Non-invasive Ventilation

Ventilator performances for non-invasive ventilation: a bench study

Abstract

Introduction A wide range of recent ventilators, dedicated or not, is available for non-invasive ventilation (NIV) in respiratory or intensive care units (ICU). We conducted a bench study to compare their technical performances.

Methods Ventilators, including five ICU ventilators with NIV mode on, two dedicated NIV ventilators and one transport ventilator, were evaluated on a test bench for NIV, consisting of a 3D manikin head connected to an ASL 5000 lung model via a non-vented mask. Ventilators were tested according to three simulated lung profiles (normal, obstructive, restrictive), three levels of simulated air leakage (0, 15, 30 L/min), two levels of pressure support (8, 14 cmH2O) and two respiratory rates (15, 25 cycles/min).

Results The global median Asynchrony Index (AI) was higher with ICU ventilators than with dedicated NIV ventilators (4% (0; 76) vs 0% (0; 15), respectively; p<0.05) and different between all ventilators (p<0.001). The AI was higher with ICU ventilators for the normal and restrictive profiles (p<0.01) and not different between ventilators for the obstructive profile. Auto-triggering represented 43% of all patient-ventilator asynchrony. Triggering delay, cycling delay, inspiratory pressure-time product, pressure rise time and pressure at mask were different between all ventilators (p<0.01). Dedicated NIV ventilators induced a lower pressure-time product than ICU and transport ventilators (p<0.01). There was no difference between ventilators for minute ventilation and peak flow.

Conclusion Despite the integration of NIV algorithms, most recent ICU ventilators appear to be less efficient than dedicated NIV ventilators. Technical performances could change, however, according to the underlying respiratory disease and the level of air leakage.

What is already known on this topic

  • With non-invasive ventilation (NIV) algorithms, intensive care unit (ICU) ventilators are more secure to deliver NIV but seem to be less efficient than NIV dedicated ventilators.

What this study adds

  • We made an up-to-date comparison between recent ICU ventilators, transport and NIV dedicated ventilators. NIV dedicated ventilators are still more efficient than ICU ventilators to deliver NIV. Technical performances could vary, however, according to the underlying respiratory disease and the level of simulated air leakage.

How this study might affect research, practice or policy

  • NIV dedicated ventilators should be first-line ventilators to deliver NIV according to our study and our results should be confirmed in a clinical study.

Introduction

Since the beginning of the 2000s, non-invasive ventilation (NIV) has been increasingly used as first-line therapy in intensive care unit (ICU) patients with acute respiratory failure (ARF) with an increasing effectiveness to prevent intubation and its complications.1 Recently, the European Respiratory Society and the American Thoracic Society jointly issued strong recommendations for the NIV management of hypercapnic ARF, particularly in acute exacerbations of chronic obstructive pulmonary disease and acute cardiogenic pulmonary oedema.2 In addition to the choice of NIV interface, and despite a high variability in the use of different ventilator’s category in daily practice,3 it is essential that those made available for the clinician are reliable in these indications of NIV to ensure effective and safe ventilation in unstable patients at risk of intubation.

Different types of ventilators can be used for NIV including transport ventilators if needed.4 Although ICU ventilators were originally designed for NIV without any intentional leakage due to a double inspiratory and expiratory circuit, significant non-intentional leakage may occur between the mask and the patient’s face leading to patient-ventilator asynchrony (PVA).5 These PVAs could generate respiratory discomfort and poor tolerance of NIV,6 increase the duration of mechanical ventilation7 8 and the risk of mortality.9

In recent years, ICU ventilator performances have been improved by integrating NIV algorithms, the so-called ‘NIV mode’, in their technological design to better compensate and manage non-intentional leakage, to generate less PVA and provide better respiratory comfort for patients.10 NIV can also be delivered with specifically designed ventilators derived from home ventilator technology. Using a turbine generator and a single inspiratory circuit, these dedicated NIV ventilators were developed to deliver pressure support ventilation via an interface integrating intentional leakage.4 Previous clinical and physiological bench test studies compared the technical performances of these different types of ventilators to deliver NIV.11–13 Dedicated NIV ventilators were found to better compensate non-intentional air leakage and provided better performance, efficacy and tolerance of NIV than ICU ventilators by reducing the risk of PVA.14 Despite integrating NIV algorithms in ICU ventilators, their performance remains inferior to compensate non-intentional air leakage, with poorer patient-ventilator synchronisation than dedicated NIV ventilators.15 16 Very recently, high-performance modern turbine-driven ventilators were also tested in bench studies using different circuit configurations and different interfaces, including the helmet.17

To our knowledge, the technical performances of the most recent ventilators have not been evaluated. The aim of our study was, therefore, to evaluate the performances of the last generations of ICU, dedicated NIV and transport ventilators, in a bench model, according to different respiratory conditions and non-intentional air leakage.

Materials and methods

This experimental study was conducted in the Medical Intensive Care Department of the Rouen University Hospital.

Patient and public involvement

It was not appropriate to involve patients or the public in our study because it was a bench study. We had designed it before patient and public involvement was common practice.

Experimental test bench and Simulations

To simulate a patient’s head, we used a 3D-printed manikin head developed and validated in previous studies18 19 (online supplemental e-Figure 1). All tested ventilators were connected to the manikin head using a Quattro FX NV, size M, (Resmed, San Diego, CA, USA) facemask without integrated leakage, and a dual inspiratory and expiratory circuit with an antimicrobial filter on each branch. The manikin trachea was connected to an ASL 5000 artificial lung (Ingmar Medical, Pittsburgh, PA, USA) (online supplemental e-Figure 2).

We simulated three patterns of underlying lung profile: (a) a restrictive pattern with a respiratory system compliance of 30 mL/cmH2O and an airway resistance of 8 cmH2O.L/s; (b) an obstructive pattern with a compliance of 50 mL/cmH2O and an inspiratory and expiratory resistance of 20 cmH2O.L/s and 25 cmH2O.L/s, respectively; (c) a normal pattern with a compliance of 60 mL/cmH2O and an airway resistance of 5 cmH2O.L/s.20–22 For each lung profile simulated, the respiratory rate was set at 15 cycles/min (cpm) and then 25 cpm, and the P0.1, that is, the airway occlusion pressure reflecting the inspiratory drive and indirectly the level of effort—usually increased in case of ARF,23 24 was set at 2.5 cmH2O.17 21 The respiratory muscle pressure resulting from these settings has been modelised in online supplemental e-Figure 3 as an example. We also simulated, as frequently observed in clinical practice,14 three levels of simulated air leakage between the interface and manikin’s face: 0 L/min, 15 L/min and 30 L/min.

Two pneumotachographs were connected to the test bench circuit to measure air leakage. Each level of simulated air leakage was calibrated using the pneumotachograph placed on an adjustable valve in derivation between the facemask and the ventilator, applying a continuous positive airway pressure (CPAP) of 10 cmH2O (online supplemental e-Figure 2). The pressure inside the facemask was measured using a DP 15–32 pressure transducer (Validyne Engineering, Northridge, CA, USA) directly connected to the mask.

Before each experimentation, the mask was fitted to minimise leakage with the aim to achieve less than 5 L/min at a CPAP level of 10 cmH2O. The pressure transducer and both pneumotachographs were also calibrated before each simulation.

All measured data were collected with a Biopac MP150 v3.9.1 analyser (Biopac Systems, Goleta, CA, USA) sequenced at 200 Hz with a 10-Hz filter and the ASL 5000 sequenced at 512 Hz.

Tested ventilators

Eight ventilators were tested: five ICU ventilators (Evita-V800, Evita-V500, Evita-XL (Dräger, Lübeck, Germany) Monnal-T75 (Air Liquide Medical Systems, Paris, France), Servo-U (Maquet, Rastatt, Germany)), two dedicated NIV ventilators (Astral-150 (Resmed, San Diego, CA, USA), Trilogy-Evo (Philips Respironics, Murrysville, PA, USA)) and one transport ventilator (Monnal-T60 (Air Liquide Medical Systems, Paris, France)), all used in our respiratory and medical ICU (online supplemental e-Table 1). Regarding the integrated technology, both dedicated NIV and transport ventilators were turbine-driven ventilators; whereas, all ICU ventilators were compressed-gas ventilators except the Monnal-T75 (online supplemental e-Table 1). For each tested ventilator, we calibrated the circuit before experimentations as proposed by the ventilator software.

All ventilators were set in pressure support ventilation (PSV) with a positive end-expiratory pressure (PEEP), that is, PSV+PEEP, each ventilator was tested with its NIV algorithm activated. For each ventilator, two PSV levels were applied, 8 and 14 cmH2O;25 whereas, the PEEP level was maintained at 5 cmH2O for each level. The sensitivity of the inspiratory trigger was set at 1 L/min or as sensitive as possible depending on the ventilator. The other preset parameters were inspiratory time (Ti) at 1.2 s, inspiration/expiration cycling (expiratory trigger) at 50%, PSV rise slope at 200 ms and the FiO2, if applicable, at 21%.

Endpoints and measurements

The primary endpoint was the patient-ventilator Asynchrony Index (AI),26 calculated as follows7 27:

AI=Number of asynchronies/global respiratory rate (ventilatory cycles+ineffective efforts) × 100 (%).

Secondary endpoints were five different categories of PVA with the following classification10 28: ineffective effort, auto-triggering, double triggering, early inspiratory/expiratory cycling, and late inspiratory/expiratory cycling.

Other secondary endpoints were represented by the following ventilatory parameters: the inspiratory flow preceding the triggering of the ventilator (L/min), the inspiratory triggering delay of the ventilator, represented by the delay between the beginning of simulated inspiratory effort and the mechanical insufflation (ms), the inspiratory rise time, represented by the delay between the onset on pressure support and the reaching of the target pressure (ms), the expiratory cycling delay of the ventilator, represented by the delay between the end of simulated inspiratory effort and the expiratory cycling of the ventilator (ms), the patient’s work of breathing, meaning the amount of work to trigger the ventilator (WOBt), indirectly expressed as pressure-time product during the triggering phase (PTPt, cmH2O.s), the peak flow delivered by the ventilator (L/min), the pressure applied in the facemask (cmH2O) and the minute volume generated in the artificial lung (L/min) (online supplemental e-Figure 4).

These parameters were evaluated for each ventilator according to the three ventilatory patterns previously described (normal, obstructive, restrictive) and, for each profile, according to two respiratory rates (15 and 25 cpm) and three levels of calibrated simulated leaks (0, 15 and 30 L/min).

All the data were processed and synchronised automatically with a custom software (Kernel Biomedical, Bois-Guillaume, France) developed with GNU Octave 8.2.18

Statistical analysis

Given their non-parametric distribution, the variables measured or calculated cycle-to-cycle were expressed as median with first and third interquartiles. Inter-ventilator comparison tests were performed with the Friedman test. In case of significance, pairwise comparisons of the ventilators were done with Dunn’s post hoc test. Comparisons between subgroups (ventilator categories, ventilatory patterns) were carried out with the Kruskal-Wallis Dunn’s post hoc test. We used the Mann-Whitney U test to assess the impact of respiratory rate and PSV level on NIV performances. Spearman’s test was used to look for correlations between different variables. A p value of 0.05 was considered statistically significant for all comparisons.

Analyses were carried out with GraphPad Prism 9.0.0 software (GraphPad Software Inc., CA, USA).

Results

Simulation comparability and experimental validity

There was no significant difference in the median total leakage value between the different ventilators (p=0.09) (online supplemental e-Figure 5 and e-Table 2). There was a correlation between the median leakage and the increase in AI for all ventilators: 0% (0; 7) for 0 L/min, 4% (0; 51) for 15 L/min and 16% (0; 76) for 30 L/min (p<0.001).

For all ventilators, AI was higher for a PSV level of 14 cmH2O compared with 8 cmH2O: 4% (0; 66) vs 0% (0; 32); p<0.01. No difference was observed in AI according to respiratory rate: 0% (0; 64) at 25 cpm vs 4% (0; 36) at 15 cpm; p=0.37.

A significant difference was observed in AI between ventilators (p<0.001). Ventilators that generated the least PVA were Servo-U (0% (0; 4)) and Trilogy-Evo (0% (0; 8)), and those that generated the most were Evita-500 (24% (0; 95)), Evita-XL (14% (0; 99)) and Evita-800 (8% (0; 93)) (figure 1). ICU ventilators generated a higher AI than dedicated NIV ventilators (4% (0; 76) vs 0% (0; 15), p<0.05) and also higher compared with the transport ventilator but not significantly (4% (0; 76) vs 0% (0; 30)), p>0.12) (figure 2).

Figure 1
Figure 1

Comparison of Asynchrony Index (AI) between ventilators. Values are aggregated across the three lung models and the two pressure support levels, and expressed as median (point) and interquartile interval (bars). *p<0.05, **p<0.01.

Figure 2
Figure 2

Comparison of Asynchrony Index (AI) between ventilator categories. Values are aggregated across the three lung models and the two pressure support levels, and expressed as median (point, square, triangle) interquartile interval (bars). *p<0.05.

Role of ventilatory patterns on AI

For the restrictive pattern, AI was significantly different between ventilators (p<0.001), with three ventilators generating significantly more PVA: Evita-XL (100% (93; 100)), Evita-500 (98% (70; 100)) and Evit-800 (96% (81; 100)) (figure 3A).

Figure 3
Figure 3

Comparison of Asynchrony Index (AI) between ventilators. Values are aggregated across the two pressure support levels, and expressed as median (point) and interquartile interval (bars). *p<0.05, **p<0.01. (A) Restrictive ventilatory pattern. (B) Normal ventilatory pattern. (C) Obstructive ventilatory pattern.

For the normal pattern, there was also a significant difference in AI between ventilators (p=0.003). The ICU ventilator generating the most PVA was Monnal-T75 (14% (1; 25)) and Astral-150 (2% (0; 74)) for the dedicated NIV ventilator (figure 3B).

AI was not found significantly different between ventilators for the obstructive pattern (p=0.93) (figure 3C).

Role of simulated air leakage on AI

For simulated air leak of 0 L/min (p=0.06) and 15 L/min (p=0.09), no significant difference was observed in AI between ventilators (online supplemental e-Figure 6A, B). However, when the leak increased to 30 L/min, the AI was significantly different between ventilators (p<0.001). A significant difference was observed between Servo-U (0% (0; 4)) and Evita-500 (64% (4; 96)), Evita-800 (60% (2; 100)) and Evita-XL (98% (68; 100)), and between Evita-XL and Trilogy-Evo (0% (0; 8)) (online supplemental e-Figure 6C).

Different types of PVA and parameters of ventilatory mechanics

Considering all ventilators, auto-triggering represented the most frequently observed PVA in 43% of cases followed by early inspiratory/expiratory cycling (25%) and late inspiratory/expiratory cycling (24%).

The PVA types for each ventilator are shown in the online supplement (online supplemental e-Table3). A significant difference was observed in ineffective efforts (p<0.01) and early inspiratory/expiratory cycling (p<0.001) between the three ventilator categories.

Inspiratory delay, PTPt (figure 4), pressure rise time, expiratory cycling delay and pressure at mask were different between ventilators (p<0.001) (online supplemental e-Table 4) and between ventilator categories (p<0.01) (table 1).

Figure 4
Figure 4

Comparison of inspiratory pressure time product (PTPt) between ventilators. Values are aggregated across the three lung models and the two pressure support levels, and expressed as median (point) and interquartile interval (bars). *p<0.05, **p<0.01.

Table 1
|
Ventilator performances (ventilatory mechanics) of each category of ventilator

Dedicated NIV ventilators (5.19 cmH2O.s (3.82; 9.05)) induced a lower PTPt compared with transport (11.1 cmH2O.s (4.75; 25.5); p<0.01) and ICU ventilators (8.8 cmH2O.s (3.78; 23.8); p<0.01) (table 1).

We also found significant differences in the performance of ventilatory parameters between ventilators based on the underlying lung profile (online supplemental e-Tables 5–7).

Discussion

Our study showed significant differences in patient-ventilator synchrony and technical performances between the different ventilators, according to underlying lung profiles. For all ventilators, regardless of the underlying lung profile, the AI appeared to be correlated with the level of leakage and PSV. Dedicated NIV ventilators generated a lower AI compared with ICU and transport ventilators. These differences in AI between ventilators were more pronounced with a restrictive ventilatory pattern (online supplemental e-Tables 6, 8 and 9), and to a lesser extent with a normal pattern (online supplemental e-Tables 7, 10 and 11); whereas, the difference between ventilators was not significant for the obstructive pattern (online supplemental e-Tables 5, 12 and 13). The most frequently observed forms of PVA were auto-triggering and cycling asynchrony. Except for peak flow and minute ventilation, other parameters of ventilatory mechanics were different between ventilators, with some heterogeneity between ventilator categories. We also demonstrated that dedicated NIV ventilators generated less WOBt than ICU or transport ventilators.

Despite the development of new generations of ventilators (ICU, dedicated NIV, transport), there is still significant heterogeneity in synchrony and technical performances between the different categories of ventilators, even within the same category, for NIV.15 16 Despite the improvement of NIV algorithms in the most recent ICU ventilators, they still appear to be less efficient than dedicated NIV ventilators and generate more PVA regardless of the underlying ventilatory profile. This was already observed with the first ‘NIV modes’ integrated in ICU ventilators.15 This lower performance was confirmed in the bench and clinical study by Carteaux et al,16 in which ICU ventilators, despite their activated NIV algorithm, had a significantly higher AI than dedicated NIV ventilators. In the bench part of this study, transport ventilators also generated more auto-triggering PVA than dedicated NIV ventilators. We did not find such a significant difference between the transport ventilator tested in our study and other ventilators. This could be explained by the difference in the generation of transport ventilators tested, the technology of the bench test used but also the lack of power of our study evaluating only one latest generation of transport ventilator as compared with the latter study.16 Furthermore, transport ventilators have been improved technologically, approaching dedicated NIV ventilator performances.26 29 30

This heterogeneity in ventilator performances in terms of AI and PVA was found between ventilators within the same category.15 Among ICU ventilators in our study, the Servo-U appeared to have the best performance, which was interesting because it is a compressed-gas ventilator and performed better than the Monnal-T75 (turbine-driven), and among dedicated NIV ventilators, the Trilogy-Evo performed better than the Astral-150. The AI and PVA rates should also be interpreted in relation to the level of leakage. The AI was significantly different between the different ventilators only for large, simulated leaks of ≥30 L/min. The lack of difference in AI between ventilators for a leak of 15 L/min may suggest, therefore, that other determinants could be involved in the genesis of PVA, such as ventilator settings.

As previously demonstrated in bench11 15 16 31 and clinical studies,6 10 16 the integration and activation of ‘NIV modes’ in ICU ventilators were not sufficient to render ICU ventilators as efficient as dedicated NIV ventilators, even with turbine-driven technology. Our bench study, using a realistic 3D manikin, confirms this feature for the latest generations of ventilators (ICU, dedicated NIV, transport). We chose not to deactivate ‘NIV mode’ in our study as we systematically use ICU ventilators with their NIV algorithm activated.6 16 Unlike previous bench studies, we also evaluated ventilator performances according to different underlying ventilatory profiles.11 15 16 31 With normal or obstructive ventilatory patterns, ventilators generated little or no PVA in our study. In contrast, with the restrictive pattern, ventilators generated more PVA with a large heterogeneity of AI between them. In addition, the AI was found to be not significantly different between ventilators for the obstructive pattern. This feature could partly be explained by the ventilator settings applied. Although we used the same settings for the different lung profiles, these settings could have been more suitable for an obstructive ventilatory pattern. Indeed, these results could be related to the underlying ventilatory profile, and we hypothesised that low compliance (restrictive profile) may be more deleterious than high resistance (obstructive profile) in terms of PVA risk. According to our results, the three ICU ventilators, Evita-XL, V500 and V800, were less suitable for restrictive profiles.

Auto-triggering and cycling abnormalities were the most frequently observed PVA in our study. Auto-triggering was also reported as the most frequent PVA in the bench study of Carteaux et al.16 In clinical studies, cycling asynchrony,10 32 ineffective efforts or even double-triggering6 were the most frequently reported PVA. These differences should take into account the evaluation conditions, that is, test bench vs clinical conditions, ventilator settings, and technological evolution over time. Auto-triggering and cycling abnormalities are closely related to inspiratory trigger sensitivity and expiratory trigger cycling setting, respectively.8 32 In our study, the inspiratory trigger was set to be very sensitive. This could have led to an increase in auto-triggering, but we did not observe any excessive auto-triggering under the various measurement conditions. We used a fixed expiratory cycling trigger without considering the underlying ventilatory pattern. Cycling abnormalities which can increase the patient’s WOBt can also be increased by inspiratory or expiratory non-intentional air leakage between mask and patient’s face.8 32 33

Differences of ventilatory performances between ventilators could be an explanation for the difference of AI. Inspiratory triggering delay, pressure rise time and expiratory cycling seem to be the main determinants of patient-ventilator synchrony during acute13 31 and chronic NIV.28 In addition to the choice of ventilator used for NIV, this highlights the importance of clinician-set settings. The measured pressure at the mask, reflecting the applied pressure, was lower with ICU ventilators than with dedicated NIV ventilators in our study; whereas, minute ventilation was comparable, suggesting similar ventilatory efficiency between ventilators for CO2 removal. Interestingly, Garnier et al34 also reported difficulties of ICU ventilators to reach the targeted pressurisation level. The PTPt, indirectly reflecting the WOBt,35 was lower with dedicated NIV ventilators than with ICU ventilators. This physiological benefit was previously demonstrated with dedicated NIV ventilators.13 16 31 Concerning the transport ventilator in our study, its technical performance in limiting PVA and improving mechanical ventilatory conditions was intermediate between dedicated NIV and ICU ventilators. Such results were reported previously with older transport ventilators.20 32

Apart from PSV level and to limit confounding factors, we applied standardised settings that could be used across the different ventilators tested, but these settings were not adapted to ventilatory conditions (air leakage, underlying lung profile particularly). Finally, our results highlight the need to individualise NIV settings, including inspiratory trigger,36 37 pressure rise time,38 expiratory cycling8 32 and PEEP,23 to each clinical situation to ensure the best compromise between the effectiveness of NIV according to underlying respiratory disease and its tolerance by patients.10 39

Our study exhibits some limitations. First, it is an experimental physiological study carried out in a bench test environment, and its results are probably not applicable to different clinical situations. Indeed, although we used a rather realistic 3D manikin head, only one manikin morphology and one type of facemask were studied, without being subjected, as in clinical practice, to the risk of interface mobilisation likely to generate or aggravate PVA. We used only a dual-limb circuit and did not test dedicated NIV ventilators with a single-limb circuit and a vented mask or an expiratory valve, even though they could be used to administrate NIV with fewer PVAs.16 As NIV practices could vary widely in terms of type of interfaces, circuits and ventilators used,3 it is obviously not possible to conduct a bench study encompassing all clinical situations, and therefore, to extrapolate our results to all clinical situations. In addition to our study, high-performance modern turbine-driven ventilators have also been tested in bench studies using different circuit configurations as well as different interfaces, including the helmet.17

Moreover, although we simulated different lung profiles, these simulations cannot represent all the ventilatory characteristics of patients treated with NIV in ICU or respiratory departments. On the other hand, we did not evaluate different levels of simulated inspiratory or expiratory leakage, which are frequently observed in clinical situations. Despite all these limitations, our results should still help clinicians selecting ventilator models and settings for NIV use, depending on the profile of the underlying respiratory pathology.

Conclusion

Our experimental bench study provides recent data on the technical performances of new generations of ventilators for NIV in ICU or respiratory wards. Despite the integration of NIV algorithms and their technical evolution, most recent ICU ventilators still appear to be less reliable than dedicated NIV ventilators. These technical performances were also heterogeneous within the same category of ventilators and could change according to the underlying respiratory disease, the level of air leakage as well as the integrated technology (compressed-gas vs driven-turbine ventilators). Clinical studies under similar ventilatory conditions are required to evaluate whether our experimental results can be extrapolated to daily practice.