The protocol was approved by Institutional Ethics Committee of Zhongda hospital (Approval Number: 2021ZDSYLL216-P01, September 1, 2021), and informed consent was obtained from the patients or next of kin. The trial “Predict Transpulmonary Pressure through ZAM”) was registered at clinicaltrials.gov (ID NCT05378802) on November 6, 2021. The procedures were followed in accordance with the ethical standards of the responsible committee on human experimentation (institutional or regional) and with the Helsinki Declaration of 1975.

Patients

Patients were eligible for inclusion if they were (i) 18–85 years old, (ii) in respiratory failure on invasive mechanical ventilation; (iii) able to tolerate pressure support ventilation, (iv) under light sedation with RASS between − 2 and 1.

Patients were not eligible if (i) there was a contraindication for nasogastric tube insertion (history of esophageal varices, gastroesophageal surgery in the previous 12 months or gastroesophageal bleeding, international normalized ratio > 1.5 and activated partial thromboplastin time > 44 s), (ii) any disease affecting spontaneous breathing.

Measurements

Patients were instrumented with a naso-gastric feeding-tube capable of measuring diaphragmatic electrical activity (Edi), esophageal (Pes) and gastric (Pga) pressures (Neurovent Research Inc, Toronto, Canada). The following measurements were recorded simultaneously: flow and ventilator pressure (PVent), Edi, Pes, and Pga. See Liu et al. [6] for more details on signal acquisition.

Protocol

The protocol (Fig. 1) consisted of the following steps:

(i)

Two NAVA periods (“Low” and “High” NAVA levels, 15 min each).

(ii)

Two NPSSIM periods (“Low” and “High” pressure support levels, 15 min each).

(iii)

Heavy sedation/paralysis.

(iv)

Step-wise increases in volume control (VC), and step-wise increases in pressure control (PC), every 20–30 s.

Fig. 1

Schematic representation of experimental protocol. After inclusion, subjects (n = 10) were randomized to initially receive either two, 15-min periods of NAVA at two NAVA levels (NAVA LOW and NAVA HIGH) or two, 15-min periods of NPSSIM at two pressures (NPSSIM LOW and NPSSIM HIGH) and then switched to the other mode. LAM breaths were performed manually (an acute lowering of the NAVA level to zero) every 3 min per 15-min ventilation period; inspiratory and end-expiratory holds were performed 1–3 times during the 15-min periods. Patients were then sedated/paralyzed and ventilated with stepwise increases in pressure control and volume control, with steps increased every 20–30 s. See “Methods” for more details

LAM breaths were performed during the NAVA and NPSSIM runs, at least 5 per 15-min period.

Patients were ventilated with a Servo-i ventilator (Maquet, Solna, Sweden). During spontaneous breathing modes, patients were studied under light sedation with RASS between − 2 and 1. The purpose of using “high” and “low” levels of assist was to ensure a wide range of respiratory efforts, and to demonstrate the validity of the new PL-estimate (PL_LAM) over a broader range.

NAVA was used conventionally as previously described [7, 8], and NPSSIM was provided using the NAVA mode at a maximum NAVA level (15 cmH2O/µV) with upper pressure-limit adjusted to target a PS level above PEEP [9,10,11]. The ventilator pressure levels studied are presented in the results (Fig. 4). There was no intention to compare the modes statistically, they were both included to test the PL_LAM estimate with different ventilator pressure profiles.

Lower Assist Maneuver (LAM): The LAM maneuver was performed by manually lowering the NAVA level to 0 cmH2O/µV, for one single breath [2]. Note that the Servo-i ventilator provides a minimal pressure of 2 cmH2O during the LAM, this is built-in to the machine and not adjustable.

Following the NAVA and NPSSIM runs, spontaneous breathing was eliminated by deeply sedating (RASS − 3 to − 4) with propofol (2–3 mg/kg h) and paralyzing by intravenous injection bolus of vecuronium bromide (2–4 mg). Control mechanical ventilation at different levels allowed us to obtain breaths for matching the spontaneous breathing titrations (in terms of flow and volume). Absence of Edi (close to 0 µV) confirmed that the diaphragm was not active during CMV.

Pressure control (PC) mode and volume control mode (VC) were used to “match” breaths of similar volumes/flow as obtained during the NAVA or NPSSIM runs (“matching” refers to later, off-line analysis, of assisted breaths and CMV breaths, see below “Analysis (off-line)”). In both modes, the assist was progressively increased in a step-wise fashion, every 20–30 s. On average for the group, 17 ± 4 steps in PC were performed between 5.5 ± 2.3 and 23.8 ± 6.5 cmH2O. In VC mode, on average 19 ± 6 steps were performed and ranged between 22.6 ± 6 and 50.7 ± 12.4 LPM.

Analysis (off-line)

As previously described [2], the analysis was performed off-line, after the experimental data was collected. Analysis was performed both automatically and manually. Figure 2 demonstrates waveforms obtained in one subject for three types of breaths during NPSSIM, one LAM maneuver, and CMV. Figure 3 shows another example of single-breath waveforms with further explanations of the analysis.

Fig. 2

Examples of recorded and calculated waveforms to demonstrate experimental protocol, breath types, and breath matching (assisted, LAM, and controlled mechanical ventilation). Waveforms are obtained from one representative subject breathing with NPSSIM. From top to bottom, PVent, Pes, PL_Pes, flow, volume, Edi. Green waveforms indicate assisted breaths, red waveforms LAM breaths, and blue waveform the CMV breath. Vertical dashed black lines indicate the start and peak of Edi waveform. Examples of matching Edi waveforms for assisted and LAM breaths are indicated, as well as the matching flow and volume for assisted breaths and the CMV breath. All signals were continuously collected from the Servo-tracker. PL_Pes was mathematically produced by digital subtraction of the Pes waveform from the PVent waveform

Fig. 3

Examples of recorded waveforms to demonstrate off-line analysis. Waveforms are obtained from one representative subject breathing with NAVA (left panels) and NPSSIM (right panels). Pressure panel: for both modes, the inspiratory portion of single breaths are displayed for PL_LAM (black solid); PL_Pes (yellow solid); PL_CMV (purple solid); PIVent (green solid), the latter calculated as the mathematical difference between PVent during assisted breath and PVent during the LAM. Volume panel: LAM breath (dashed blue line), assisted breath (blue solid line) and volume control breath (purple solid line) are plotted. PVBC was calculated as LAM volume/assisted volume (at peak Edi). For NAVA, assisted volume (blue solid) and matched volume control (purple solid) are superimposed to demonstrate matching of the breaths. For NPSSIM, volume during pressure control (purple solid) are superimposed on the assisted breath (solid blue). Flow panel: the corresponding flow waveforms are plotted for spontaneous modes (solid blue) and controlled modes (solid purple) and show examples of matching waveforms. Edi matching: Edi curves for LAM breaths (dashed blue) and assisted breaths (solid blue) were evaluated for their similarity (regression analysis with inclusion as defined in “Methods”). Vertical dashed red line indicates the time point of peak Edi, to where most variables are calculated, with the exception of PL_Pes, calculated to its peak (yellow arrow)

Automated analysis

From the Edi waveform, two time-points (start and peak of Edi, Fig. 2, most bottom panel, vertical dashed lines) for the assisted breaths and the LAM breaths were detected automatically. Edi onset and Edi peak (also shown in Fig. 3, bottom panels, vertical dashed red line) was obtained by finding the “state” of the ventilator (a digital signal collected from the Servo-I ventilator). Hence, the onset of ventilator pressure and inspiratory flow, as well as their values at peak Edi, were also automatically detected and stored.

Initial screening criteria for including LAM breaths and assisted breaths in the analysis were: tidal volume > 100 ml, and peak Edi > 1 µV. Regression analysis of (inspiratory) Edi between the LAM and the assisted breath must have had R2 > 0.8 and must have included a minimum of at least 192 ms. The slope values needed to be 0.7–1.4 and the intercept less than ± 2 µV. The assisted breaths that had an “Edi-matched” LAM were then stored for later comparison to the breaths obtained during the PC or VC mode.

Calculation of PL_LAM: For Edi-matched breaths, for each LAM period, the LAM-based estimate of PL was calculated as previously described by Liu [2]:

$$}\_} = }/\left( }^ } \right),$$

where PIVent = PVent (assist) − PVent (LAM), and PVBC = Volume LAM/Volume Assist [5, 6].

PL_LAM was calculated to peak Edi. All breaths with PVBCs up to 0.85 were included. When PIVent (numerator of the equation) was 1–5 cmH2O, PVBC up to 0.90 was permitted.

Calculation of PL_Pes: During assisted modes (NPSSIM/NAVA), the PL_Pes waveform was created by digital subtraction of Pes from PVent, examples are provided in Figs. 2 and 3 [3]. The start of the PL_Pes waveform was adjusted to zero for each breath (i.e., driving pressure). PL_Pes was calculated to its peak (see Fig. 3, yellow arrow indicates peak PL_Pes). The influence of expiratory muscles’ relaxation on inspiratory Pes was minimized by an algorithm based on Pes baseline values just prior to inspiration for the LAM and the assisted breaths. If Pes baseline difference was > 2 cmH2O, the samples were discarded. For PL_Pes and PL_LAM calculations, any negative deflection in airway pressure (PVent) immediately prior to the onset of assist, was added to both.

Manual analysis

Manual analysis was performed to validate the automated analysis. Comparing the automated analysis to the manual analysis also allowed the minimization of selection bias.

The steps in the manual analysis were:

(i)

Comparing LAM breaths and assisted breaths for matching Edi, by visual inspection of superimposed Edi curves for both breath types

(ii)

Comparing CMV breaths to selected assisted breaths (from above) for matched flow/volume curves.

More specifically, flow and volume from the different step-wise increases in CMV were superimposed and compared to the flow and volume curves from each assisted breath, until the best match was obtained visually (see Fig. 2). A new average was calculated for PL_LAM and PL_Pes for the matched assisted breaths. As with PL_Pes, PL_CMV was calculated as Pvent-Pes (above PEEP). PL_CMV was measured up to the volume where the peak Edi occurred during the matched assisted breath (see Fig. 3, see vertical red dashed lines). At least 10 breaths visually matching flow and volume curves were required.

Respiratory system compliance and resistance were calculated from the data collected during VC at the end of the study protocol, see Electronic Supplementary Material, Methods section.

Statistics

For comparing PL estimates to PL_CMV, and PL_LAM to PL_Pes, Bland–Altman plots were used, and bias is reported as well as the limits of agreement = 1.96 × standard deviation (1.96SD). Regression analysis was used, and we report determination coefficients (R2), slopes and intercepts. Statistical analysis was performed using Sigma Stat (v.10).