Tidal Volume Estimation during Helmet Noninvasive Ventilation: an Experimental Feasibility Study

We performed a bench (BS) and human (HS) study to test the hypothesis that estimation of tidal volume (VT) during noninvasive helmet pressure support ventilation (nHPSV) would be possible using a turbine driven ventilator (TDV) coupled with an intentional leak single-limb vented circuit. During the BS a mannequin was connected to a lung simulator (LS) and at different conditions of respiratory mechanics, positive end expiratory pressure (PEEP) levels and leaks (30, 50 and 80 L/min). All differences were within the 95% limits of agreement (LoA) in all conditions in the Bland-Altman plot. The overall bias (difference between VT measured by TDV and LS) was 35 ml (95% LoA 10 to 57 ml), 15 ml (95% LoA −40 to 70 ml), 141 ml (95% LoA 109 to 173 ml) in the normal, restrictive and obstructive conditions. The bias at different leaks flow in normal condition was 29 ml (95% LoA 19 to 38 ml). In the HS four healthy volunteers using nHPSV had a pneumotachograph (P) inserted through a mouthpiece to measure subject’s VT.The bias showed a scarce clinical relevance. In conclusions, VT estimation seems to be feasible and accurate in all conditions but the obstructive one. Additional leaks seem not to affect VT reliability.

Noninvasive ventilation (NIV) is widely used worldwide to support ventilation and oxygenation in patients with respiratory failure 1 . The helmet is an interface that may increase comfort, reduce skin breakdown, and increase success rate during noninvasive helmet pressure support ventilation (nHPSV) compared to face mask [1][2][3][4][5][6][7] .
Measurement of tidal volume (V T ) still remains an unresolved issue during nHPSV due to the mechanical properties of the helmet 4,6,[8][9][10] . The availability to assess V T during nHPSV might be of importance because it was previously found that high V T during noninvasive ventilation (NIV) in hypoxemic patients may increase lung injury [11][12][13] . Nevertheless, no study has shown the possibility to assess V T during nHPSV.
Intentional leak single-limb vented circuit configuration allows during face mask or nasal ventilation to make an estimation of the delivered V T in all bi-level ventilators as well as in some intensive care unit (ICU) ventilators mainly dedicated for NIV use [14][15][16] .
We hypothesized that a turbine driven ventilator (TDV) coupled with intentional leak single-limb vented circuit and whose leak location is set at the helmet expiratory port would allow to estimate patient's V T during nHPSV 15,16 . The aim of the present study was to test this hypothesis on a bench and healthy volunteer study. condition). Its functioning is based on a direct-drive screw-driven piston, which moves inside a cylinder according to the equation of motion of an active respiratory mechanics system 17 . Its settings during the bench study, using a single-compartment model, was as follows: (1) Normal condition: resistance 4 cmH 2 O/L/s, compliance 60 mL/cmH 2 O, inspiratory muscle pressure (Pmus) -5 cmH 2 O (semisinusoidal waveform with rise time of 25%, inspiratory hold of 5%, release time of 25%) and LS respiratory rate set at 15 breaths/min; (2) Obstructive condition: resistance 15 cmH 2 O/L/s, compliance 80 mL/cmH 2 O, Pmus -12 cmH 2 O (semisinusoidal waveform with rise time of 20%, inspiratory hold of 5%, release time of 30%) and LS respiratory rate set at 25 breaths/min; (2) Restrictive condition: resistance 7.5 mH 2 O/L/s, compliance 30 mL/cmH 2 O, Pmus -12 cmH2O (semisinusoidal waveform with rise time of 25%, inspiratory hold of 5%, and release time of 25%), and LS respiratory rate set at 30 breaths/min.
The mannequin head was connected to the LS through the mannequin's trachea after having positioned and secured a helmet (Castar, Next Intesurgical, size small, Mirandola, Italy) 18 . The helmet inspiratory port was connected to the TDV (Bellavista 100 ICU, Buches, Switzerland) via a single-limb circuit while the expiratory port was closed with a cap having a 5.5 mm hole to allow the intentional leak. This configuration was not provided with any type of inspiratory or expiratory valves and carbon dioxide flushes, as a function of mask pressure through an intentional leak port set in the circuit or in the interface itself. Neither inspiratory nor expiratory V T is measured but they are estimated taking into account the leak according to a mathematical leak model as a function of mask pressure. This mathematical model is used to make an estimation the instantaneous leak and flow which are subtracted from the total flow output of the ventilator (which is the only measured flow) to calculate V T 19 . Figure 1 shows the set-up. During nHPSV, the ventilator was set at the fastest pressure rise time, cycling-off flow threshold of 25% in the normal and restrictive condition and 40% in the obstructive one. Inspiratory flow trigger initially set at 2 L/min was then always tuned at the lowest value not determining auto-triggering 20 . Inspiratory pressure above positive end expiratory pressure (PEEP) was set to reach a V T of about 300/500 ml in all conditions and did not change throughout each condition. Each condition was simulated at PEEP of 5, 8, 10 cmH 2 O. PEEP 12 cmH 2 O was only used in the restrictive condition. During nHPSV leaks were avoided by accurately fit the helmet to the mannequin's neck. During normal condition at PEEP 8 cmH 2 O and inspiratory pressure (above PEEP) 0 cmH 2 O we also tested the reliability of V T estimation at different leak flows (30,50 and 80 L/min) using a calibrated hole. During normal condition at PEEP 8 cmH 2 O we also tested the reliability of V T estimation at different leak flows (30,50 and 80 L/min) using a calibrated hole.
Healthy volunteers study. After approval from our Institutional Ethics Committee (Comitato Etico Palermo 1 -approval N° 07/2018) and after obtaining written informed consent, four healthy volunteers, 1 female, 3 male, mean age 33 ± 4 years, mean body weight 70 ± 15 kg were ventilated in nHPSV (PSV 8 cmH 2 O and PEEP 8 cmH 2 O) via a helmet (Castar, Next Intesurgical, size small or medium, Mirandola, Italy), using the same TDV configuration used in the bench study. all experiments were performed in accordance with relevant guidelines and regulations. A mouthpiece was inserted in the volunteer's mouth and connected to pneumotachograph (P) (V T mobile FLUKE, Germany) to measure subject's airflow and V T . A nose clip was positioned on volunteer's nostrils to avoid leaks from the nose. The set-up is showed in Fig. 1. All the healthy volunteers were trained, using a metronome, to maintain an imposed respiratory rate of about 12/15 breaths per minute 20 . During nHPSV, ventilator was set at the fastest pressure rise time and at cycling-off flow threshold of 25%. Inspiratory flow trigger was set at the lowest value not determining autotriggering starting from 2 L/min. During nHPSV, leaks were accurately avoided choosing the right helmet size and by appropriately fitting the helmet to the volunteer's neck.
Measurements. Bench study. Data were collected by the TDV and by LS software. Airflow (V.), the airway pressure (Paw), PEEP as well as the V T and respiratory rate delivered to mannequin were collected over 3 minutes. Differences in V T between TDV and LS were compared during the last 20 breaths of each trial to ensure the stability of the system across settings modification (e.g. increasing, mechanical pattern and pressure level) 16 . During the test at different leak flows, each measurement was performed after 1 minutes after the institution of the new leak to let the TDV to adapt to the new setting. Autotriggering was determined as a mechanical insufflation in absence of inspiratory effort on the LS 17 .
Healthy volunteers. Each experimental condition was maintained for 3 minutes, to ensure the stability of the system across the modification of settings (e.g. increasing leak, respiratory pattern) 16 . Paw, PEEP obtained from the TDV and V T and respiratory rate of both TDV software and pneumotacograph (P) were collected. V T measured by TDV and P were compared during the last 20 breaths of each trial.
Statistical analysis. Normality of data distribution was checked graphically and by D' Agostino-Pearson test. Data are expressed as mean ± standard deviation (SD) or as median and interquartile range (IQR) when appropriate. Bland-Altman graphs were used to plot differences between the V T from the TDV and LS against the average of the two measurements.
Bland-Altman plot is the standard graphical statistical method to evaluate the agreement between quantitative measurements from two different devices, assays or raters. Bland-Altman plots reported the bias (defined as the average of the differences between measurements) and 95% limit of agreements (LoA), namely ± 1.96 SD of the bias 21 . We assumed as clinically relevant a difference of >15% of the LS.
Differences between V T (TDV-LS) at different PEEP levels or leaks were compared using ANOVA for repeated measures and Student-Newman-Keus test was used for all pairwise comparisons. Paired sampled t-test was used to compare V T measured by V and P in the healthy volunteers. P values < 0.05 were considered statistically significant. We used Prism 7 (GraphPad software; San Diego, CA) and Microsoft Excel (version 2013; Microsoft corporation, Redmond, CA).

Results
Bench study. Normal, restrictive, obstructive condition. The level of nHPSV to reach the default V T was 8, 10 and 12 cmH 2 O respectively in the normal, restrictive and obstructive condition.
Differences in V T between TDV and LS at different levels of PEEP and pairwise comparisons are shown in Table 1. All plotted differences were within the 95% LoA in all conditions. The overall bias was 35 ml (95% LoA 10 to 57 ml), 15 ml (95% LoA -40 to 70 ml), 141 ml, (95% LoA 109 to 173 ml) respectively in the normal restrictive and obstructive conditions (Figs. 2-4). In the normal condition, the bias reached the least clinical relevance at PEEP 10 cmH 2 O (Fig. 2) while in the restrictive one at PEEP 10 and 12 cmH 2 O (Fig. 3). In the obstructive condition, the bias increased at PEEP 5 cmH 2 O (Fig. 4). Autotriggering was never detected. Tables S1-S3 in the supplementary file 1 reported average V T at different PEEP levels in the different simulated conditions. Leaks simulation. Data on the difference in V T between TDV and LS at different levels of leak flow are shown in Table 2. The only significant differences were found between 80 L/min vs. 30 L/min and 80 L/min vs. 50 L/min. The overall bias was 29 ml (95% LoA 19 to 38 ml) (Fig. 5). Table S4 in the supplementary file 1 reported average V T at different levels of leak flow.  www.nature.com/scientificreports www.nature.com/scientificreports/ Autotriggering was never detected in any experimental records but at the beginning of the new leak condition before the complete TDV compensation (e.g. switching leak from 30 to 50 L/min). Table 3 shows V T measured by the TDV and the P for the four health volunteers. Mean difference between V T measured by TDV and P was 45 ml (±32) for volunteer 1, 11 ml (±38) for volunteer 2, -35 ml (±37) for volunteer 3 and 58 ml (±40) for volunteer 4. Figure S1 in the supplementary material shows the distribution of V T measured by TDV and P for each volunteer.

Discussion
The main findings of the bench study are: (1) No clinically relevant differences were found in V T between TDV and LS in normal and restrictive conditions at all simulated level of PEEP; (2) The difference between TDV and LS remained stable across the different tested leak flows; (3) There was a large, clinically relevant, difference in V T in the obstructive condition between TDV and the LS.
To the best of our knowledge this is the first study able to demonstrate the possibility of estimate V T during helmet ventilation. The same circuit configuration was tested in a previous study, which used nHPSV in single-circuit vented intentional leak configuration to measure carbon dioxide (CO 2 ) rebreathing 15 . Interestingly, although V T was not measured by the ventilator software in this study, this kind of circuit configuration should theoretically allow to estimate V T as during non invasive intentional leak mask ventilation 16 . This estimation has been demonstrated to be pretty reliable in different condition of respiratory mechanics and linear non intentional leaks 16 .
Our bench study was carried out with a TDV with an intentional leak vented single-limb circuit because direct measurement of V T was not possible due to the intrinsic mechanical features of the helmet, namely its large   www.nature.com/scientificreports www.nature.com/scientificreports/ internal volume 15 . Besides, a TDV with an intentional leak vented single-limb circuit better compensate for air leaks up to its leak compensation capability 16,22 .
The helmet is an interface that may increase comfort, reduce skin breakdown, and increase success rate during nHPSV compared to face mask [1][2][3][4][5][6][7] . The TDV was coupled to a new commercially available helmet. This new helmet is featured by an annular extendable ring placed under an inflatable large cushion that secures the interface without the need for armpit support 23 . This helmet has been found to reduce to a large extent its upward displacement during the ventilator support thus improving patient-ventilator interaction 23 .
Our results found a very good agreement between differences in V T estimated by the TDV and that measured by the LS in the normal and restrictive condition using the Bland Altman plot. Although the difference between  www.nature.com/scientificreports www.nature.com/scientificreports/ V T measured by the LS and the TDV was statistically different among the groups at the different PEEP levels, this difference was of poor clinical relevance. Of note, Lyazidi et al. found that when setting a V T of 6 ml/kg of predicted body weight, a difference of 1-2 ml/kg with the actually delivered V T would be commonly found 24 .
We found, both in the normal and in restrictive condition, that the higher the PEEP the less the difference between the V T measured by the TDV and the LS. This result is very appealing being the restrictive lung one the most common application of helmet ventilation 4,25 . Conversely, in the obstructive condition we found, although still in the 95% LoA, a clinically relevant discrepancy between the two V T with an overestimation showed by the TDV. Interestingly, as opposite to our findings, Lujan et al. found that built-in software was underestimating the V T in two of the tested ventilators 16 . This may suggest that different built-in software may behave differently in presence of an obstructive pattern. One can also speculate that understimation might be linked to the reassessment of the zero flow value by the TDV in presence of air trapping. However, we can also hypothesize that during the bench study in this condition, the simulated patient's high resistances could have generated small non intentional leaks. While the ventilator software was not affected by the presence of the leak because it was able to compensate it, the monitoring system of the ASL 5000 did so reading smaller V T .
Autotriggering was never detected in any trial. This result supports previous finding of the TDV capability in intentional leak vented configuration to better compensate for leaks when a flow trigger is used 22 . Autotriggering could also not be found due to the use of the new helmet, which presents less upward displacement during nHPSV. Only when simulated a new leak flow, few auto-triggered breaths were detected due to the need of the ventilator to compensate the loss in air flow.
Non intentional leaks are the most frequent drawback during NIV 26 . We evaluated the reliability of V T estimation increasing the values of the intentional leak flows to 30, 50 and 80 L/min during nHPSV in the normal simulated condition at PEEP of 8 cmH 2 O. This range was selected because it includes the minimal value to avoid rebreathing even in absence of the inspiratory support 7 . Again, the difference in V T at different leak flows was clinically negligible.
Software may underestimate both leaks and V T with the greater the leak, the greater the difference between the estimated and actual V T 27 . However, the ventilator software we used was provided with an algorithm that improves accuracy compensating for the difference between the estimated leak measured at the distal and proximal sides of the circuit 16 . The interesting clinical message is that being leaks a possible drawback of helmet ventilation this way of estimating V T could theoretically assures V T estimation even in presence of additional leaks.
The main findings of the healthy volunteer study are that although the differences in V T were significantly different in all the volunteers but the second one, this difference was of scarce clinical relevance. During the human study, we used the same set-up that we previously described in other study 15 . We applied PEEP at 8 cmH2O because it is one of the most common value used in the clinical setting 18,28,29 .
Estimation of V T has been the only mode of measuring V T in TDV in intentional leaks single-limb vented configuration since the introduction of these ventilators at the beginning of the 90 16,27 . The built-in software of these TDV should be able to accurately estimate V T , even in presence of increased non-intentional leaks. Of note, using pressure controlled modes with compressed air driven intensive care unit (ICU) ventilator, the higher the non intentional leak the lower the expiratory V T 22,30 . In addition, by using an helmet, which has a larger internal volume, the displayed V T , when using a non vented configuration with a high pressure driven ICU ventilator is significantly higher than actual delivered V T

.
We can speculate that, using a TDV in this configuration, it would be possible to ventilate hypoxemic patients 4 via an helmet by knowing that delivered V T 11 is pretty closed to what the patients is really breathing in. The present study has several limitations. Firstly, this is a bench and human volunteers study. Therefore, these results need to be confirmed clinically. However, direct comparison of real measurements and its estimation at standardized condition of respiratory mechanics and respiratory rate in the same patient seems impossible for both practical and ethical reasons. Moreover, it should be noted that volunteers had healthy lungs. Second, we measured V T under ambient temperature and pressure and dry gas conditions (ATPD). However, the V T in the patient's lungs is at body temperature and pressure, and saturated with water vapor (BTPS). The conversion from ATPD to BTPS would increase V T by about 10.4% for TDVs at 20 °C and at the vapor pressure of H 2 O 17.5 mmHg 16 . We could speculate that the small gap we found between TDV and LS would be potentially further reduced by converting ATPD in BTPS. Third, we used to simulate an increase in leaks an increase in linear non intentional leak 31 . As previously demonstrated by Sogo et al., the presence of unintentional random leaks may lead to an error in the measurement of V T provided by the ventilator software 31 . In addition, we did not measure the pressurization time at the different simulated leaks. It is possible that by increasing the amount of the leak (intentional or non intentional) we approached the compensating capability of the TDV thus increasing the time necessary to achieve the preset level of pressure support 5 . Fourth, we did not measure CO 2 rebreathing in the Volunteer N°TDV V T (ml) P V T (ml) P value  Table 3. Tidal volumes (V T ) measured by turbine driven ventilator and pneumotacograph in the human study. All measurements were done at PEEP and nHPSV of 8 cmH 2 O. Data are expressed in ml and reported as mean ± SD; PEEP: positive end expiratory pressure; TDV: V T measured by turbine driven ventilator; P: V T measured by pneumotachograph. (2019) 9:17324 | https://doi.org/10.1038/s41598-019-54020-5 www.nature.com/scientificreports www.nature.com/scientificreports/ human study. However, we previously demonstrated that the same setting of the present study improved CO2 rebreathing compared to a non-vented circuit configuration using an air compressed ICU ventilator 15 . Lastly, due to different algorithms used to estimate V T in the build-in software of TDV, we cannot be sure that our data can be extrapolated to other TDVs in vented configuration 16,27 .
In conclusion, this feasibility bench and human study demonstrated that V T estimation during nHPSV seems to be feasible and accurate in all the conditions except the simulated obstructive condition. Additional leaks seem not to affect V T reliability. Further studies are required to identify the magnitude of discrepancy between the proposed method and gold standard in bench and clinical setting.

Data availability
The datasets analysed during the current study are available from the corresponding author on reasonable request.