Automated patient-ventilator interaction analysis during neurally adjusted non-invasive ventilation and pressure support ventilation in chronic obstructive pulmonary disease

Adult patients with acute respiratory failure and a medical history of COPD, admitted to the ICU for non-invasive ventilation were eligible for inclusion in the study. Patients with a known neuromuscular disorder, severe hypoxemic failure (PaO₂/FiO₂ < 100 mmHg), or hemodynamic instability requiring high-dose norepinephrine (>0.5 μg/kg/min) were excluded. The study was approved by the ethics committee of the Radboud University Medical Center (NL33351.091.11) and is in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. All patients gave their informed consent prior to the study.

All patients undergoing NIV in our hospital receive a nasogastric tube to allow adequate feeding and prevent gastric hyperinflation. COPD patients undergoing NIV receive a nasogastric tube with a multiple array of electrodes placed at the distal end (NAVA catheter, 12 French; Maquet Critical Care, Solna, Sweden). Correct positioning was established by use of dedicated software. After enrollment and clinical stabilization, each patient received three 30-minute ventilation protocols in the following order:

BiPAP Vision uses the Auto-Trak Sensitivity algorithm to trigger and cycle off the ventilator and cannot be set individually. With NAVA, the back-up mode for triggering was set at flow triggering. In order to reduce the amount of leakage on ventilator performance, we chose to use a tightly fitted oronasal mask (Respironics PerforMax, Philips, Best, The Netherlands), a frequently used interface [21]. Switching between ventilators required modifications in measurement setup and short disconnection of the patient from the ventilatory circuit. In order to minimize discontinuation of assist, the order of interventions were not randomized.

At the end of each ventilator mode, respiratory discomfort was scored by use of a Visual Analog Scale (from 0 mm (no discomfort) to 100 mm (maximal imaginable discomfort)) and arterial blood gas analysis was performed from an indwelling arterial line.

Flow, airway pressure (Paw) and EAdi were acquired from the serial port of the Servo-I at a sampling rate of 62.5 Hz and recorded using dedicated acquisition software (Neurovent Research Inc., Toronto, ON, Canada). The BiPAP Vision does not have a data output port. Therefore, flow was acquired by placing a pneumotachograph (Fleisch no. 3, Phipps & Bird, Richmond, VA, USA) between the airflow port of the ventilator and its tubing, and Paw was acquired by placing a connection piece between the end of the tubing (after the leakage port) and the face mask, connected to a pressure transducer (range ±50 kPa, Freescale Semiconductor, Tempe, AZ, USA). Both Paw and flow were recorded at a sampling rate of 62.5 Hz and synchronized with the EAdi using dedicated acquisition software (Neurovent Research Inc., Toronto, ON, Canada).

Study parameters were calculated from a stable 5-minute period at the end of each mode on a breath-by-breath basis using a software routine developed for Matlab (Mathworks, Natick, MA, USA). Measuring tidal volume by expiratory flow integration is not precise in the presence of leaks, therefore, tidal volumes are not presented in the manuscript. Neural respiratory rate was calculated as the number of EAdi peaks/min.

Patient-ventilator interaction was evaluated by comparing Paw and EAdi waveforms with automated computer algorithms [20]. Trigger and cycle-off error (that is dyssynchrony) were calculated as percentages of neural inspiratory and expiratory time periods, respectively. Events where EAdi and Paw were completely dissociated (that is asynchrony), such as wasted efforts, auto-triggering, multiple assist during EAdi peak (double triggering) and multiple EAdi peaks during assist, were assigned 100% error. To estimate the overall extent of asynchrony and dyssynchrony, we calculated the NeuroSync index by averaging the percentage errors for all breaths.

The D'Agostino and Pearson test was used to test the normality of distribution. NIV-PSV_Vision, NIV-PSV_Servo-I and NIV-NAVA were compared using the Friedman test with Dunn's post hoc testing. Exponential regression analysis using a least squares fit was performed to test the relationship between the NeuroSync index and wasted efforts. A P value <0.05 was considered significant. Statistical analyses were performed with Graphpad Prism 5 (Graphpad Software, San Diego, CA, USA). Results are reported as median with interquartile ranges.

Results for breathing pattern and respiratory drive are presented in Table 3. EAdi amplitude was higher with NIV-PSV_Servo-I compared to NIV-PSV_Vision (P <0.05). Peak airway pressure and peak flow were higher with NIV-PSV_Vision (P <0.05) compared to NIV-NAVA and NIV-PSV_Servo-I.

Figure 1 depicts median values for trigger delays and cycling-off error during each mode for all individual patients. NIV-NAVA showed lowest trigger delay compared to NIV-PSV_Vision and NIV-PSV_Servo-I (P <0.0001). NIV-PSV_Vision and NIV-PSV_Servo-I had comparable trigger delays, but NIV-PSV_Servo-I showed more early cycling off (P <0.05). In absolute values, NIV-PSV_Vision (95 ± 22 ms) and NIV-PSV_Servo-I (91 ± 19 ms) showed more cycling-off error compared to NIV-NAVA (12 ± 6 ms); P <0.05.

Patient-ventilator interaction, calculated with the NeuroSync index, was significantly higher (larger error) with NIV-PSV_Vision (24 (interquartile range (IQR) 15 to 30) %) and NIV-PSV_Servo-I (21 (IQR 15 to 26) %) compared to NIV-NAVA (5 (IQR 4 to 7) %; P <0.001).

Figure 2 depicts the correlation between the number of wasted efforts and the NeuroSync index. The relationship shows as timing errors progressively increased with NIV-PSV_Servo-I and NIV-PSV_Vision a positive association with the number of wasted efforts, which was certainly more pronounced above 20% error.

For all three modes of NIV studied, Figure 3 shows a plot of the relative timing errors of triggering (Y-axis) versus the relative timing error for cycling off (X-axis), for every breath, in all patients. Based on the data from Figure 2, we have inserted a box suggesting `acceptable' synchrony to be ≤20% of neural timings, whereas larger errors (>20%) represent dyssynchrony.

In the form of a pie chart, Figure 4 plots the distribution of synchrony (≤20% error, that is inside the box in Figure 3), dyssynchrony (>20% error, that is outside the box in Figure 3), and asynchronies for each mode. Wasted efforts were the most prevalent type of asynchrony and differed between ventilator modes (P <0.001). Post hoc analysis indicated significantly more wasted efforts with NIV-PSV_Vision compared to NIV-NAVA. Other asynchronies, such as multiple EAdi during assist, double triggering and auto-triggering were uncommon.

There were no differences in blood gas values (Table 4) and respiratory discomfort with NIV-PSV_Vision (45 (IQR 31 to 69) mm), NIV-PSV_Servo-I (60 (IQR 41 to 65) mm), and NIV-NAVA (45 (IQR 33 to 75) mm).

For effective unloading of the respiratory muscles with NIV, the ventilator should cycle in synchrony with the patient's neural respiratory drive [5]. Our results are consistent with previous studies that showed improved patient-ventilator interaction with neurally compared to pneumatically controlled mechanical ventilation [15]-[18], however, several differences between these and the current study should be noted. First, we included only patients with COPD, which are more likely to exhibit poor patient-ventilator interaction [19]. Second, dedicated NIV-NAVA and NIV-PSV software was used instead of software for invasive ventilation in the previous studies [16],[17]. This is important as the software for invasive ventilation lacks leakage compensation thereby allowing auto-triggering at high leakage. Indeed, auto-triggering up to 6 breaths/min was found with NIV-NAVA using the invasive software [16], whereas we found only up to 1 breath/min. Third, a dedicated NIV ventilator was evaluated in the present study. In bench-test comparisons, including the ventilator used in our study, PSV delivered by dedicated NIV ventilators allowed better patient-ventilator interaction than ICU ventilators with NIV algorithms [10],[22]. Lastly, an automated analysis method for quantifying patient-ventilator asynchronies and the more subtle dyssynchronies was used [20], allowing a more objective detection of patient-ventilator interaction.

The present study showed a small trigger delay with NIV-NAVA, which substantially increased with NIV-PSV_Servo-I and NIV-PSV_Vision. These findings agree with previous work comparing NIV-PSV_Servo-I and NIV-NAVA [15], but oppose a previous bench test showing longer trigger delay for NIV-PSV_Servo-I compared to NIV-PSV_Vision[22]. NAVA triggers on the increase in EAdi and thus represents the duration to increase EAdi, to process the signal and to open the inspiratory valve. Our average trigger delay of about 50 ms with NIV-NAVA is in the range previously reported for NIV-NAVA [15],[16],[18]. In contrast, pneumatic triggering is more complex and considerably affected by leakage, which can only be partly compensated for by dedicated NIV algorithms [7].

Synchronized termination of assist is another key component to maintain good patient-ventilator interaction. As depicted in Figure 1, NIV-PSV_Vision showed large inter-subject variability in early and late cycling off, whereas NIV-PSV_Servo-I showed primarily early cycling off. Cycling-off error in NIV-NAVA was negligible, which could be anticipated since its definition for cycling off is similar to the algorithm used to quantify cycling-off error (70% of peak EAdi). These findings agree with previous suggestions that NIV algorithms for ICU ventilators tend to increase the incidence of premature cycling off [7].

Wasted efforts are inspiratory efforts not rewarded by ventilatory assist, which can increase the work of breathing [5],[6]. In the present study, 4.3% and 2.5% of inspiratory efforts were unnoticed by the ventilator for NIV-PSV_Vision and NIV-PSV_Servo-I, respectively. In contrast, NIV-NAVA effectively prevented wasted efforts, confirming previous studies [15],[16],[18]. Furthermore, as depicted in Figure 2, we found that wasted efforts increase drastically after timing errors reach 20%. This suggests that the limits of the NeuroSync index and the definition of `acceptable' synchrony should be kept below 20%, as indicated by the centered boxes in Figure 3.

EAdi in the present study was higher with NIV-PSV_Servo-I compared to NIV-PSV_Vision, which is difficult to explain. Lack of difference in blood gases or respiratory rates contradict that increased EAdi with NIV-PSV was ventilation related. Premature cycling off with NIV-PSV_Servo-I could be a probable cause for increased EAdi, since this results in unassisted inspiration in the last part of inspiration. It should also be noted that the design of the respiratory circuit and assist delivery of the BiPAP Vision is fundamentally different from the Servo-I. For example, the BiPAP Vision system has a large intentional leakage. Consequently, from Ohm's law it follows that higher flow is required to maintain the preset pressure level (Table 3). Higher flow might have resulted in higher CO₂ clearance in the interface and upper airways and a consequent reduction in dead space leading to reduced respiratory drive.

Good patient-ventilator interaction is one of the key factors for clinical success of NIV, thus solving poor patient-ventilator interaction in COPD patients is of potential clinical value. In our study, we demonstrate that progressive mismatch between timing of the patient's neural drive and the response of the ventilator is associated with increased number of wasted efforts. It is tempting to speculate that improving synchrony between patient neural effort and ventilator assist improves outcome in COPD patients, but it should be noted that our study is a short-term physiological study performed in a center with extensive experience in NIV, both with PSV and NAVA. In addition, a limitation of the present study is the limited number of patients, which hamper drawing generalized conclusions.

Differences in patient-ventilator interaction between ventilator modes did not affect blood gas values, in particular pH, and respiratory discomfort. In part, this results from the timing of study inclusion, after initial stabilization on NIV. At inclusion in the study, blood pH (around 7.38) was already increased, making it more difficult to detect changes in pH and respiratory discomfort caused by different ventilator modes. In this context it should also be mentioned that NIV modes were not performed in a random order. Nevertheless, we performed our measurement after initial stabilization on NIV making it unlikely that the patients' respiratory status was worse at the beginning of the study than at the end. Future studies, which randomize between NAVA and PSV at admission, are necessary to ascertain whether or not improved patient-ventilator interaction in the acute phase of NIV translates to better NIV outcomes.