1 Introduction

In normal subjects, the respiratory centre in the brainstem controls the respiratory pattern namely inspiratory time and expiratory time, inspiratory flow, hence tidal volume (Vt). Action potentials generated in the brainstem represent the patients’ inspiratory demand. From the brainstem, these action potentials reach the diaphragm, through phrenic nerves then initiating diaphragmatic depolarization and contraction resulting in a flow and pressure variation in the airways which is the beginning of inspiration.

Mechanical ventilation is a widely used therapy in intensive care units (ICUs) for acute respiratory failure [1]. Partial assist ventilation modes that preserve some of the patient’s spontaneous respiratory activity are widely used [2, 3]. The most used partial assist ventilation mode is the pressure support ventilation (PS) [2, 3]. In PS, each ventilator-delivered cycle is initiated by a pneumatic signal (flow or pressure), detected at the upper airway as produced by the patient’s inspiratory effort. The pressure delivered is set by the clinician, and the transition of the (assisted) breath into expiration occurs when the inspiratory flow decreases to a predetermined level [4].

Thus, where normal spontaneous breathing patterns show high variability, respiratory variability is reduced under PS. In PS, a constant pressure is delivered by the ventilator regardless of the patient’s relative inspiratory effort. As this constant pressure produces the majority of resulting tidal volume (Vt) supply, PS is expected to result in reduced Vt variability.

Neurally adjusted ventilatory assist (NAVA) is an assisted mode that uses electrical activity of the diaphragm (Eadi), an expression of the patient’s inspiratory demand, to trigger and cycle off the ventilator, as well as to deliver pressure in direct proportion to Eadi [5]. Compared to PS, NAVA improves patient-ventilator interaction by reducing trigger delay, improving expiratory cycling and reducing the number of asynchrony events [6–8]. NAVA also increases respiratory variability in Vt and flow related variables compared to PS [6, 9]. However, no studies make these comparisons relative to the inspiratory demand (Eadi), which may vary between modes and patients. Thus, to determine the true impact of NAVA on Vt variability, Eadi must be accounted for in the analysis.

It is hypothesized that the magnitude of Vt under NAVA would show both better correlation with magnitude of Eadi (A better matching between Vt with Eadi) and greater variability, than with PS. This study aims to confirm these hypotheses by analysis of flow-time and Eadi-time curves for PS and NAVA using a simple, new metric (Range90) that quantify the match of Eadi demand and Vt supply to shows which mode provided a better matching for each patient.

2 Methods

This research analyses prospectively recorded Eadi-time and flow-time curves and derived parameters in a study exploring patient-ventilator synchrony on clinically based criteria [7] at the University Hospital of Geneva (Switzerland) and Cliniques Universitaires St-Luc (Brussels, Belgium). The study protocol was approved by the Ethics committee of both participating hospitals.

2.1 Data recordings

Patients admitted to the ICU, intubated because of acute respiratory failure, and ventilated with PS mode were eligible for the study if they had none of the following exclusion criteria: (1) severe hypoxemia requiring an FIO2 ≥ 50 %; (2) hemodynamic instability; (3) known esophageal problem (hiatal hernia, esophageal varicosities); (4) active upper gastro-intestinal bleeding or any other contraindication to the insertion of a naso-gastric tube; (5) age ≤16 years old; (6) poor short term prognosis (defined as a high risk of death in the next seven days); or (7) neuromuscular disease. All included patients were ventilated with a Servo-I ventilator (Maquet, Solna, Sweden) equipped with the commercially available NAVA module and software, which delivers both PS and NAVA. The main ventilator settings are reported in Table 1.

Table 1 Main ventilator settings
Full size table

After written informed consent was obtained, the patient’s standard nasogastric tube was replaced with NAVA tube. Airway suctioning was performed before the beginning of the protocol. A 20 min continuous recording (~300–500 breaths) was carried out during standard PS with clinician determined ventilator settings, and then 20 min during NAVA with the NAVA level (proportionality factor between recorded Eadi and ventilator delivered pressure) set to obtain similar peak airway pressure as the total pressure obtained in PS using previsualization system built in the ventilator. Eadi and flow traces were acquired from the Servo-I ventilator and recorded at a frequency of 100 Hz by Servo-tracker V4.0 (Maquet, Solna, Sweden). During the entire period, the pressure support level in PS, and NAVA gain in NAVA were kept constant. Equally, positive end expiratory pressure (PEEP), FiO2, inspiratory trigger, and cycling off settings were maintained constant.

2.2 Data analysis

Flow-time and Eadi-time signal obtained from the Servo-tracker were used to perform the analysis using MATLAB software (The Mathworks, Natick, Massachusetts, USA). A breath was determined by the flow signal, and was defined to commence when the flow signal became positive, and terminate when the flow signal became negative. The flow-time signal was integrated to obtain Vt. Vt < 50 ml were discarded from analysis. The Eadi signal was integrated between the same two time points to obtain the corresponding integrated Eadi value (ʃEadi), to represent total inspiratory demand [9]. Peak Eadi value was captured and compared with ʃEadi to ensure no loss of information. ʃEadi is the time integral of Eadi signal and this parameter carries the information on the change of Eadi with time, while retaining information on peak Eadi. Finally the inspiratory time, Ti (Time when flow becomes positive until time when flow became negative) and neural inspiratory time Ti_ Neural (Time when flow became positive until time when peak Eadi occurs) were calculated as shown in Fig. 1.

Fig. 1
figure 1

Example of a patient’s inspiratory ʃEadi and resulting ventilatory Vt

Full size image

2.2.1 Correlation analysis

Pearson’s linear correlation coefficients were calculated for each patient between peak Eadi value and ʃEadi, and also between Vt and ʃEadi. Two-sample non parametric Kolmogorov–Smirnov goodness-of-fit hypothesis tests were then used to compare PS and NAVA groups.

2.2.2 Variation analysis: robust coefficient of variation

Traditionally, the Coefficient of Variation (CV = standard deviation/mean) in Vt, ʃEadi and Ti over all breaths for each patient in each ventilator mode was calculated [4, 9] for variation analysis. However, normality was assessed using a Kolmogorov-Smirnow test that showed that some parameters were not normally distributed. Hence, the robust coefficient of variation (CVR) was used instead (CVR = median absolute deviation/median). CVRs for the PS and NAVA group are also compared using a two-sample non parametric Kolmogorov–Smirnov goodness-of-fit hypothesis test.

2.2.3 Matching analysis: Range90 of supply over demand

Vt and ʃEadi were determined and the ratio Vt/ʃEadi was assessed for each breath in both ventilation mode. Vt is the tidal volume given to patients and ʃEadi is defined as an expression for patient’s ventilatory demand. The ratio of Vt/ʃEadi is previously defined as Neuroventilatory efficiency, by Passath et al. [10], and this the ratio also carries the information of ventilator supply over patients demand (A form of patient-ventilator interaction).

For each patient, the empirical cumulative distribution function (CDF) of Vt/ʃEadi ratio in different ventilation modes was plotted. Range90 is calculated as the range 5th–95th Vt/ʃEadi ratio (Range90 = 95th Vt/ʃEadi − 5th Vt/ʃEadi). A smaller range indicates better matching of the response Vt and, importantly, its variability to the inspiratory ʃEadi demand and its variability. A larger range indicates a lesser ability to match Vt and ʃEadi for each breath regardless of the underlying patient-specific variability in ʃEadi. Thus, if Vt variability were equally matched to variability in ʃEadi, then the ratio would be more consistent (Range90 will be smaller), and a larger Range90 thus indicates an inability to consistently match Vt to ʃEadi demand. Hence, this ratio normalizes differences in Eadi demand within and between patients and-or ventilatory modes in analysing the resulting Vt variability. The ratio of Vt/ʃEadi for each breath and the analysis of its variability (Range90) over a given mode thus allows fair comparison between modes (for a patient) and between patients, where Eadi demand may vary significantly. The detail description of Range90 and case examples can be found in the electronic supplemental file.

3 Results

Twenty-two patients were included in the study. Their major clinical characteristics were the following (mean ± SD): age 66 ± 12 years, BMI 23.4 ± 3.1 kg/m2, recording at 3 ± 2 days post intubation, PaO2/FIO2 ratio at inclusion 194.8 ± 58.1 mmHg. One patient had a history of pulmonary restrictive disease, and 8 patients had a history of pulmonary obstructive disease. For the 22 patients, median [interquartile (IQR)] ʃEadi were not different between PS and NAVA: 4.10 μVs [2.55–5.99] vs 3.97 μVs [2.59–6.64]. Ti was 0.97 s [0.70–1.15] in PS and 0.80 s [0.65–0.92] in NAVA. Median Vt was 468 ml [418–514] in PS and 431 ml [378–472] in NAVA. Peak inspiratory pressure (PIP) was 21.44 cmH2O [18.57–24.11] in PS and 21.63 cmH2O [19.61–24.56] in NAVA. No significant difference was found between the median these values between PS and NAVA. Table 2 shows the summary of median [IQR] values for ʃEadi, Vt, Ti_ Neural , Ti and peak pressure for all 22 patients for PS and NAVA. The box-whisker plot for ʃEadi, Vt, Ti and peak pressure are shown in Fig. 2.

Table 2 Summary of ʃEadi, Vt, Ti_ Neural , Ti and peak pressure or 22 patients median [IQR]
Full size table
Fig. 2
figure 2

Summary of ʃEadi, Vt, Ti and PIP for all 22 patients in PS and NAVA

Full size image

3.1 Correlation analysis

Across all 22 patients, peak Eadi and ʃEadi are correlated (median [IQR]) with R = 0.93 [0.88–0.96] for PS and R = 0.87 [0.78–0.90] for NAVA. The correlation of ʃEadi and Vt (median [IQR]) for PS is R = 0.50 [0.05–0.65]. For NAVA, R = 0.85 [0.78–0.90]. A two sample Kolmogorov–Smirnov goodness-of-fit hypothesis test shows that the NAVA and PS correlation datasets for ʃEadi and Vt are significantly different (p < 0.005), illustrating the much better correlation seen for NAVA which thus better matches these two variables than PS.

3.2 Variability analysis

The coefficients of variation (CVR) in Vt, ʃEadi and Ti are shown in Table 3 for the 22 patients. Using a two-sample Kolmogorov–Smirnov goodness-of-fit hypothesis test, the PS and NAVA CVR in Vt are significantly different (p < 0.005), with the NAVA being more variable. Conversely, no significant difference was observed between ʃEadi or Ti_ Neural CVR, with p = 0.563 and p = 0.332, respectively. More importantly, these results are reported for the population of 22 patients. Individual patients could exhibit a very different variability between PS and NAVA, and the specific variables reported.

Table 3 Robust coefficient of variation (CVR) in Vt, ʃEadi, Ti_ Neural , Ti and PIP (median [IQR])
Full size table

3.3 Range90 matching analysis

For NAVA, the median [IQR; 5th–95th percentile] Range90 = 71.0 ml/μVs [36.5–153.6; 15.0–531.3]. For PS, Range90 = 129.9 ml/μVs [64.0–341.5; 19.2–645.2]. These results indicate significant variability in the matching of Vt and ʃEadi for both modes, but a consistently lower range for NAVA. The Range90 for both PS and NAVA in all patients are shown in Table 4.

Table 4 PS and NAVA Range90 (ml/μVs) for all patients
Full size table

As a larger Range90 value indicates less matching and since the Range90 value is normalized, the value for each mode can be compared. Examining the ratio of Range90 for PS/NAVA, the median [IQR; 5th–95th percentile] of this ratio is 1.72 [1.39–3.14; 0.97–7.18], showing that NAVA consistently had much smaller values than PS. The lower quartile of this Range90 ratio is [0.68–1.39], where only 1 patient had a value less than 1.0, at 0.68. These results show that 21 of 22 patients had better matching of Vt and ʃEadi for NAVA than for PS. 4 patients had values greater than, but near to, 1.0, indicating relatively comparable matching of Vt and ʃEadi between NAVA and PS.

Figure 3 show the CDFs for the ratio of Vt to ʃEadi for 3 specific patients, along with dashed lines indicating the Range90 value. These three patients show a typical case where NAVA better matches Vt and ʃEadi than PS, the one case where PS better matches these variables than NAVA, and one case where they are similar showing the patient with ratio of Range90 values (PS/NAVA = 1.24) that is close to 1.0.

Fig. 3
figure 3

Cumulative distribution function (CDF) plots of Vt/ʃEadi for both NAVA and PS for three patients. The CDFs show the ~ 300–500 such values per patient and mode. The dashed lines show the variability (along x-axis) in this ratio or matching, where a narrower band is a smaller Range90 value and thus better matching of Vt and ʃEadi. a Patient 3: NAVA is better than PS; b patient 4: NAVA and PS are similar; c patient 21: PS is better than NAVA

Full size image

Figure 4 shows two patients, also shown in Fig. 3, whose results highlight the differences in ability to match Vt to ʃEadi. In particular, the figure plots ʃEadi and Vt for each breath for both modes and each patient. In these figures, it is clear that PS provides a far more constant range of (output) Vt despite similar variation in ʃEadi for both ventilatory modes.

Fig. 4
figure 4

Vt–ʃEadi plots for NAVA and PS. a Patient 3 with PS/NAVA ratio = 1.51. b Patient 4 with PS/NAVA ratio = 1.24. The dashed lines around the data capture the middle 90 % of the data for both ʃEadi and Vt for each mode. In both patients, the outcome tidal volume range for the middle 90 % of breaths is 50–100 mL wide for PS and much wider for NAVA, despite similar input ranges of ʃEadi. The smaller ratio of width over height of the box between modes is similar to having a smaller Range90

Full size image

4 Discussion

Peak Eadi and ʃEadi were well correlated under both PS and NAVA ventilation modes. Hence, peak Eadi and ʃEadi are essentially identical and could be used to characterize Eadi in future studies. ʃEadi is used instead of Eadi value in this study as it is more physiologically relevant as total inspiratory demand and readily calculated [9].

Significantly better correlation was observed between Vt and ʃEadi under NAVA than PS, indicating that the delivered ventilation is possibly more physiological to the patient under NAVA than PS by better matching the patient’s demand and its variability.

Variability in Vt, ʃEadi and Ti were determined using the traditional method of reporting the robust coefficient of variation (CVR). It was shown that there is no statistically significant difference in the variability of ʃEadi, Ti_ Neural and PIP between PS and NAVA over the population of 22 patients. However, the CVRs in Vt and Ti were significantly different between PS and NAVA (p < 0.005), with NAVA tidal volumes and inspiratory time more variable. As increased variability has previously been associated with improved ventilation/perfusion ratio and improved oxygenation both in animal models [11–13] and humans [14], the ability of NAVA in maintaining higher variability in tidal volume compared to PS could be of potential clinical interest and will require further investigation. Moreover, overall, it is hypothesized that greater variability in breathing pattern is a healthier patient condition. Thus, it is thought that better patient outcomes could be expected for those patients with a higher variability in ʃEadi when it is equally matched by a high variability in Vt. The results presented show that such a result is significantly more likely under NAVA than PS.

In this study, Range90 values were significantly smaller in NAVA than in PS. Of 22 patients, 21 had lower Range90 with NAVA, and only 1 showed a higher ratio with a better match of ʃEadi and Vt by PS. Equally, the comparison of Range90 values showed that 4 of the 5 patients comprising the lowest quartile had values near to 1.0 with the 25th percentile value showing the match of Vt and ʃEadi was 1.40× better for NAVA than for PS in Range90 value. Hence, ventilation under NAVA is probably more physiological and adapted to the patient’s inspiratory demand than under PS.

The importance of normalizing Vt variability to its ʃEadi variability is highlighted in when it is observed that although there is no significant difference in ʃEadi variability over the population in Table 2, ʃEadi can be substantially different between PS and NAVA on a patient specific basis. Hence, the patient-specific comparisons of Range90 are particularly relevant, and, equally, account for all breaths directly rather than via a grouped statistic.

The correlation between Range90 with CVR in ʃEadi and Vt; For PS, correlation of coefficient R was 0.71 for ʃEadi, and 0.67 for Vt, compared to NAVA, R = 0.16 for ʃEadi, and 0.02 for Vt. These results indicated that Range90 is an alternate analysis different from variability analysis, and thus a higher variability does not necessarily mean better ‘matching’. The Range90 metric is very simple. Hence, the ventilator Vt in response to (not necessarily equally) patient’s inspiratory demand can be calculated in real-time. Thus, it could be used to monitor response to ventilator settings and possibly (in future) adapt ventilator settings.

Overall, the results show that PS does not adapt or respond to ʃEadi variation, as it provided relatively constant tidal volumes regardless of the magnitude of ʃEadi. This point is illustrated in Fig. 4. In both cases of Fig. 4 the range of tidal volumes seen is very narrow for PS, particularly relative to the ʃEadi range. Thus, PS was unable to match Vt to the ʃEadi.

Prior to this study, Piquilloud et al. [7] examined the patient-ventilator asynchronies in pressure support and NAVA, namely ineffective effort, auto-triggering, premature cycling, delayed cycling and double triggering. This work further extends the comparison of PS and NAVA in patients-specific level. In particular, Range90 is a novel method of determining the matching (Synchronisation) of the magnitude of the ventilator supply and magnitude of patient’s demand (Neuroventilatory efficiency, Vt/ʃEadi ratio). The resulting 90 % range (Range90) value for each mode shows how well outcome Vt was matched by the ventilatory mode to inspiratory demand, ʃEadi. A smaller Range90 indicates better matching, thus better response to variable demand with equally variable Vt. This approach thus provides a patient-specific comparison of which mode better matched Vt and ʃEadi. Equally, the comparison of Range90 values for a given patient enable one to quantify how much better one mode matched inspiratory demand and resulting tidal volume. Overall, what makes this work and Range90 unique is its patient-specificity, enabling patient-specific comparison where prior works [6–9, 15] make comparison on a global or cohort level between modes of ventilation or ventilation strategies.

4.1 Limitations

One possible limitation of this study is the definition of a minimum tidal volume that defines a breath versus an artefact, where those breaths with Vt < 50 ml were ignored. The selection was made through post hoc analysis of the Vt distribution indicating a bimodal distribution, one mode representing physiological respiratory activity and a small mode (Vt < 50 ml) likely corresponding to artefacts in Eadi. This statement is based on the observation that there was no significant ventilator pressurisation associated with Vt lower than 50 ml. A re-analysis of the data with a limit of 100 ml showed no change in the overall results, wherein NAVA provided a better match to patient variability in ʃEadi for 21 of 22 patients, and the lower quartile still had a PS/NAVA ratio of 1.40. Thus, the analysis presented is robust to this choice.

Range90 only considers the matching of Vt magnitude of a ‘known’ breathing cycle towards the corresponding ʃEadi. Thus, if Vt > 50 ml and ʃEadi exist, it is used, and any asynchrony is seen as a mismatch of the Vt and ʃEadi magnitude. If Vt < 50 ml and ʃEadi exist, this information is not included. However, 300–500 ‘known’ breathing cycles are analysed for every patient in each ventilation mode. An average of only 30 breaths of ~500 per patient (6 %) were discarded in each phase. Thus, the loss of this potential data, assuming they are true breaths, is negligible.

It is important to note that, while Range90 shows better matching in NAVA than PS, it is yet to be applied as a bedside monitoring tool. Range90 in this study used 20 min of breathing pattern of a patient in each ventilation mode, and the total duration fort the study is not clinically feasible. In addition, the availability of Eadi signal for Range90 analysis is dependent on the availability of NAVA nasogastric tube. This study is a proof of concept and patient respiratory adaptation time to different ventilation mode was taken into consideration [16]. Thus, the results indicate that the use of Range90 as a bedside monitoring tool warrants further investigation with shorter monitoring time.

Another limitation is that setting NAVA gain to obtain similar peak pressure as PS, is an approximation based on the built in previsualisation system and only one NAVA gain value is used to compare with PS. The amplitude of delivered pressure during NAVA is variable, as it is proportional to Eadi signal. Therefore, it is possible that the delivered peak pressure may be higher or lower in different patients when comparing PS and NAVA. An online supplement file is included to show peak pressure delivered in different ventilation for every patient. However, it was found that there was no significant difference between peak pressure between these two ventilation modes (p > 0.005) as shown in Table 2. Thus, for this cohort of 22 patients, the peak pressure in both modes can be considered as ‘similar’.

In this study, only one NAVA level and one PS level was tested and the influence of increasing NAVA level towards Vt/ʃEadi was not determined. The effect of Vt and ʃEadi related to increase in NAVA level have been extensively described by Passath et al. [10]. Based on the results, we can assume that increasing NAVA level will result in a consecutive decrease in Eadi and in a small and only initial increase in Vt. As a consequence, we can assume that with increased NAVA level, Vt/ʃEadi ratio will also increase, but this point must be formally explored in future studies to test this hypothesis.

5 Conclusions

Compared to PS, NAVA allowed better match between Vt and Eadi as well as higher variability in Vt. As higher variability has been associated with improved oxygenation and as higher variability is present in healthier systems, the ability of NAVA in not reducing patient’s intrinsic variability could be of potential clinical interest. Future work is needed to explore if there was a potential effect of this maintained variability obtained with NAVA on patients’ outcome.