Assisted Ventilation in Patients with Acute Respiratory Distress Syndrome: Lung-distending Pressure and Patient–Ventilator Interaction

IN patients with acute respiratory distress syndrome (ARDS), mechanical ventilation with positive-end expiratory pressure (PEEP), limited plateau airway pressure (Paw), and low tidal volume (Vt) is widely accepted as the ventilation strategy of choice to limit ventilator-induced lung injury.^1,2  The use of assisted mechanical ventilation in ARDS is a subject of debate.^3–5  Among other beneficial effects, assisted ventilation preserves diaphragm function and better resembles natural respiratory variability when compared with controlled mechanical ventilation.^3,4  This may be important, as increased variation in breathing pattern improves oxygenation and lung mechanics and enhances tidal distribution to the dependent lung regions.^6,7  However, during assisted ventilation, spontaneous breathing contributes to the transpulmonary pressure (Ptp), potentially increasing the risk of ventilator-induced lung injury if either is excessive.⁴ 

Neurally adjusted ventilator assist (NAVA) is a ventilator mode that uses electrical activity of the diaphragm (EAdi) to control timing and level of support during inspiration, thereby introducing inherent feedback loops between patient and ventilator.⁸  It has been shown in patients with acute respiratory failure that increasing levels of support with NAVA unloads the respiratory muscles but limits pressure increases due to down-regulation of EAdi.^9–11  This suggests that intrinsic lung-protective feedback mechanisms may limit Vt and lung-distending pressure. Recently, it has been shown in an animal model that not only the magnitude but also the duration of mechanical stress within the respiratory cycle (inspiratory-to- expiratory time ratio) determines the severity of lung injury.¹²  This emphasizes the importance of synchrony between neural and mechanical inspiration.

Overall, it is rational to investigate the effects of increased patient freedom to control the ventilator on respiratory mechanics in patients with ARDS. Up to now, data related to these effects are scarce in patients with ARDS.^10,13  Therefore, the aim of the current study was to investigate the effects of three modes of assisted ventilation on lung-distending pressures, Vt, breathing variability, and patient–ventilator interaction in patients with ARDS. The selected ventilator modes cover a wide spectrum of patient freedom to control ventilator assist: (1) pressure control ventilation (PCV): a pressure-limited mode of both mandatory and triggered assist of fixed duration; (2) pressure support ventilation (PSV): a pressure-limited mode with flow-trigger and variable cycling-off, allowing the patient more freedom to time assist delivery to inspiratory effort; and (3) NAVA: a mode where the patient regulates both timing and magnitude of assist via neural inspiratory effort. We reasoned that intrinsic lung-protective feedback mechanisms maintain lung- protective ventilation when patient freedom to control the ventilator is increased. Accordingly, the first hypothesis of our study was that with NAVA (maximal “patient freedom”) Vt and lung-distending pressure remain within clinically acceptable limits. The second hypothesis was that breathing pattern variability and patient–ventilator interaction in patients with ARDS improve with increased freedom to control the ventilator.

Twelve adult patients, fulfilling the Berlin definition of ARDS,¹⁴  were studied in a physiological study on the intensive care unit of the Radboud University Medical Center. Exclusion criteria were hemodynamic instability, contraindications to changing a nasogastric tube (i.e., recent nasal bleeding, upper airway/esophageal pathology, or surgery), and previously known neuromuscular disorders.

The protocol was approved by the institutional review board (CMO regio Arnhem-Nijmegen, NL31557.091.10) and is in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. Patient surrogate decision makers gave informed consent before study inclusion.

Patients were ventilated with a Servo-I ventilator (Maquet Critical Care, Sweden). EAdi and esophageal pressure (Pes) were obtained with a multielectrode esophageal catheter with balloon (Neurovent Research Inc., Canada) as described previously.^15,16  Care was taken with correct positioning and inflation of the balloon using the Baydur method.¹⁷  Additional details on catheter positioning are provided in Supplemental Digital Content 1, https://links.lww.com/ALN/B147.

Initially, patients were ventilated in PSV mode for 30 min to verify the feasibility of assisted ventilation. Subsequently, in a cross-over design, PCV, PSV, and NAVA were randomly applied for 30 min each. PSV and PCV levels were matched to target a Vt of 6-ml/kg predicted bodyweight. With PCV, mechanical respiratory rate was set just below patient’s respiratory rate to allow patient-triggered breaths. NAVA level (cm H₂O/μV) was set by the use of a dedicated window on the Servo-I ventilator. In brief, in either PSV or PCV, the actual Paw is presented with an overlay of the predicted Paw (as if the patients were in the NAVA mode). The NAVA level was adjusted manually to try to match the predicted peak pressure to the actual peak Paw. PEEP was maintained constant throughout the study period and set according to the higher PEEP/lower inspired oxygen fraction (Fio₂) arm of the ARDSnet consensus.¹⁸  Additional details on the ventilator settings are provided in Supplemental Digital Content 1, https://links.lww.com/ALN/B147. At the end of each mode, arterial blood was sampled for analysis. Drugs, including sedatives and fluids, were unchanged during the entire study period.

Flow, Paw, and EAdi were acquired (sampling rate, 100 Hz) from the serial port of the Servo-I and resampled to 2 kHz. Pes was acquired (sampling rate, 2 kHz) by connecting the balloon to a pressure transducer (range ±375 mmHg; Freescale, USA) and A/D converter (DT3004; Data Translation, USA). All signals were acquired synchronously using a dedicated software (Neurovent Research Inc.).

Data were analyzed offline. All variables were calculated from a stable 5-min period at the end of each mode on a breath-by-breath basis using a software routine developed for MATLAB (MathWorks, USA). For each breath, we defined the following variables during the inspiratory phase (as shown in fig. 1):

• Vt, as the integral of the inspiratory flow over time.

• Peak EAdi.

• Mean, peak, and end-inflation airway pressure (Paw_mean, Paw_peak, and Paw_{end inflation}).

• Mean, peak inspiratory deflection, and end-inflation Pes, as change from baseline (ΔPes_mean, ΔPes_peak, and ΔPes_{end inflation}). Baseline was defined as Pes at the start of neural inspiration.

• Mean, peak, and end-inflation transpulmonary pressure, as change from baseline (ΔPtp_peak, ΔPtp_mean, and ΔPtp_{end inflation}). Transpulmonary pressure was calculated as the difference between Paw and Pes. Baseline was defined as transpulmonary pressure at the start of neural inspiration.

• Transpulmonary pressure–time product (PTP_tp) was calculated as the integral of transpulmonary pressure over time. PTP_tp is a measure of duration of mechanical stress within the respiratory cycle. To determine the effect of patient–ventilator dyssynchrony, this variable was also calculated during the following phases of inspiration: (1) trigger delay; (2) synchronous overlap period between neural and mechanical inspiration; and (3) cycle-off error.

Neural respiratory rate was calculated as the number of EAdi peaks per minute and mechanical respiratory rate as the number of Paw peaks per minute. Breath-by-breath variability was assessed by calculating the coefficient of variation (CV; SD/mean × 100) for Vt and EAdi.

Patient–ventilator interaction was evaluated by comparing Paw and EAdi waveforms with a recently validated automated computer algorithm that reports the timing error between Paw and EAdi.^19,20 

In brief, automatic detection of the start of neural inspiration (EAdi_ON) was obtained by detecting a 0.5-μV increase in EAdi. The end of neural inspiration (EAdi_OFF) was automatically detected by finding when the EAdi had decreased to 70% of its peak. The onset of pressure support (P_ON) was automatically detected by searching for an increase in Paw of greater than 3 cm H₂O. The termination of pressure support (P_OFF) was automatically detected by searching for the decrease in Paw (see fig. 1 for an example of these timings).

Electrical activity of the diaphragm and Paw timings were used to calculate the NeuroSync index as following. The trigger error (P_ON − EAdi_ON) and cycle-off error (P_OFF − EAdi_OFF) were calculated as a percentage of the neural inspiratory detection period and neural expiratory detection period, respectively. Neural inspiratory detection periods were defined as segments from one detected EAdi_OFF to the next EAdi_OFF. Neural expiratory detection periods were defined as segments from one EAdi_ON to the next EAdi_ON. Thus, an early trigger error could range between −100 and 0% and a late trigger error could range between 0 and 100%. In the same manner, an early cycle-off error could range between −100 and 0% and a late cycle-off error could range between 0 and 100%. Breaths with an absolute error of more than 33% were defined as dyssynchronous breaths and breaths with an absolute error of less than 33% as synchronous breaths. Asynchronous breaths, defined as a complete dissociation between EAdi and Paw (wasted efforts, autotriggering, and double triggering), were assigned 100% error. The NeuroSync index was then calculated by averaging the errors for all breaths per patient per mode.

To compare modes, one-way ANOVA for repeated measures was performed or the Friedman test as its nonparametric equivalent. Post hoc analysis was performed with the Student–Newman–Keuls test or Dunn test, as its nonparametric equivalent, to correct for multiple comparisons. The effect of patient–ventilator dyssynchrony on PTP_tp was analyzed using two-way ANOVA for repeated measures with Bonferroni posttest. Linear regression analysis was performed to test associations. For all tests, a two-tailed P value less than 0.05 was considered significant. Data are described as mean ± standard error for parametric data or median (interquartile range) for nonparametric data. Statistical analyses were performed with Prism 5 (GraphPad Software, USA).

Table 1 reports patient characteristics and table 2 reports ventilator settings for PCV, PSV, and NAVA during the study.

Tidal volumes were equal between modes (fig. 2A). One patient having lowest pH of 7.18 revealed excessive Vt both in PSV and NAVA (patient 4, fig. 2A) but not in PCV. There was no difference between modes for neural and ventilator respiratory rates; however, respiratory rates varied widely between subjects (fig. 2B). The coefficient of determination (r²) between neural and ventilator respiratory rates was high with NAVA (0.99, P < 0.0001), 0.70 (P = 0.0007) with PCV, and 0.61 (P = 0.0026) with PSV. Minute ventilation was equal between modes (table 3). Peak EAdi ranged from near-zero to 30 μV between patients (fig. 2C). Post hoc analysis did not reveal any significant differences between modes for EAdi. Mean Paw (Paw_mean) during inspiration was lower with NAVA compared with both PCV and PSV (fig. 2D). Changes in pleural pressure during inspiration are expressed by ΔPes_mean and ranged from positive to high negative values. Group mean values for ΔPes_mean were higher during PCV, that is, less negative swing, compared with PSV and NAVA (fig. 2E). Lung-distending pressure, expressed by ΔPtp_mean, was highest during PSV, lower during PCV, and lowest during NAVA (fig. 2F). Peak and end-inflation ΔPes and ΔPtp during inspiration were slightly lower during PCV compared with PSV and NAVA, whereas these values were equal for Paw (table 3).

The CV of EAdi was equal between modes, whereas the CV of Vt was higher with NAVA (table 3). Linear regression analyses between CVs of EAdi and Vt are presented in figure 3, which shows no relation with PCV, a poor relation with PSV, and a good relation with NAVA.

There was a significant difference in PTP_tp between modes (one-way ANOVA, P = 0.0018). Post hoc analysis showed that PTP_tp was significantly lower (P < 0.05) with NAVA (6.9 ± 0.8 cm H₂O s) compared with PSV (8.3 ± 0.7 cm H₂O s) and PCV (9.5 ± 0.8 cm H₂O s).

The percentage of synchronous breaths were higher and trigger delay shorter with NAVA compared with PCV and PSV (table 4). Cycle-off error was higher with PCV compared with PSV and NAVA (table 4). Wasted efforts and double triggering were overall uncommon and equal between modes, whereas the rate of autotriggering was different between modes (table 4). The latter is explained by the fact that during PCV, 87.6 ± 16.8% of the breaths were patient triggered and 12.4 ± 16.8% were ventilator controlled. The NeuroSync index, depicted in figure 4A (0% error = perfect patient–ventilator interaction and 100% error = zero patient–ventilator interaction), indicated more than 20% error in seven patients during PCV, which improved during PSV and was less than 20% in all patients during NAVA. The higher cycle-off error with PCV resulted in an increased PTP_tp, as depicted in figure 4B.

Table 5 shows arterial blood gas values at the end of each 30-min interval. Pao₂ and Paco₂ were significantly higher with NAVA compared with PCV (table 5).

The current study reports the effects of different modes of assisted ventilation on lung-distending pressure, Vt, breathing pattern variability, and patient–ventilator interaction in patients with ARDS. The three selected modes allow for different degrees of patient freedom to control the ventilator. The main findings of this study were that with increasing freedom to control the ventilator (1) Vt remains within the limits of lung-protective ventilation; and (2) lung-distending pressures remain similar between modes. Even with NAVA, when the patient has a high degree of freedom to control the amount of support, Vt and lung-distending pressures were similar to assist control ventilation, whereas breathing pattern variability and patient–ventilator interaction improved compared with PCV and PSV. Furthermore, prolonged inspiratory time during PCV resulted in a longer duration of mechanical stress.

In the current study, Vt could be considered lung protective in the majority of patients and was not affected by ventilator mode. Peak and end-inflation Paws (table 3) were equal between modes, whereas mean Paws were lower during NAVA than during PCV and PSV (fig. 2D). Lower mean ventilator pressures with NAVA could be explained by the slower rate of pressurization with NAVA compared with a squared pressure waveform with PSV and PCV. These lower mean Paws do not necessarily reflect a too low level of assistance during NAVA. The latter is supported by the similar levels of EAdi with the three modes (fig. 2C), and the similar level of mean Pes compared with PSV. The lower level of mean Pes during PCV mainly reflects a too long inspiratory time during PCV. Although there were statistical differences in ΔPtp_mean, ΔPtp_{end inflation}, and ΔPtp_peak between modes, these were small in absolute values. Based on a post hoc analysis of the low Vt ARDSnet study,¹  Hager et al.²¹  concluded that no safe plateau pressure exists. It can be derived from the first figure in their article that reduction in plateau pressure from 30 to 25 cm H₂O is associated with reduction in mortality of approximately 4%. In our study, differences in airway and transpulmonary pressures among modes were 3 cm H₂O or less and therefore probably of limited clinical significance.

Protection against excessive Ptp and Vt is likely mediated by vagal afferents sensitive to lung distension.^22,23  Indeed, our data demonstrate similar Vt and lung-distending pressures during PCV, PSV (both delivering fixed pressure), and NAVA (proportional control of pressure). Prevention of high Vt and Ptp may be mediated by the Hering-Breuer inspiratory-inhibiting (vagal) reflex, where lung inflation at critical volume (i.e., lung stretch) terminates inspiration.²⁴  However, in the current study in one patient, a Vt greater than 9 ml/kg was observed during PSV and NAVA. This patient had the lowest blood pH (7.18), consistent with the view that acid–base homeostasis is the primary goal of the respiratory centers in the brain stem and may be achieved at the expense of high Vt or Ptp.²⁵  On the basis of this observation, we suggest to be very cautious with the use of assisted breathing modes in patients with a very low pH.

Preservation of natural breathing variability and complexity during mechanical ventilation is considered important, as it may improve oxygenation, lung mechanics, and gas distribution to the more dependent regions.^6,7  Breath-by-breath variability of Vt with NAVA was higher in comparison with PCV and PSV, whereas EAdi variability was similar. Accordingly, the ventilator output responded to the patient’s neural breathing effort during NAVA and not during pressure-targeted assist (fig. 3), which is in accordance with previous findings.^9,26  By definition, a larger respiratory variability with NAVA increases the incidence of breaths with a high Vt.⁹  It is unknown whether this periodic delivery of breaths with a higher Vt during assisted mechanical ventilation is harmful. It could also be regarded as a type of sigh or recruitment maneuver, which have been shown to have moderate success for improving lung mechanics and gas exchange.^27–29  In the current study, oxygenation improved slightly with NAVA compared with PCV, but not compared with PSV. This might be explained by better lung recruitment of the dependent regions with NAVA.⁶  Paco₂ was slightly lower during PCV compared with NAVA. This might have resulted from small differences in Vt, respiratory rate, and Vt distribution, nevertheless blood pH was unaffected.

With NAVA patient–ventilator interaction was with minimal error, whereas prevalence of patients with poor patient–ventilator interaction increased during PSV and PCV. This is consistent with recent findings in patients with ARDS^10,13  and in other patients with acute respiratory failure.^30,31  This is of potential clinical relevance, as patient–ventilator asynchrony has been associated with adverse clinical outcome, including prolonged mechanical ventilation.^32,33  For example, reversed triggering or breath stacking in assist-control mode may contributing to ventilator-induced lung injury^34,35  (see figure, Supplemental Digital Content 1, https://links.lww.com/ALN/B147, for an example of breath stacking caused by reversed triggering in the current study). Experimental animal data suggest that not only the magnitude but also the inspiratory cycle duration of lung- distending pressure during PCV is associated with lung injury of increased severity.^12,36  In the current study, late cycling-off resulted in prolonged and unnecessary lung-distending pressures with PCV and to a lesser extent with PSV (fig. 4B). Thus, our results support the notion that poor patient–ventilator interaction and fixed duration of assist during spontaneous breathing is less lung protective than synchronized assist delivery in proportion to neural inspiratory effort.

The role of assisted ventilation in ARDS has been extensively debated in recent years.⁴  The most striking evidence in opposition to assisted ventilation is the effect of muscle paralysis in the first 48 h in patients with ARDS with Pao₂/Fio₂ less than 150 mmHg, which has been associated with improved outcome.³⁷  Our findings may have implications for mechanical ventilation in mild-to-moderate ARDS after these first 48 h. First, our results suggest that it is reasonable from a physiological point of view to use an assisted ventilation mode with good patient–ventilator interaction without increasing the risk of excessive Vt and lung-distending pressure. Our data are supported by recent findings in a pig model of ARDS that higher levels of spontaneous breathing reduce global stress and strain.³⁸  Second, in terms of the patient’s inspiratory drive and effort, the current study revealed a high variability in EAdi and Pes values, ranging from overassist to strenuous efforts (fig. 1). Also, a great dispersion in neural respiratory rate implies a large variety in respiratory drive in patients with ARDS. These findings suggest that adjusting ventilator assist to target low Vts resulted in different levels of neural drive in each study patient, probably due to a complex interaction between sedation, ventilatory mode, and level of assist. Accordingly, monitoring Vt in conjunction with EAdi and/or Pes can aid to optimize ventilation in ARDS to prevent overassist or insufficient assist and improve patient–ventilator interaction.³⁹  The capabilities of EAdi for monitoring purposes were recently well demonstrated in a trial to monitor the dynamic intrinsic PEEP.⁴⁰ 

Several limitations of the current study should be addressed. First, Pes was used as an estimate of pleural pressure. The validity of Pes for this purpose has been discussed before in the literature.^41,42  Second, the real lung-distending pressure is transpulmonary pressure obtained under static conditions (zero-flow) to eliminate the resistive component in the equation of motion. Achieving and verifying static conditions in spontaneous breathing under assisted ventilation in critically ill patients is very cumbersome. Therefore, our measurements of lung-distending pressures were performed during dynamic conditions and cannot directly be compared with static transpulmonary pressures. As peak pressure is highly influenced by airway resistance and is not a reliable estimate of lung-distending pressure, mean transpulmonary pressure during inspiration may be a better estimate under dynamic condition. Last, it should be recognized that our study is a single-center, physiological study, conducted in a heterogeneous group of patients with ARDS with a wide range of sedation scores. Different conditions leading to ARDS, such as pulmonary versus extrapulmonary, might influence the effect on lung-protective ventilation while changing ventilator modes, when studied in large subgroups. In our study, each mode was studied for a period of 30 min. This allows for physiological comparisons and improves our understanding in the effects of these different modes on respiratory mechanics, patient–ventilator interaction, and gas exchange. Large clinical trials that evaluate these different ventilator modes for several days in large groups of patients are needed to determine whether one mode is superior in outcome compared with the other modes.

In conclusion, in patients with mild-to-moderate ARDS, increasing patient freedom to control the ventilator maintains lung-protective ventilation in terms of Vt and lung-distending pressure but improves patient–ventilator interaction and preserves respiratory variability.

The authors thank Norman Comtois from the Department of Critical Care Medicine, St. Michael’s Hospital, Toronto, Ontario, Canada, for his technical assistance.

This study was investigator initiated and financed by institutional resources of the Radboud University Medical Center, Nijmegen, The Netherlands. Dr. Sinderby was funded in part by the R. S. McLaughlin Foundation, Toronto, Ontario, Canada.

Drs. Sinderby and Beck have made inventions related to neural control of mechanical ventilation that are patented. The license for these patents belongs to Maquet Critical Care, Sölna, Sweden. Future commercial uses of this technology may provide financial benefits to Drs. Sinderby and Beck through royalties. They each own 50% of Neurovent Research Inc. (NVR, Toronto, Ontario, Canada). NVR is a research and development company that builds the equipment and catheters for research studies. NVR has a consulting agreement with Maquet Critical Care. Drs. Sinderby and Beck are married. St. Michael’s Hospital has a research agreement with Maquet Critical Care AB and receives royalty and overhead from this agreement. The other authors declare no competing interests.