at 1.9 bar) was not included due to bubble formation in system. Dashed lines show highest flux efficiency achieved by previous authors with extreme mixing (10k rpm) and more modest pulsing methods (300 BPM).

As mentioned, the fluxes observed in both the CSTR and tubular systems are significantly higher than those reported by previous groups (Baskaran, Nodelman, & Ultman, 1998; High et al., 1994; A J Makarewicz, Mockros, & Anderson, 1993; Anthony J. Makarewicz et al., 1994; Nodelman et al., 1998; Snider et al., 1994; Vaslef, Mockros, & Anderson, 1989) who developed devices for intravascular gas exchange, including Hattler et al. whose team designed and tested some of the most efficient devices to date. The Hattler catheter systems relied on oxygen pulled through porous polypropylene membranes by a vacuum, remaining at sub-atmospheric pressures, with a variety of mixing mechanisms included to improve mass transfer. Figure 5 and 6 include two benchmarks, 140 (Hattler et al., 2002) and 251 (Eash et al., 2007) mL min⁻¹ m⁻² flux, representing a device with a balloon rapidly inflating/deflating at 300 beats per minute (BPM) and their other device with fibers rotating up to 10,000 rpm respectively. These comparisons illustrate that non-porous hollow fiber membranes (which allow for hyperbaric oxygen flowing within) can achieve greater fluxes than previously obtained with porous fibers that require low (and even sub-atmospheric) pressure oxygen to be used to prevent convective flux of gaseous oxygen and bubble formation.

3.2. Bubble investigations

It is well known that large vascular air emboli (bubbles) cause significant morbidity, such as cardiovascular collapse if venous gas migrates to the pulmonary circulation, or tissue ischemia and necrosis if gas is lodged in the arterial vessels. However, the impacts of air micro-emboli, particularly pure oxygen micro-emboli, are not well understood. To our knowledge, no studies have examined the effect of pure oxygen gas micro-emboli in vivo, with only some case studies of large oxygen gas emboli with varying degrees of morbidity (Barak & Katz, 2005; Haller, Faltin-Traub, Faltin, & Kern, 2002). At the conditions reported herein, bubbles were only observed visually on some fibers when one of the following occurred: 1) two or more fibers were touching in the multifiber cartridges, 2) the high pressure (1.9 bar) was used, 3) when low water flow (0.5 L min⁻¹) was used in the tubular system (which inherently has less mixing), 4) when the bulk liquid DO was higher than >2 mg L^-1.

These observations serve as the foundation as we develop prototype devices and define safe operating parameters for our system. Our experiments show that future prototypes will require HFM spacing such that fiber-fiber interactions are accounted for, and limited as is feasible, which will reduce bubble risks and allow for higher operational pressures. Periodic fiber movement, bulk mixing, or even fiber physical features or coatings in key areas are potential ways to limit the areas of highest fiber-fiber interaction (such as the beginning and end of the fiber bundles). Additionally, methods to enhance mixing in and around the fibers will be required not only to further increase oxygen flux, but also to enhance convective transfer of oxygen to the bulk liquid to reduce bubble formation. Finally, the bubbles seen with experiments performed at 1.9 bar show that enhanced mixing of the bulk and/or reduction of nucleation sites during higher oxygen fluxes may be an important improvement of future prototypes to ensure safe operation. However, it is likely that pure oxygen in small bubbles would be quickly bound by hemoglobin or metabolically consumed.

Beyond direct (visual) observation of bubbles, a proof-of-concept of high frequency ultrasound imaging for bubble detection was conducted during testing of the tubular system. Results are reported for one condition (tubular device, 2.0 L min⁻¹ and 0.24 bar) in Figure 6. They revealed the presence of a low concentration of small bubbles in the experimental circuit at baseline even before the introduction of oxygen (Figure 7). Further, bubble counts remained below the maximum baseline in all but one data point during the hyperbaric oxygenation (Figure 7B). Videos of bubble tracking and their motion are shown in SI. They show our ability to track and quantify bubbles in real time. A ~3 mL bubble was released into the circuit (not from the HFM) and broken down into smaller bubbles by the circulation pump resulting in a clear increase in the bubbles detected by ultrasound prior to being caught in the bubble trap. This served as a positive control for bubble detection. This ultrasound work confirms the delivery of oxygen from the HFM is solely possible via dissolved oxygen rather than through nucleated bubbles under select hyperbaric conditions. Secondly, this demonstrates that our bubble quantification method provides a quantitative means to monitor bubbles in future experiments especially in in-vitro blood testing where direct bubble visualization in the system will be difficult given the opaque nature of blood.

Figure 7.

Experimental observation of gas bubbles in the tubular device using ultrasounds. A) DO evolution vs. time during the test runs (2.0 L min⁻¹ 0.24 bar). Markers indicate times for ultrasound bubble assessment. B) Unique bubbles counted within 100 ms sampling window. Baseline prior to oxygenation shown at time <0. Positive control of bubble detection shown at end of each run (triangle symbols). C) Sample image after non-linear coherence processing with three identified bubbles.

3.3. Modeling oxygen transfer in the devices and model parametric sensitivity

The conceptual model of the device was used to determine the convective mass transfer coefficients for the CSTR and tubular systems. A set of replicate runs was used in each case (CSTR 400 rpm, CSTR 1200 rpm, tubular device 0.5 and 2 L min⁻¹), and the model was fitted to the experimental DO data with the mass transfer coefficient k as the fitted parameter and averaged across the replicates. The mass transfer coefficients values obtained were 9.54 × 10⁻³ and 1.45 × 10⁻² for the CSTR at 400 rpm and 1200 rpm, respectively and 3.65 × 10⁻³ and 6.42 × 10⁻³cm s⁻¹ for the tubular device at 0.5 and 2 L min⁻¹, respectively. These values illustrate the impacts of external transport and the fact that the tubular device was subject to more external mass transfer limitations than the CSTR. Figure 8A and 8C demonstrate the model fit to a single replicate condition. The fitted k values were then used to predict DO time series for other pressure conditions (Figure 8B and 8D). The results illustrate the model predictability for select CSTR and tubular experiments, showing good match with the experimental data especially below 3 mg L⁻¹ which is most physiologically relevant. However, the model did not predict the 400 rpm CSTR well. The poor fit for the 400 rpm CSTR is potentially the result of external forces providing oxygen to the system not taken into account in the model, which is magnified under low total oxygen flux into the system. The other conditions tested had higher total oxygen transfer from HFMs, thus reducing this impact.

Figure 8.

Model fitting for the convective mass transfer coefficient for the CSTR (A) and the tubular device (C) for a single replicate, and model predictions vs. experimental data for experiments with the CSTR (B) and tubular device (D). Figures B and D use a fraction of the total data set to increase clarity. Model fitting used replicates of 0.24 bar experiments to find an average k value (one set of CSTR and tubular) for each mixing scenario (400/1200 rpm and 0.5/2.0 L min⁻¹). Figures for the 400 rpm and 0.5 L min⁻¹ fit are provided in SI.

The sensitivity of the model to key parameters was investigated next. Figure 9 reports the sensitivity of both the CSTR (A) and tubular (B) systems to changes in the convective mass transfer coefficient (k) for lumen pressures ranging from 0.25 to 2 bar of oxygen. Examination of the results highlights the different mass transfer limitation regimes. At low k values, the oxygen flux is sensitive to increases in k, an indication that external mass transfer is in part limiting, and that mixing will increase the oxygen flux. On the other hand, at high k values, the flux becomes independent to k values, as oxygen transport is primarily limited by diffusion through the membrane. For all conditions, as expected, the flux increases with increases in pressure, as it increases the overall driving force for oxygen transfer. The vertical bars in Figure 9 indicate the nominal values for the actual experiments. They highlight that all the systems tested would benefit from greater mixing. The difference between the CSTR and tubular systems sensitivity was not discernable. This is because the DO increase along the tubular device remains relatively low, thus that the impact of the axial decrease of the oxygen gradient driving the transfer is insignificant for single fiber systems.

Figure 9.

Sensitivity of instantaneous oxygen flux to the mass transfer coefficient k and to membrane oxygen pressure. The reported fluxes are for a bulk (CSTR) or outlet liquid (tubular) partial pressure of 30 mmHg. The fiber length was set to 30 cm in both cases while pressure inside the fiber was held constant across the entire length. Vertical lines represent actual mass transfer coefficients for CSTR (A) and tubular (B) systems at fitted rpm or flow rate. Note the log scale for the x axis.

Figure 10 illustrates the potential benefits of reducing membrane wall thickness for various convective mass transfer conditions (k = 0.001 to 0.1 cm s⁻¹). As expected, when the external mass transfer becomes less limiting (higher k values), reducing the wall thickness has a greater impact on the system flux. Conversely, for thick membranes (>70 µm), the resistance to transport is higher in the membrane than liquid boundary layer, and the impacts of k are limited. This figure also shows that there will be significant benefits to decrease the membrane thickness in the current prototypes.

Figure 10.

Sensitivity of instantaneous flux to membrane wall thickness of the CSTR model and to the convective mass transfer coefficient (0.001 to 0.1 cm s⁻¹) at a bulk liquid partial pressure of 30 mmHg. The fiber oxygen pressure was held constant across the length of fiber at 0.25 bar. Actual membrane wall thickness of prototypes was 89 µm (vertical bar).

The sensitivity of oxygen transfer to the membrane material permeability to oxygen is reported in Figure 11. The material used in this work was Teflon AF 2400, which has one of the highest permeabilities to oxygen. Thus, limited increases can be expected, although Figure 11 shows that improvements to permeability would result in much greater oxygen fluxes, especially when convective transport is high. Altogether, Figure 10 and 11 show that the total resistance to mass transfer from the membrane will be a key factor in sizing intravascular oxygenation devices. In particular lowering of the membrane wall thickness will both increase the flux across the fiber as well as reduce the total volume of fibers required to achieve a target oxygen flow.

Figure 11.

Sensitivity of instantaneous flux to membrane permeability of the CSTR model and to the convective mass transfer coefficient (0.001 to 0.1 cm s⁻¹) at a bulk liquid partial pressure of 30 mmHg. The fiber oxygen pressure was held constant across the length of fiber at 0.25 bar and membrane thickness was 89 µm. The vertical bar shows the permeability of the Teflon AF 2400 membrane (9.9 × 10⁻⁸ cm² s⁻¹ cm⁻¹_Hg).

Finally, we used the model to determine if a device using high pressure oxygen could potentially achieve physiologicaly relevant levels of oxygen transfer, while maintaining low total surface area. Figure 12 shows two contour plots reporting the total system oxygen transfer rate at 30 mmHg oxygen partial pressure in water, over a range of membrane surface area and lumen pressure. At a low mass transfer coefficient (tubular system value used), representing low mixing, oxygen transfer rates as high as 35 mL min⁻¹ are predicted while staying under 0.15 m² total surface area and below 2.0 bar of oxygen pressure (A). A hypothetical increase in the mass transfer coefficient by 10 times by a simulated increase in mixing more than triples that to over 100 mL min^-1. These model results show the potential of introducing hyperberic oxygen through non-porous HFM as a means to deliver a clincally signifacant amount of oxygen. This novel approach to intravascular oxygenation allows for a more compact device with higher oxygen transfer rates when compared to previous works.

Figure 12.

Contour plot for total oxygen transfer rate as a function of pressure and total fiber area, for two different mass transfer coefficients (k). A) k value for fitted tubular system at 2 L min⁻¹. B) 10-fold increase simulating significant increased mixing.

4. Conclusions

Alternative treatments for hypoxic respiratory failure that function independent of the lungs are clearly needed. This has become more apparent with the worldwide COVID-19 pandemic which has overwhelmed ICUs around the world. An inexpensive, easy to deploy device that provides oxygen directly to the bloodstream, augmenting mechanical ventilation and possibly reducing the need for ECMO, would find broad and immediate use in a variety of clinical settings.

The work presented here describes fundamental engineering research in support of such a device which could be deployed intravascularly and provide a portion of a patient’s oxygen needs irrespective of the degree of lung injury. High oxygen fluxes were observed with our non-porous membrane devices reaching over 550 mL min⁻¹ m⁻² under hyperbaric oxygen supply. The work presented here demonstrates that operation under hyperbaric conditions allowed us to achieve oxygen fluxes greater than any previous work. Experiments showed that factors limiting the transfer of oxygen were both internal (diffusion through the membrane) and external (convection around the fibers). The amount to which these factors affected flux could be captured in a conceptually correct analytical model of the system. The model facilitated the conduct of a sensitivity analysis on the system and was used in a predictive fashion to determine the surface area required for a clinically meaningful device. Sensitivity analyses to model parameters showed differentiation between key mass transfer regimes and demonstrated that such a system will need significant mixing to take advantage of higher oxygen pressures. Our model predicts this approach has potential to deliver significant amounts of oxygen in a device that readily fits intravascularly. As we scale up, it can be expected that the membrane efficiencies will decrease as surface area increases though the system will be able to compensate for some amount of this lost efficiency by increasing oxygen pressure. We show that an important limitation in any intravascular oxygenation device that achieves such high fluxes is the risk of bubble formation, which will ultimately limit the maximum pressure at which a device could safely operate and, therefore, limit the upper bounds of safe oxygen transfer. These studies constitute the foundation for a new generation of intravascular oxygenation devices which could supply clinically significant amounts of oxygen to patients with hypoxic respiratory failure by overcoming mass transfer limitations with the use of hyperbaric pure oxygen flowing through non-porous hollow fiber membranes.