Introduction

The rapid discontinuation of mechanical ventilation (MV) is a major objective of intensive care unit medicine. Current guidelines recommend that weaning from the ventilator be achieved using a two-step strategy [1]. First, there should be an assessment of the readiness of the patient for weaning; this is to be followed by the second step, which should be a spontaneous breathing trial (SBT) as a diagnostic test to determine the likelihood of successful extubation [2]. Various weaning predictors derived from breathing pattern analyses have been developed to enable the likelihood of a successful weaning to be assessed as early and accurately as possible [36]. Since the success of weaning depends on the capacity of respiratory muscle to overcome the load that impedes the respiratory system, most of these indices are surrogates of the load capacity balance. The most commonly studied of these indices is the ratio of respiratory frequency (respiratory rate) to tidal volume (RR/V T), otherwise known as the rapid shallow breathing index (RSBI) [3], and one of the most accurate is the variability of breathing pattern descriptors [6]. Central respiratory drive, which is increased during weaning failure, constitutes an alternative approach for predicting weaning. Occlusion pressure (P0.1) and P0.1-derived parameters are among its surrogates [712], but the performance of these parameters remains controversial [13]. In addition, although P0.1 is a marker of respiratory drive, it is influenced by muscle strength. This is in contrast to the diaphragm electromyographic activity (EAdi)-derived indices, which have recently become easier to determine with the availability of the neurally adjusted ventilatory assist (NAVA) catheter [14]. Deemed to be used in conjunction with the NAVA mode, which subordinates the ventilator cycle and level of assistance to EAdi, the NAVA catheter consists of a nasogastric feeding tube equipped with a multiple-array esophageal electrode that provides real time access to EAdi.

In the study reported here, we hypothesized that EAdi would provide reliable predictors of weaning success. Our aim was therefore to compare patients successfully separated from the ventilator and the endotracheal tube with those who were not, in terms of various EAdi-derived indices recorded during the SBT. We also compared the performance of these indices to the performance of the RR/V T. Finally, because it has been demonstrated that indices of breathing variability can discriminate weaning success from weaning failure [6], we also sought to evaluate the indices of EAdi variability to predict weaning success.

Materials and methods

The study was conducted over a 12-month period (1 November 2010 to 1 November 2011) in a ten-bed Intensive Care Unit (ICU) within an 1,800-bed university hospital. The protocol was approved by the institutional review board of the French Learned Society for Intensive Care Medicine (no. 11–321). Informed consent was obtained from the patients.

Patients

Patients intubated and ventilated for at least 24 h were eligible for inclusion in the study if (1) they had been previously mechanically ventilated with NAVA mode and the NAVA catheter had been left in place [the preconditions for NAVA are listed in the Electronic Supplementary Material (ESM)] and (2) they met the predefined readiness-to-wean criteria on the daily screening (see ESM). Patients with neuromuscular diseases and those for whom life support would be withheld or withdrawn were not included in the study.

Spontaneous breathing trial protocol

As soon as the patients fulfilled the readiness-to-wean criteria, a 30-min SBT was performed with the patient connected to the ventilator (pressure support level 7 cmH2O, zero end expiratory pressure). The SBT was considered to be a failure if at least one the following criteria was present: (1) blood oxygen saturation (SpO2) of <90 % with a fraction of inspired oxygen (FiO2) of ≥50 %; (2) acute respiratory distress (RR ≥ 40/min, agitation, cyanosis); (3) systolic arterial blood pressure of ≥180 mmHg; (4) cardiac arrhythmias; (5) respiratory acidosis [pH < 7.32 with an arterial carbon dioxide tension (PaCO2) of ≥50 mmHg]. If none of these failure criteria was present, the SBT was considered as successfully completed and the patient was extubated. This decision was taken by the physician in charge of the patient. The patient was reconnected if signs of intolerance (see above) were present. The separation from the ventilator and the endotracheal tube was considered a success when spontaneous breathing could be sustained without any form of ventilatory support at 48 h after extubation. Conversely, cases of failure included patients who failed the SBT and patients requiring reintubation or any form of ventilator support (including non-invasive ventilation for post-extubation acute respiratory failure) during the first 48 h after extubation. Patients were only studied once, at the first SBT.

Data acquisition

Diaphragm electromyographic activity (obtained from the NAVA catheter; see ESM for detailed EAdi signal processing), airway pressure and flow were acquired at 100 Hz from the ventilator via a RS232 interface connected to a computer using commercially available software (Servo-I® RCR ver. 3.6.2; Maquet Critical Care, Solna, Sweden). Physiologic variables were measured at baseline and at 3, 10, 20 and 30 min during the SBT. Arterial blood gases were sampled at the end of the SBT.

Fig. 1
figure 1

Schematic representation of the airways pressures and diaphragm electromyographic activity (EAdi) signals. Upper panel Airway pressure recording, lower panel EAdi recording. Dotted line no. 1 indicates the start of the rise of the EAdi signal, dashed line no. 2 shows the peak of the EAdi (EAdi max ). Grey area represents the area under the curve of EAdi (EAdi AUC )

Data analysis

Respiratory rate (RR), tidal volume (V T), maximum EAdi (EAdimax) and the area under the curve of the EAdi during inspiratory time (EAdiAUC, integrated from baseline to peak; Fig. 1) were calculated at the very beginning of the SBT (baseline) and at 3, 10, 20 and 30 min during the SBT. The rapid shallow breathing index (RR/V T), the EAdimax/V T, the EAdiAUC/V T and the difference between EAdimax and the lowest EAdi value reported to the V T ratio (ΔEAdi/V T) were also computed.

The coefficient of variation (CV), defined as the standard deviation (SD) reported for the mean of both V T and EAdimax, was calculated at 3 and 30 min for a duration of 3 min at each time-point using a software routine developed for Matlab (Mathworks, Natick, MA).

Statistical analysis

The statistical analysis was performed with Prism ver. 4.01 software (GraphPad Software, San Diego, CA). The results are expressed as the median [25–75 % interquartile range (IQR)]. The normality of the distribution was evaluated with the Kolmogorov–Smirnov test. For a given time-point, patients successfully (success group) and unsuccessfully (failure group) separated from the ventilator were compared using Student’s t test or the Mann–Whitney U test as appropriate. Within a given group (success or failure), the five time-points (baseline, 3, 10, 20 and 30 min) were compared using analysis of variance (ANOVA) for repeated measures or the Friedman Test. Categorical variables were compared with the χ2 test or Fisher’s exact test. Differences were considered to be significant when p < 0.05. The predictive performance of RR/V T and EAdi-derived indices at the 3 min time-point was finally assessed with receiver operating characteristic (ROC) curves [15]. For RR/V T, the cut-off of 100 breaths min−1 l−1 was retained [3], while for EAdi-derived indices, the cut-off was obtained for each index by determining the better sensitivity and specificity.

Results

Patients’s characteristics

A convenience sample of 57 consecutive patients were enrolled in this study (see flow chart in ESM Fig. E1), of whom 35 and 22 patients were successfully and unsuccessfully separated from the ventilator, respectively. In the failure group, the endotracheal tube was not removed at the end of the SBT in 19 patients, and the endotracheal tube was removed but ventilatory support was required within 48 h for three patients. The two groups were similar in terms of demographic and clinical characteristics, but there were more patients with chronic respiratory failure in the failure group (Table 1). Patients in the failure group were longer on mechanical ventilation, but ICU mortality was similar in both groups (Table 1).

Table 1 Patients’ characteristics

Breathing pattern at baseline and during the SBT

At the initiation of the trial, V T and RR were similar in patients in both groups, but the RR/V T was higher in patients in the failure group (Table 2). Figure 2 depicts the V T, RR and RR/V T at baseline and at 3, 10, 20 and 30 min during the SBT (see also ESM Table E1). During the SBT, RR increased significantly in the failure group and decreased significantly in the success one. Simultaneously, V T decreased in the failure group, whereas it remained stable in the success group. The RR/V T ratio did not change over time in the success group, while it increased significantly in the failure group. Overall, 3 min after the beginning of the SBT, V T, RR and RR/V T were significantly different between patients who were successfully weaned and those who were not.

Table 2 Breathing pattern and diaphragm electromyographic activity-derived indices at baseline
Fig. 2
figure 2

Breathing pattern during the spontaneous breathing trial (SBT). Tidal volume (V T ) (upper panel), respiratory rate (RR) (middle panel) and RR/V T ratio (lower panel) at baseline and during the SBT (3, 10, 20 and 30 min) in patients successfully separated from the ventilator (white boxes; success group) and those who were not (grey boxes; failure group). Line inside the boxes median, limits of the boxes 75th and 25th percentile of the data (interquartile range), whiskers minimum to maximum. *p < 0.05 vs. failure group, #p < 0.05 vs. baseline within the failure group, $p < 0.05 within the success group

EAdi-derived indices at baseline and during the SBT

Table 2 presents the EAdi-derived variables at baseline and Fig. 3 depicts the EAdimax, EAdiAUC, EAdimax/V T, EAdiAUC/V T and ∆EAdi/V T at baseline and during the SBT at 3, 10, 20 and 30 min (see also ESM Table E1). At the initiation of the trial, all of the EAdi-derived variables were different between patients who were successfully weaned and those who were not (Table 2). These differences remained significant during the entire SBT. However, during the SBT, EAdi-derived variables remained stable throughout the SBT in both the success and failure group with the exception of EAdimax/V T and ∆EAdi/V T, which increased significantly in the failure group (Fig. 3).

Fig. 3
figure 3

Diaphragm electromyogaphic activity-derived indices during the SBT. EAdimax (upper left panel), EAdiAUC (middle left panel), amplitude between the maximal and lowest EAdi (∆EAdi)/V T ) ratio (lower left panel), EAdimax/V T ratio (right upper panel) and the EAdiAUC/V T ratio (middle right panel) at baseline and during the SBT (3, 10, 20 and 30 min) in patients successfully separated from the ventilator (white box) and those who were not (grey box). Line inside the boxes Median, limits of the boxes 75th and 25th percentile of the data (IQR), whiskers minimum to maximum. *p < 0.05 versus failure group, # p < 0.05 versus baseline within the failure group, $ p < 0.05 within the success group

Performance of RR/V T and EAdi-derived indices at 3 min into the SBT

Figure 4 presents the ROC curves and performance of the RR/V T (threshold of 100 breaths min−1 l−1), EAdimax/V T, EAdiAUC/V T and ∆EAdi/V T at 3 min into the SBT. The best values of sensibility, specificity and AUC were obtained with a threshold of 28 μV l−1 for EAdimax/V T, 11 μV2 l−1 for EAdiAUC/V T and 26 μV L−1for ∆EAdi/V T. The sensitivity and the specificity of these two indices were similar to the performance of RR/V T.

Fig. 4
figure 4

Receiving operating curves (ROC) for breathing pattern and EAdi-derived indices during the SBT. ROC curves for the RR/V T ratio (upper left panel), EAdimax)/V T ratio (upper right panel), EAdiAUC/V T ratio (lower left panel) and ∆EAdi/V T ratio (lower right panel) to predict the success or failure of the separation of patients from the ventilator at 3 into the SBT. Threshold, sensibility, specificity and area under the ROC curve (AUC) are provided in the table

Variability of the V T and the EAdimax

The CV of the V T at the onset of SBT was similar in both the success and failure patients [18 (IQR 13–28) vs. 17 (IQR 11–34) %, respectively]. This decreased significantly in both groups during the SBT to reach, by the end of the SBT, 12 (IQR 8–25) % in the success group patients and 12 (IQR 9–21) % in the failure group patients. The CV of EAdimax was also similar in both groups at onset of SBT [success group: 30 (IQR 24–43) %; failure group: 27 (IQR 18–36) %]. However, while the CV of EAdimax decreased during the SBT in success group patients [18 (IQR 9–32) %], it remained unchanged in failure patients [27 (IQR 18–36) %].

Tolerance of the SBT

By the end of the SBT, patients of the failure group had a higher systolic blood pressure, a lower pH and a higher PaCO2 (see ESM Table E2).

Discussion

To our knowledge, this study is the first to report a direct and physiologic approach with NAVA of neuromechanical coupling during the weaning of mechanical ventilation. Our major findings can be summarized as follows: (1) several of the EAdi-derived indices measured during the SBT were significantly different between patients successfully separated from the ventilator and those for whom separation failed; (2) these differences occurred early during the SBT and were even present at baseline; (3) 3 min after the beginning of the SBT, the performance of EAdi-derived indices of neuromechanical coupling was similar to—but not better than—the performance of the RR/V T ratio; (4) variability of the EAdi signal is not a reliable predictor of weaning failure.

Breathing pattern and EAdi-derived indices during SBT

During the SBT, we observed what has previously been widely described: an increase in the RR with a concomitant decrease in the the V T in the failure group and no change in the success group [16, 17]. These changes resulted in the development of a rapid shallow breathing pattern in patients who failed the SBT, as suggested by the increased RR/V T ratio. The RR/V T ratio was not used as a SBT failure criterion because it might certainly have caused a bias in its comparison with the EAdi-derived indices.

Interestingly, EAdimax and EAdiAUC were higher in the failure than in the success group from the very beginning of the SBT, suggesting that patients who fail the SBT may have a higher respiratory drive at baseline [16]. However, the EAdi did not change significantly during the course of the SBT in either group. This latter observation suggests that the central respiratory drive to the diaphragm did not increase in failure group patients despite these patients developing a rapid shallow breathing. Taken together, these data suggest that during weaning failure, extradiaphragmatic inspiratory muscles, such as sternomastoid and ribcage muscles, are proportionally more recruited than the diaphragm. This suggestion is consistent with the well-observed preferential recruitment of extradiaphragmatic inspiratory muscle during loaded breathing [18]. Hence, EAdi might not be the best surrogate of inspiratory muscles failure during weaning. Of course, the steady EAdi concomitant with the development of a rapid shallow breathing strongly suggests an altered neuro-mechanical coupling.

Beyond ventilatory drive, neuro-mechanical coupling as a marker of weaning failure

We computed various indices of neuro-mechanical coupling, namely EAdimax/V T, EAdiAUC/V T and ∆EAdi/V T. These indices reflect the ability of the diaphragm to convert respiratory drive into ventilation. We showed that patients in the weaning failure group had a severe impairment of the neuromechanical coupling of the diaphragm, which was not only present from the beginning of the SBT, but also increased progressively during the entire SBT. All of these indices were able to efficiently discriminate success from failure patients as early as 3 min after the beginning of the SBT. However, the performance of these indices was not superior to the performance of RR/V T. The fact that the SBT was performed with a pressure support of 7 cmH2O may have lowered the performance of EAdi-derived indices to discriminate patients who were successful in the SBT from those who failed as 7 cmH2O of pressure support lowers the performance of the RR/V T. Indeed, even if the good performance of RR/V T is well demonstrated while the patient is disconnected from the ventilator and breaths room air without pressure support [2], this is not the case under pressure support [19, 20].

Our results differ slightly from those of Wolf et al. [21] who recently observed in children that the V T/∆EAdi ratio after 1 h of SBT decreased dramatically in patients who were successfully weaned while it remained unchanged in those who were not. The different duration of the SBT (1 h vs. 30 min) and the different age of the population (children have indeed a lower rib cage compliance than adults) may explain this discrepancy.

Variability of EAdi

Variability in the various descriptors of the breathing pattern has been shown to be reliably predict weaning success [6]. We therefore sought to evaluate the performance of EAdi variability to predict weaning failure. First, we did not observe an earlier described difference in the CV of the V T between patients who were successfully separated from the ventilator and those who were not [6]. This is not highly surprising since we performed SBT with pressure support and not while the patient was disconnected from the ventilator [22]. Pressure support indeed unloads the respiratory system and therefore alters the load capacity balance of the respiratory system, of which the breathing pattern variability is a surrogate [23]. Secondly, the variability of the EAdi was similar among patients who succeeded and those who failed the SBT. This result is consistent with the findings of a previous study conducted by our group, which showed that the underlying variations in central respiratory neural output activity seem not to be particularly sensitive to the mechanical load [24].

Study limitations

Our study has several limitations. First, it was a single-centre study with a relatively small sample size. Second, to keep the study observational, we only included patients who had been previously ventilated in NAVA, which constitutes a selection bias. Indeed, in our unit, NAVA is given to patients with a poorer tolerance for pressure support ventilation and a longer duration of mechanical ventilation; these patients are eventually more likely to be difficult to wean. In addition, we restricted the study to patients ventilated for more than 24 h, which could explain why chronic obstructive pulmonary disease (COPD) patients were overrepresented in our patient cohort. Third, the SBT was performed with a minimal pressure support of 7 cmH2O and not with the patient disconnected from the ventilator and breathing room air without pressure support. Compared to a SBT without pressure support (“T-piece”), pressure support ventilation with or without positive end-expiratory pressure (PEEP) is known to markedly modify the breathing pattern and inspiratory muscle effort [20, 25]. Hence, the performance of EAdi-derived indices might have been higher during a SBT performed with a T-piece or PEEP without pressure support. However, in our unit and for safety reasons, the standard of care is to perform the SBT with the patients connected to the ventilator. Because we wished to perform a “real life” study, we did not change the SBT procedure standardly used at our hospital [1]. Fourth, there were more COPD patients in the failure group than in the success group. In these patients, dynamic hyperinflation may have altered neuromechanical coupling and diaphragmatic performance [2628]. Moreover, patients with chronic respiratory diseases are known to strongly recruit their extradiaphragmatic inspiratory muscles when facing increased inspiratory load [29]. Fifth, the NAVA catheter has been designed to trigger and adjust ventilatory assist and not for weaning. Thus, it cannot provide information, such as the ratio of high to low frequencies of the electromyograph (H/L), which predicts diaphragm fatigue or excessive loading [30, 31]. Moreover, although we have some information, we do not know the whole signal processing of the technique and have unfortunately no access to raw data. In addition, there might be an inter-subject variability in EAdi recording, depending on many factors and, to date, there is no study that has compared EAdi in large populations of patients. Finally, the study was not strictly blinded. Indeed, the physician in charge of the patients could see the EAdi recordings during the SBT, which may have influenced his/her decision regarding extubation. For this reason, we used strict criteria of SBT success and failure, which are actually those used in our ICU in daily clinical practice. Of note, we did not consider RR/V T as a criterion for extubation despite its good performance as an index of SBT success or failure as this would have biased the comparison of the performance of this index to the performance of EAdi-derived indices.

Conclusions

EAdi-derived indices collected via the NAVA catheter during the SBT provide reliable predictors of successful separation from the ventilator. These indices can discriminate early in the SBT the ongoing success or failure of separation from the ventilator. However, the performance of these indices is not better than the performance of the rapid shallow breathing index. Therefore, the benefit of EAdi-derived indices during the SBT remains unclear.