Estimation of the diaphragm neuromuscular efficiency index in mechanically ventilated critically ill patients

This prospective cohort study was performed in an academic ICU. Adult patients, with a dedicated EAdi catheter (Maquet critical care, Solna Sweden) in situ and mechanically ventilated in neurally adjusted ventilator assist (NAVA) mode were eligible for inclusion. The institutional ethical committee approved the study protocol and informed consent was waived due to the non-invasive nature of the study and negligible risks.

The EAdi catheter was positioned according to the manufacturer’s instructions using a dedicated software tool on the Servo-i ventilator. The catheter position was verified before data acquisition. When patients exhibited a stable breathing pattern (i.e. no coughing, hiccups or disproportional differences in respiratory rate), measurements to assess respiratory muscle function were performed: (1) an end-expiratory occlusion maneuver (obtained by activating the expiratory hold button on the ventilator for one inspiratory effort) for measurement of NMEoccl and (2) a “zero assist breath” in which inspiratory support was decreased to zero for one single breath to calculate patient-ventilator breath contribution (PVBC) [25, 26]. Both maneuvers were repeated five times with at least a one-minute interval. The number of breaths between maneuvers was variable to prevent anticipation by the patient. Measurements were recorded and stored for offline analysis at time T = 0 and after 12 h (T = 12), 24 h (T = 24) and 72 h (T = 72).

EAdi, flow and Paw waveforms were acquired from the Servo-i ventilator via a RS232 serial port connected to a laptop with dedicated software (Servo Tracker version 4.1, Maquet, Solna, Sweden). Maximum inspiratory pressure (MIP) measurements were performed with a manovacumeter (Micro Respiratory Pressure Meter, Carefusion, Yorba Linda, CA, USA) connected to the endotracheal tube [27].

A software routine developed for MatLab (version R2016b, MathWorks Inc., Natick, MA, USA) was used for offline analysis.

NMEoccl was calculated in three different ways (Fig. 1): (1) by dividing ∆Paw by ∆EAdi [21], in which ∆Paw is the difference in pressure between the lowest Paw during end-expiratory occlusion and the preceding end-expiratory pressure level, (2) by dividing the area under the curve (AUC) of Paw and EAdi, and (3) by dividing Paw and EAdi at fixed points (steps of 3 μV) on the EAdi waveform. The inspiratory time (Ti) was defined as the period between the onset of EAdi and 70% peak of EAdi. To investigate the effect of an occlusion on the inspiratory time, we compared Ti during occlusion with the Ti of three preceding unloaded breaths. To estimate Pmus under clinical conditions, the mean EAdi of five tidal breaths (before the end-expiratory occlusion) was multiplied by NMEoccl/1.5 [21]. The correction of 1.5 compensates for the fact that in the presence of flow, the diaphragm generates less pressure for the same EAdi than during an occlusion [21]. The tension-time index (TTI) was calculated as Pmus/MIP multiplied by the ratio of Ti to total respiratory cycle time (T_tot) [28].

The repeatability coefficient (RC) represents the absolute value by which two repeated measurements in one subject will differ in 95% of cases. The formula developed by Bland and Altman was used to calculated RC:

1.96 × √2 × Within-subject standard deviation (SD) [29].

One-way analysis of variance (ANOVA) was used to obtain the within-subject SD with the subject as dependent factor and the repeated NMEoccl measurements as independent factors. Since the NMEoccl variability increased as the magnitude of NMEoccl increased, the ratio of a single NMEoccl measurement to the mean NMEoccl of five repeated measurements was used (see Additional file 1) [29]. The correlation coefficient with repeated observation was used to investigate the within-subject correlation between Paw and EAdi [30, 31] (IBM SPSS Statistics version 22).

The Pearson correlation coefficient was calculated to test correlation between Ti and NMEoccl, and between TTI and Pmus. One-way ANOVA was used to investigate the changes in NMEoccl and inspiratory muscle pressure over time.

A representative NMEoccl maneuver is shown in Fig. 1. At T = 0, 149 maneuvers were obtained in 31 patients, 6 maneuvers were lost due to technical issues. Mean ∆Paw was 14.1 ± 7.9 cmH₂O and mean ∆EAdi 14.8 ± 9.9 μV. Mean NMEoccl was 1.22 ± 0.86 cmH₂O/μV (ranging from 0.41 to 3.56 cmH₂O/μV), with a RC of 82.6%. This implies that when NMEoccl is 1.0 cmH₂O/μV, it is expected with a probability of 95% that the subsequent measured NMEoccl will be between 0.17 and 1.83 cmH₂O/μV. When the EAdi of this patient during normal breathing is 10 μV, the estimated pressure generated by the inspiratory muscles would be somewhere between 1.1 cmH₂O and 12.2 cmH₂O (calculated as EAdi * NMEoccl/1.5, see “Methods”). In addition, the RC of NMEoccl calculated by the AUC and at fixed points on the EAdi curve (Fig. 2) remained high: 87.7% for AUC and 85.5–175.9% at fixed points. The RC of the separate components of NMEoccl was 95.7% for EAdi and 73.9% for Paw. Figure 2 shows there was moderate correlation between ∆EAdi and ∆Paw in the individual patient (mean r = 0.52, p < 0.0005; ranging from r = − 0.90 to r = 1.0).

Upon visual inspection it appeared that some of the EAdi tracings exhibit a rather non-physiological shape: a plateau during inspiration (while an increase would be expected) or at maximum inspiration, or a delayed increase in EAdi relative to the decrease in Paw (Fig. 3). It was reasoned that non-physiological-appearing EAdi waveforms contribute to the high NMEoccl variability. Additional file 2 shows every occlusion maneuver obtained in this study. Different mathematical approaches were used in order to try to objectively detect and exclude non-physiological EAdi waveforms (See Additional file 3). Despite these mathematical approaches the NMEoccl repeatability remained high with a RC of 63.4%.

Finally, we pragmatically selected three out of five occlusions with the lowest variability and averaged the three values to obtain a single NMEoccl for each individual patient. This approach reduces the influence of erroneous values, irrespective of the origin, and will result in a more reproducible NMEoccl value. As a result, the RC improved to 29.8%. This approach was used for subsequent analysis.

The median CoV NMEoccl at T = 0 was 23.1% (IQR 18.7–29.9%). After dividing the study population into two groups with CoV higher or lower than median, respectively, no significant differences in clinical parameters were found (see Additional file 4). Median Ti during occlusion was 0.51 s (IQR 0.35–0.65 s) and of the preceding breaths 0.65 s (IQR 0.48–0.84 s). On average, in 48% of the measurements within a patient, Ti of the occluded breath was longer compared to Ti of the preceding breaths. Only moderate negative correlation was found between Ti and NMEoccl (r = − 0.219).

Twelve patients completed the 72 h study period. In these patients, NMEoccl at T = 0 was 0.8 cmH₂O/μV (IQR 0.7–1.1 cmH₂O/μV) and did not change over time (p = 0.75); 0.7 cmH₂O/μV (IQR 0.4–0.9 cmH₂O/μV), 0.8 cmH₂O/μV (IQR 0.6–1.1 cmH₂O/μV), 0.9 cmH₂O/μV (IQR 0.6–1.3 cmH₂O/μV) for T = 12, T = 24 and T = 72, respectively.

The mean EAdi of five unloaded breaths before the end-expiratory occlusion were used to calculate Pmus. At T = 0, median Pmus was 9.4 cmH₂O (IQR 6.0–12.8 cmH₂O) and did not change over time (p = 0.58); 10.4 cmH₂O (IQR 5.4–15.4 cmH₂O), 10.7 cmH₂O (IQR 4.8–13.1 cmH₂O) and 8.2 cmH₂O (IQR 6.0–15.9 cmH₂O) for T = 12, T = 24 and T = 72, respectively. The median Pmus varied widely among patients, ranging from 1.8 to 36.0 cmH₂O. Based on previous literature [21, 23], we defined a physiological Pmus between 5 and 10 cmH₂O. At T = 0, 6 patients (19.4%) had a mean Pmus < 5cmH₂O and 12 patients (38.7%) > 10cmH₂O, indicating ventilator over-assist or under-assist, respectively.

MIP was obtained in 15 patients in the week before or after the NME measurements, which allowed us to calculate TTI. Median MIP was 29 cmH₂O (IQR 24–38 cmH₂O) and median TTI was 0.12 (IQR 0.08–0.17). Figure 4 shows the relationship between TTI and Pmus.

Variability of both components of NMEoccl (Paw and EAdi) is expected given the variability in inspiratory drive, even during occlusions. However, NMEoccl itself is independent from respiratory drive and should therefore be more stable from breath to breath.

Beck and colleagues reported a linear relationship between EAdi and Pdi in healthy subjects, at least up to 75% of maximum force [32]. Their subsequent study also demonstrated a linear relationship between EAdi and Pdi in patients with acute respiratory failure [20]. Bellani et al. showed that NMEoccl derived from airway pressure closely reflects NME during normal breathing (NMEdyn) derived from Pes and concluded that calculation of NMEoccl allows a clinically valuable estimate of inspiratory effort [21]. In this latter study, two end-expiratory occlusions were obtained in each patient, but repeatability was not reported. Furthermore, they report that despite changes in level of support, NMEoccl remained rather stable within individual patients, as supported by the linear relationship between Pmus and EAdi (r² = 0.78) [21]. This is in apparent contrast with our findings, as we found only moderate correlation between Pmus and EAdi (Fig. 2). However, differences in data analysis should be acknowledged. Bellani used the Pearson correlation coefficient to test the correlation between Paw and EAdi, in which all measurements from different patients were analyzed together as if they were from a single patient. However, the variability of between-subject measurements is different compared to the variability of the within-subject measurements [33]. Therefore, calculating the correlation coefficient of repeated observations, as in the current study, seems more appropriate [30, 31]. Indeed, Fig. 2 demonstrates that the slope of Paw and EAdi in the individual patient is highly variable. This is consistent with the results of Bellani, exhibiting high inter-individual variability [21].

The high NMEoccl repeatability reported in the current study precludes its application in clinical practice. Based on visual inspection of the EAdi waveforms, we proposed that suboptimal filtering and replacement of cardiac electric activity by the ventilator software are important. Several techniques for waveform analyses were applied, but did not improve repeatability, suggesting that in addition to suboptimal filtering other factors might be involved. Beck et al. showed that an increase in volume from functional reserve capacity (FRC) to total lung capacity (TLC) reduces Pdi by 60% for a given EAdi [34]. Similarly, muscle weakness may affect NMEoccl, but both are unlikely to explain the high variability in our study, given that all measurements were obtained in a time window of 5–10 min.

As none of the techniques for waveform analyses resulted in improved NMEoccl repeatability, a more pragmatic approach was explored that is also feasible in clinical practice. In thermodilution cardiac output measurements, three repeated measurements are averaged, provided that these values are within 10% of their average. If not, a total of five measurements are performed in which the lowest and highest value are eliminated [35]. In our study, this strategy reduced the influence of erroneous NMEoccl values, irrespective of the cause, and will result in a more reproducible value of NMEoccl.

Theoretically, the NMEoccl could be helpful to titrate ventilator support in patients on partially supported modes. NMEoccl is calculated by dividing ∆Paw by ∆EAdi during an end-expiratory occlusion. During an occlusion ∆Paw equals ∆Pmus and therefore NMEoccl can be obtained without direct measurement of Pmus (requiring an esophageal balloon). Rearranging this formula to Pmus = NMEoccl * EAdi allows calculation of Pmus breath by breath, after dividing this value by 1.5 to correct for differences in NME obtained under static and dynamic conditions [21]. Interestingly, NMEoccl may be used to evaluate respiratory muscle function over time. A decrease in NMEoccl indicates that the respiratory muscles are less efficient in converting electrical activity into pressure. Possible causes for variability in NMEoccl require further studies, but may include intrinsic positive end-expiratory pressure (PEEP) and impaired function of the contractile proteins.

In our study population there were no significant changes in NMEoccl over time (p = 0.75), which corresponds to the results of Bellani et al. [23]. However, it should be noted that during our study period the ventilatory settings were not fixed, which might explain why NMEoccl did not change over time.

In our study Pmus varied among patients but remained relatively stable in individual patients over time. In some patients estimated inspiratory effort was > 20 cmH₂O. An important question is whether a relatively high inspiratory pressure generated by the respiratory muscles may result in the development of contractile fatigue. A TTI ≥ 0.15 puts the diaphragm at risk of development of fatigue [36]. In our study, all patients except for two, with a Pmus < 12 cmH₂O, had a TTI < 0.15. This might suggest that titration of ventilatory support to a pressure < 12 cmH₂O could limit the risk of fatigue development. However, this has to be studied before it can be applied in clinical practice.

The strengths of our study are the high number of occlusions analyzed and the fact that at each time point five repeated occlusions were obtained. This allows thorough analysis of the repeatability of NMEoccl and provides methods to improve its variability under clinical conditions. In addition, several waveform analysis techniques were performed to evaluate the high NMEoccl variability; however, this did not improve the RC for NMEoccl. It was suggested that suboptimal filtering might be important. Software engineers should further improve ventilator software for EAdi signal filtering.

Several limitations should be acknowledged. First, patients in our study did not have an esophageal balloon in situ and therefore we could not validate our measurements against the gold standard. However, previous studies have shown excellent correlation between ∆Paw and ∆Pes during an occlusion maneuver [37‐39]. Second, our study was conducted in a single center and in a selected group of ICU patients. The generalizability of the findings needs to be assessed.