Introduction

Non-invasive ventilation (NIV) defines a method of mechanical ventilation that does not require endotracheal intubation. NIV delivers pressurized gas to the airways through one of a variety of interfaces, such as a mouthpiece or a nasal or facial mask. Over the last two decades, the use of NIV has increased markedly in the treatment of acute and chronic respiratory failure [1, 2]. In an acute setting, NIV reduces the risk of endotracheal intubation [3] and facilitates the weaning process from mechanical ventilation [4]. Nevertheless, for various reasons, NIV is not always successful. The rate of failure for this procedure in acute patients ranges from 7 to 49% [2, 57]. One of the reasons reported in the literature to explain this failure rate is the poor patient tolerance to the NIV interface [8, 9]. Mask discomfort, sensation of excessive air pressure, intolerance, skin lesions, leakage and rebreathing have been cited as the most relevant parameters to explain the interface-related failure of NIV. A large variety of interfaces are presently available ranging from the simple nasal or oronasal mask to the helmet recently proposed [10, 11]. To our knowledge, the efficacy of these different interfaces has essentially been tested in clinical trials or in bench studies based upon the comparison of clinical indexes such as breathing pattern, airway pressure, neuromuscular drive, inspiratory muscle effort, work of breathing (WOB), arterial blood gases and dyspnea [1, 9, 1218]. All these interfaces allow successful non-invasive ventilation. To a great extent, clinical efficacy was not found to be different during NIV conducted with interfaces whose volumes very much differed [1, 9, 12, 18]. Only WOB and inspiratory muscle effort were slightly less decreased with the helmet, in comparison to the other interfaces [12, 18], but these differences were abolished by a specific ventilator setting [18]. At first sight, this fact is surprising because some of these interfaces have a large internal volume associated with dead space that should not permit successful ventilation. In theory, the dead space should remain as small as possible because any increase of it induces a rise in patient’s WOB [19].

We postulated that due to the streaming effects of the gas passing throughout the interface, the effective dead space of the interface could be quite different from the volume delimited by the interface and the part of the body embedded in this interface (called interface gas region). To test this assumption, we used numerical simulations with computational fluid dynamics (CFD) software to describe pressure, flow and gas composition in four types of interfaces regularly used to deliver NIV in the intensive care unit (ICU). In contrast with clinical trials, CFD has the advantage of allowing this set of interfaces to be tested under strictly identical conditions. In a previous in vitro bench study, Saatci et al. [20] estimated the interface dead space without flow or pressure measurements. However, these authors tested several face masks but not the helmet interface.

Materials and methods

We studied four interfaces commonly used in the ICU (see Table 1 for interface characteristics): two oronasal face masks (masks #1 and #2), an integral mask (mask #3) and a helmet-type interface which covers the entire head and neck (mask #4). Each interface was successively positioned upon the same adult mannequin head made of polystyrene. Each set-up, i.e. the mannequin–interface, was numerically reconstructed in three dimensions by using computed tomography (CT) images. In each three dimensional (3D) geometry, breathing flow was simulated for successive cycles (i.e. inspiratory + expiratory phases) by using CFD software (FLUENT® Fluent Inc., Lebanon, TN, USA) until an “equilibrium state” was reached. The equilibrium state is defined as the moment where the parameters (i.e. pressure, flow and gas composition) become identical from one breathing cycle to the next breathing cycle. Details about 3D reconstruction and CFD procedure are described in the online electronic supplementary material (ESM). The inspiratory phase consisted of blowing a constant 600 ml/s flow rate during 1.25 s through the inlet of the interface. The expiratory phase consisted of blowing a constant 300 ml/s flow rate during 2.5 s through the mouth of the mannequin. This setting corresponds to a breathing pattern of 750 ml tidal volume (V T), 16 beats/min frequency and a ratio of 2 between inspiratory and expiratory times. During the inspiratory phase, the insufflated gas called “fresh air” was normal air, fully humidified and heated at 37°C (Fig. 1). During the expiratory phase, the gas blown in the interface, called “exhaled air” in Fig. 1, was air enriched with CO2 (see Table 2 for the exact compositions of fresh air and exhaled air). At the initial time of the first breathing cycle, the interface gas region was totally filled with fresh air. The CFD procedure allows one to estimate gas composition, flow and pressure at any locus of the interface gas region and at any time of the breathing cycle. Gas composition was described as the mass fraction of each gas species, especially for oxygen (\( {\text{MF}}_{{{\text{O}}_{2} }} \)) and carbon dioxide (\( {\text{MF}}_{{{\text{CO}}_{2} }} \)). Then, we determined the effective dead space (V D) of the interface gas region defined as the rebreathed gas volume resulting from the exhaled gas trapped in this interface gas region as well as the volume of fresh air (V fresh) that is inhaled by the mannequin during each inspiratory cycle.

Table 1 Non-invasive ventilation interfaces
Fig. 1
figure 1

Schematic of the set-up showing the interface positioned on the polystyrene mannequin head. CFD computational fluid dynamics

Table 2 Gas compositions

The pressure in the interface is composed of a mean pressure (e.g. atmospheric pressure, pressure support, positive end-expiratory pressure) and the variation of pressure induced by the flow. This variation of pressure, which is computed by the CFD procedure, does not depend on the mean pressure as long as the thermodynamic properties of the gas are not affected by the mean pressure. Therefore, our results theoretically apply to the different pressure regimes or ventilator settings, constant or intermittent, used in clinical practice during NIV.

Results

The equilibrium state was estimated from the time course of \( {\text{MF}}_{{{\text{O}}_{2} }} \) at the mouth from one cycle to the following cycle. For the first three interfaces shown in Table 1, the equilibrium state was quickly reached: the \( {\text{MF}}_{{{\text{CO}}_{2} }} \) curves become identical after less than six breathing cycles. To reach the equilibrium state with the helmet took a little more time: approximately 2 min (i.e. after 32 breathing cycles).

During the inspiratory phase, \( {\text{MF}}_{{{\text{O}}_{2} }} \) at the mouth of the mannequin globally increases with time, while \( {\text{MF}}_{{{\text{CO}}_{2} }} \) decreases (see Fig. 2). \( {\text{MF}}_{{{\text{O}}_{2} }} \) represents the quantity of oxygen available for the lung at the equilibrium state, whereas \( {\text{MF}}_{{{\text{CO}}_{2} }} \) quantifies the amount of CO2 rebreathing. With the two oronasal masks (#1 and #2), i.e. the smaller interfaces, the \( {\text{MF}}_{{{\text{O}}_{2} }} \) tended to reach the maximum theoretical value (22.2%, i.e. the oxygen concentration of the total fresh air) by the end of the inspiration time. By contrast, with the two other interfaces (#3 integral and #4 helmet), \( {\text{MF}}_{{{\text{O}}_{2} }} \) at the mouth of the mannequin was never above 20% during the inspiratory cycle. The \( {\text{MF}}_{{{\text{CO}}_{2} }} \) behaviour was to decrease throughout the inspiratory phase from its maximum value (7.2%) to zero (0%) for the two oronasal masks, and to remain always positive for the two other masks: 2.4% for integral mask and 3.2% for helmet (see Fig. 2).

Fig. 2
figure 2

Evolution of the mean mass fraction of oxygen (\( {\text{MF}}_{{{\text{O}}_{2} }} \) on left vertical axis) and carbon dioxide (\( {\text{MF}}_{{{\text{CO}}_{2} }} \) on right vertical axis) at the mouth of the mannequin during the inspiratory period once the equilibrium state is reached between two successive cycles (see text). The inspiration begins at time (t = 0) and stops at time (t = 1.25 s). Dashed lines indicate the mass fractions of fresh and exhaled air

The effective dead space characterizing the four interfaces varied from 110 to 370 ml (see Fig. 3). The larger V D corresponds to the helmet, and represents only 4% of the volume of the interface gas region. In this case, the effective dead space increases linearly with time and inspired volume (see Fig. 4). The smallest V D was obtained with the oronasal mask #1 which has the smallest interface gas volume, and the V D was about 100% of the interface gas volume (see Fig. 3). For this oronasal mask #1, V D was attained within the first 0.35 s of the inspiratory cycle, before following a plateau during the remainder of the cycle (see Fig. 4). The two others masks, i.e. oronasal mask #2 and integral mask, behaved intermediately. The V D profile (see Fig. 3) of oronasal mask #2 during the inspiration was close to that of oronasal mask #1, with a V D reaching 93% of the interface gas volume. The integral mask behaviour was not far from that of the helmet, with a V D equal to only 35% of the interface gas volume (see Fig. 3).

Fig. 3
figure 3

Effective dead space volume, volume of inspired fresh air and interface gas volume for the four masks presented in Table 1. These data are given once the equilibrium state is reached between two successive cycles

Fig. 4
figure 4

Evolution of the effective dead space volume during the inspiratory period once the equilibrium state is reached between two successive cycles. The inspiratory flow begins at time (t = 0) and stops at time (t = 1.25 s)

For all tested interfaces, the pressure field was relatively homogeneous with a maximum pressure variation not exceeding 0.3 cmH2O.

Discussion

The main finding of this study is that the effective dead space, which is never higher than the interface gas volume, can be much smaller due to the streaming effect presently calculated by CFD. For all the interfaces the effective dead space was under 370 ml for a tidal volume of 750 ml. This is of clinical interest because it suggests that even interfaces with extremely high internal volumes (10 l) might yet be used in patients. This holds true to the extent that patients are capable of inhaling an inspired gas slightly enriched with CO2 (a maximum 2.1% of CO2 in volume for the helmet). The finding that effective interface dead space is not linked to the interface gas volume is consistent with several clinical studies. These studies have reported a lack of differences, in terms of patient’s respiratory effort, breathing pattern and gas exchange, among interfaces with disparate internal volumes [1, 9, 16, 18, 21, 22].

Importantly, our results also show that with a tidal volume of 750 ml, the less voluminous mask (oronasal mask #1) clearly remains the most favourable in terms of rebreathing, because the V D is reduced by a factor of 3.4 when compared to the V D of the most voluminous masks (helmet) (Fig. 3). Nevertheless, this beneficial effect decreases quasi-linearly as tidal volume decreases (Fig. 4). For example, with a tidal volume of 400 ml, the previous factor is as low as 1.8. This means that the choice of interface based only on the effective V D is probably questionable. Factors such as comfort, tolerance and leaks should also be taken into account during the decision-making process.

In the extreme case of tidal volumes below 200 ml, the most efficient interface in terms of rebreathing becomes the most voluminous (the helmet). Of course, such a small tidal volume does not fit standard adult ventilation. However, if we consider the non-invasive high-frequency oscillatory ventilation at low tidal volume, as recently proposed in neonatology [23], our results suggest that a helmet interface could be potentially more effective than a nasal or oronasal interface due to the lower CO2 rebreathing. This result is not in disagreement with the fact that the helmet is not the interface of choice for NIV in COPD patients with acute exacerbation. Indeed, a 200-ml tidal volume is too low in breathing adult COPD patients to result in adequate alveolar ventilation. With these patients the required tidal volume is higher, i.e. in range where the helmet is the less efficient interface in terms of rebreathing.

The streaming effect behind effective dead space can be physically understood on the basis of the ratio between tidal volume and interface gas volume. When the interface gas volume is much larger than the tidal volume, the convective flows are reduced and the gas composition in the interface, i.e. the mass fraction of each gas species, is poorly affected by a single breathing cycle. The quasi-constancy of \( {\text{MF}}_{{{\text{O}}_{2} }} \) and \( {\text{MF}}_{{{\text{CO}}_{2} }} \) at the mouth in the helmet mask throughout the entire inspiratory phase illustrates this phenomenon (Fig. 2). This assumption was already implicitly developed by Taccone et al. [17] who described the helmet functions as a semi-closed environment. Accordingly, the corresponding mass fraction becomes, at the equilibrium point, the mean between the mass fractions of fresh air and exhaled air and the V D is half the tidal volume (e.g. V D = 49% of V T in the helmet). On the basis of a negligible convective flow it suggests that this result (V D ≅ 50% of V T) may be generalized to the other helmets.

By contrast, when the interface volume is small compared to the tidal volume (oronasal mask #1 and #2), the convective flows are important. In this scenario, the recirculation velocities are relatively high and are also present in the quasi-totality of the interface gas region. In this case, the interface gas volume is washed out by the breathing flow (i.e. both the inhaled and the exhaled gas) and the dead space gets closer to the interface internal volume (e.g. 100% of the interface gas volume for the oronasal mask #1 and 93% for the oronasal mask #2). On this basis, we can predict that the “small” masks (oronasal) have a V D close to their internal volume.

With the integral mask, where the gas entry is at the level of the chin, the role of convection is intermediate and the tidal volume is close to the interface gas volume. In the first part of an inspiration (i.e. for the first 0.35 s, corresponding to about 200 ml of inspired gas), the inferior part of the mask (from the chin to the mouth) is rapidly washed out. During this time, \( {\text{MF}}_{{{\text{O}}_{2} }} \) and V D are close to the values observed with the two smallest masks (see oronasal masks #1 and #2 in Figs. 2 and 4). After this short period of time, \( {\text{MF}}_{{{\text{O}}_{2} }} \) at the mouth reaches a plateau resulting from gas mixture made of the gas flow coming directly from the ventilator (fresh air) and partial entrainment of the gas residing between the mouth and forehead (see the video showing the time course of the \( {\text{MF}}_{{{\text{O}}_{2} }} \) during the inspiration and expiration in the online ESM). The fact that V D is impacted by both convective and diffusive phenomena suggests that the results of our simulation cannot be generalized to the other integral mask to predict the V D. Changing geometry may change the balance between convective and diffusion phenomena that will result in a different V D.

It is noteworthy that the physical streaming effects do not require important pressure variations in the interface gas region. This observation confirms that these interfaces are slightly resistive and are not responsible for a significant amount of resistive work. Chiumello et al. [12] found that the helmet induced an increase of WOB associated with an increase of the time to reach the pressure support without any increase of the tidal volume, whereas Vargas et al. [18] found that using a specific setting (increase of the pressure support and PEEP with the highest pressurization rate) allowed one to abolish this greater inspiratory muscle effort. The latter results suggest that this increased WOB observed with the helmet is explained in part by trigger and pressurization difficulties and not solely by a large dead space problem. Such an assumption is in agreement with the relatively low value of V D that we obtained.

The present findings can be partially extended to clinically relevant changes in oxygen and carbon dioxide concentrations. In this study we used normal air, heated and humidified, as inspiratory gas and we used air enriched with 5% of CO2 as the expired gas. Oxygen therapy as well as O2–CO2 exchange within the lung modify gas composition. As long as physical and thermodynamic properties of the gas are not affected by gas composition and/or temperature, the effective dead space inferred in this study remains unchanged. This means that if convective and diffusive phenomena are unchanged, the effective dead space is also unchanged. Considering the slight modification in thermodynamic properties of the gas (<10% for gas density and less than 4% for gas kinematic viscosity) induced by either a variation of O2 concentration (from 15 to 100%) or a variation of CO2 concentration (from 0 to 5%), we can postulate that our results remain valid whatever the oxygen therapy level administered. Note that a change in O2 concentration in the fresh air only requires one to change the maximal and minimal values corresponding to fresh and exhaled air, in the scales of Fig. 2.

One of the limitations of this study was that the numerical tool did not allow us to simulate non-intentional leak. Nevertheless, it is clear that a bias flow leak tends to decrease the rebreathing problem, but it also tends to induce auto-triggering and to decrease the effective tidal volume inspired by the patient [24, 25]. However, the importance of these phenomena is extremely contingent upon a set of conditions such as ventilator mode, trigger algorithms or maximum flow deliverable by the ventilator. Another limitation of this study is that we implicitly assumed the rigidity of the wall interface. This assumption is particularly questionable with the helmet. However, a recent study [18] concluded that increasing both pressure support level and positive end-expiratory pressure may be clinically advisable when providing NIV via a helmet. Such a higher pressurization, which reduces the compliance of the helmet, tends to justify our rigidity assumption.

In summary, we have shown that mathematical modelling and computational fluid dynamics can estimate the effective dead space of NIV interfaces by analysing the pressure field/flow pattern and variations in gas composition inside these interfaces. New research efforts will be dedicated in the future to the translation of these mathematical models to clinical environments so as to increase efficiency while decreasing the side effects of the clinical treatments. The effective dead space of the interfaces used for NIV was not directly related to the internal volume of the interface. For the most voluminous interfaces, effective dead space is limited to half the tidal volume, whereas the effective dead space is close to the interface gas volume only for the less voluminous interfaces. These findings allow physicians to choose among a large number of masks that may be suitable for an individual patient and also allow physicians to take into consideration other parameters such as comfort and synchrony when choosing an interface.