Introduction

In patients with acute respiratory failure, the placement of an endotracheal tube to allow mechanical ventilation is associated with significant complications related to local airway injury and increased risk of ventilator-associated pneumonia [1]. Although noninvasive mechanical ventilation delivered by pressure support ventilation (PSV) decreases the percentage of complications and is better tolerated than mechanical ventilation delivered by volume-cycled ventilators, poor patient-ventilator interaction has been advocated as a frequent cause of PSV failure [2, 3].

Proportional assist ventilation (PAV) is a recently developed mode of assisted ventilation where pressure is applied by the ventilator in proportion to the patient-generated volume and flow. In this mode, there is therefore automatic synchrony between the end of the patient's effort and the ventilatory cycle [2, 3, 4, 5]. This approach has several potential benefits, including decreased pressure requirements (and hence increased potential for avoiding intubation), continued engagement of patient's own control mechanisms, increased comfort, and decreased weaning problems [6, 7, 8, 9, 10, 11].

A few short-term studies in a limited number of patients with chronic respiratory insufficiency have shown that noninvasive PAV is able to improve arterial blood gases and to unload the inspiratory muscles, as well as to decrease the need of endotracheal intubation [12, 13, 14, 15, 16, 17]. In a recent pilot study to test the feasibility of treating acute respiratory insufficiency with PAV delivered noninvasively [18], mortality and intubation rates were similar in patients randomised to receive PAV or PSV, but refusal rate was lower, reduction in respiratory rate was more rapid, and there were fewer complications in the PAV group. In view of the advantages of PAV, a prospective randomised study was designed to compare the effects of PSV and PAV during noninvasive ventilation in the treatment of acute respiratory failure.

Materials and methods

Between March 2000 and December 2001, adult patients with acute respiratory failure who consecutively attended the intensive care unit (ICU) of Hospital General Universitario de Alicante (Spain), were eligible to participate in a prospective randomised study. The protocol was approved by the institutional review board, and written informed consent was obtained from all enrolled subjects or their next of kin. The study was conducted according to the Declaration of Helsinki.

Patients were included if they had acute respiratory failure with worsening dyspnoea and at least one of the following: (1) a respiratory rate 30/min; (2) use of accessory muscles or abdominal paradox; (3) hypercapnia [arterial carbon dioxide tension (PaCO2) >45 mmHg] with pH <7.35; and (4) hypoxaemia with arterial oxygen tension (PaO2) <55 mmHg or a ratio of PaO2 to fraction of inspired oxygen (FiO2) = 200.

Exclusion criteria were as follows: need for immediate intubation; loss of consciousness; inability to protect the upper airway or cleaning secretions; inability to cooperate; haemodynamic instability; and anatomic abnormality of the airways. Enrolled patients were randomised to receive either PAV or PSV delivered using the BiPAP Vision Ventilator (Respironics, Pittsburgh, Pa., USA). Randomisation was achieved using sealed envelopes. Patients were fitted a standard oronasal or facial mask (Respironics).

PSV was begun at an expiratory pressure of 4 cmH2O and an inspiratory pressure of 10 cmH2O gradually adjusted for a tidal volume (VT) = 6 ml/kg. For all patients receiving PAV therapy, standard initial settings included flow assist (FA) = 2 cmH2O·l·s, volume assist (VA) = 5 cm H2O/l and end-expiratory pressure of 4 cmH2O. Then, leaving FA unchanged, VA was increased slowly by steps of 2 cmH2O/l until the pattern of the "runaway" was seen [14, 15, 17]. "Runaway" is identified when the inspiratory phase of the ventilator extends well beyond the inspiratory time observed at lower levels and goes into the neural expiration. This occurs when the pressure generated by the ventilator during patient's inspiration is greater than that needed for compensating patient's elastic resistances. At this point, VA is equivalent to the patient's elastic resistances. It is important to establish a VA lower than the "runaway" value in order to synchronise that the inspiratory phase of the ventilator will terminate when the patient's inspiratory effort will finish. We considered a level of VA corresponding to 80% of the value of VA at the last step preceding the "runaway" to provide sufficient assistance without the risk of overassistance and disadaptation. Whereas the VA "runaway" occurs at end-inspiration, the FA "runaway" occurs early in the inspiration and this is more difficult to recognize with certainty [15].

Following the method proposed by Patrick et al. [17] and starting from FA of 2 cmH2O·l·s, step-by-step increases of 1 cmH2O·l·s were applied. At each step, the patient was asked whether he/she felt more comfortable and, in case of affirmative responses, increases in FA continued until the level at which the patient felt less comfortable. FA was then programmed at 80% of the level corresponding to the previous step. Final end-expiratory pressure parameters and FiO2 were adjusted to obtain an arterial oxygen saturation (SaO2) = 88%. Assisted ventilation was maintained during the first 24 h until an improvement in oxygenation and clinical condition occurred. Then, assisted ventilation was progressively reduced according to the degree of clinical improvement, by decreasing respiratory support or by initiation of periods of oxygen therapy without assisted ventilation. In hypoxaemic respiratory failure, noninvasive ventilation was withdrawn when the patient maintained a respiratory rate <30 breaths/min with PaO2 ≥ 75 mmHg and FiO2 = 0.5 in the absence of ventilatory support. In hypercapnic respiratory failure, noninvasive ventilation was withdrawn when the patient maintained stable arterial blood gases with pH >7.35 after 24 h of discontinuation of assisted ventilation.

Outcome measures

Major outcome measures were mortality and need of intubation of the trachea. Criteria to decide intubation included respiratory arrest or gasping; inability to protect the upper airway due to Glasgow coma score <9; inability to cough and cleaning abundant secretions; inability to maintain PaO2 = 65 mmHg with FiO2 = 0.6 in patients with hypoxaemic respiratory failure without hypercapnia; inability to maintain PaO2 = 45 mmHg, pH = 7.20 or worsening of pH compared to the initial value in patients with hypoxaemic respiratory failure and hypercapnia; haemodynamic instability (systolic blood pressure <70 mmHg); and psychomotor agitation preventing nursing care. Intolerance to noninvasive ventilation was defined as the patient's refusal to continue ventilation mode because of discomfort caused by facial mask (pain, cutaneous ulcer, claustrophobia) and/or discomfort related to ventilatory pressure despite repeated attempts at gaining acceptance. Failure of noninvasive ventilation was defined as need of intubation and/or mortality.

Secondary outcome measures included changes of heart rate, respiratory rate, systolic blood pressure (SBP), arterial blood gases, and PaO2:FiO2 1–24 h postinitiation, as well as subjective responses including dyspnoea and comfort and hospital lengths of stay. Subjective indices (dyspnoea and comfort) were measured using a visual analogue scale (1 least–10 most). ICU and hospital lengths of stay were also recorded.

Data were collected systematically at the beginning of noninvasive ventilation and at 1 h, 8 h, and 24 h. At these intervals, after carefully adjusting the mask to prevent leaks, dyspnoea and comfort scores were registered (as a single measure); peak inspiratory pressure and tidal volume were the mean for 1 min. During the remaining time periods, decisions regarding re-assessment of patients and the necessity to change setting were taken by the physician's in charge.

Statistical analysis

Although at the time of designing the study it was already known that there was no possibility of decreasing type II errors in the detection of differences in the need for endotracheal intubation and mortality, to detect a difference of 2 points in subjective indices of dyspnoea and comfort, 55 patients in each arm were required considering an alpha error of 5% and a statistical power of 80%. The SPSS (version 8.0) computer package was used for statistical analysis. Mean and standard deviation (SD) were calculated for continuous variables and distribution of proportions for categorical variables. The Student's t-test and the chi-square test were used for the comparison of quantitative and qualitative variables, respectively. The analysis of variance (ANOVA) was used for within-group comparisons. Need for endotracheal intubation and survival were calculated by the Kaplan-Meier method. Survival and avoidance of intubation curves were compared with the log-rank test. Differences were considered significant when P <0.05.

Results

Of 269 patients who were assessed for eligibility, 130 patients were randomised (65 PSV, 65 PAV). The remaining 139 patients were excluded because inclusion criteria were not met (n = 133) or because of patient's refusal to participate in the study (n = 6). In addition, six patients assigned to the PSV group and seven to the PAV group required immediate intubation and did not receive the assigned mode of ventilatory support. Therefore, 117 patients (59 PSV, 58 PAV) were enrolled in the study (Fig. 1).

Fig. 1.
figure 1

Flow-chart of the study

As shown in Table 1, baseline characteristics were similar in both groups except for a significantly higher respiratory rate in the PAV group (37±6 breaths/min) than in the PSV group (35±6 breaths/min) (P = 0.05). Ventilatory parameters in both groups are shown in Table 2. In patients assigned to the PAV group, peak inspiratory pressure was significantly lower at 8 h and 24 h, whereas tidal volume was significantly higher and end-expiratory pressure significantly lower at 1 h, 8 h, and 24 h.

Table 1. Baseline characteristics and principal diagnoses contributing to respiratory failure in patients undergoing noninvasive mechanical ventilation with pressure support ventilation (PSV) mode or proportional assist ventilation (PAV) mode. Data as n (%) or mean±SD
Table 2. Mean settings for PSV and PAV. Data as mean±SD

With regard to respiratory rate and gas exchange, the arterial oxygen saturation (SaO2), pH, PaCO2, and PaO2:FiO2 in the two groups were similar throughout the first 24 h, but in each group, statistically significant differences between baseline values and measurements during the 24-h study period were observed (Table 3). Moreover, between-group comparisons for SaO2 showed significant differences in favour of the PAV group at 1 h, 8 h, and 24 h of noninvasive ventilation.

Table 3. Haemodynamic and respiratory parameters during the first 24 h of noninvasive ventilation with pressure suppor ventilation (PSV) mode or proportional assist ventilation (PAV) mode. Data as mean±SD

All patients showed a significant improvement in dyspnoea with any mode of noninvasive ventilation (Fig. 2), but at the end of the study, statistically significant differences in VAS scores between PSV (5.17±4.98) and PAV (3.38±2.34) groups were not found. With regard to comfort (Fig. 3), VAS scores were significantly higher in the PAV group than in the PSV group (24-h values, 6.73±1.76 vs 2.94±1.21, P <0.001). On the other hand, patients in the PAV group rated their level of comfort steadily higher, whereas in the PSV group, level of comfort was rated progressively lower (Fig. 3).

Fig. 2.
figure 2

Dyspnoea improvement on the 0–10 visual analogue scale (VAS). Comparison of PSV and PAV modes at start of non-invasive ventilation (baseline) and at 1 h, 8 h, and 24 h (Student's t test, NS). Start of noninvasive ventilation vs 24 h for within group comparisons (PSV and PAV) (ANOVA, *P <0.001)

Fig. 3.
figure 3

Comfort on the 0-10 visual analogue scale (VAS). Comparison of PSV and PAV modes at start of non-invasive ventilation (baseline) and at 1 h, 8 h, and 24 h (Student's t test, *P <0.001). Start of non-invasive ventilation vs 24 h for within group comparisons in the PSV mode (ANOVA, ++ P <0.001) and PAV mode (ANOVA, NS)

Complications occurred more frequently in the PSV group than in the PAV group (nasal ulcer, 27% vs 12%, P = 0.04; conjunctivitis, 8.5% vs 0%, P = 0.06). Four patients (6.8%) in the PSV mode and one in the PAV mode presented abdominal distension (P = 0.36).

Intolerance to noninvasive ventilation occurred in nine (15.2%) patients (COPD exacerbation six, pneumonia three) in the PSV group and in two (3.4%) patients (COPD exacerbation two) in the PAV group (P = 0.03). Patient's refusal occurred within the first hour in two PSV patients, between 1 h and 8 h in five PSV patients, and between 8 h and 24 h in the remaining PSV patients and in both PAV patients.

There were no statistically significant differences between both groups regarding the need of endotracheal intubation (37% vs 34%), ICU mortality rate (22% vs 27%), and in-hospital mortality rate (29% vs 28%). Intubation was required because of haemodynamic instability (two2 patients in each group), persistence of respiratory insufficiency with increase in dyspnoea (17 PSV, 16 PAV), and inability to clean secretions (three PSV, two PAV).

Avoidance of intubation was 62.7% in the PAV group and 65.5% in the PSV group (χ2 test, P = 0.85). As shown in Fig. 4, the Kaplan-Meier curves for avoided intubation were similar in both groups (log-rank test, P = 0.69). The percentage of survivors was similar in both groups (72.4% PAV, 71.2% PSV) and, as shown in Fig. 5, the Kaplan-Meier survival curves were not significantly different (log-rank test, P = 0.87). Primary causes of death were progressive respiratory insufficiency (nine in each group), septic shock (two in each group), and heart failure (three PSV, four PAV). Three patients in the PSV group and one patient in the PAV group subsequently died during the hospitalization.

Fig. 4.
figure 4

Kaplan-Meier analysis for avoided endotracheal intubation in PSV and PAV modes

Fig. 5.
figure 5

Kaplan-Meier survival curves for PSV and PAV modes

The mean length of ICU stay was 8.6±9.5 days for the PSV group and 8.9±1.2 days for the PAV group (P = NS). The corresponding figures for duration of hospitalisation were 21±22 days vs 24±19 days, respectively.

Discussion

Noninvasive ventilation delivered by PSV can avoid endotracheal intubation in acute respiratory insufficiency due to different causes [19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31]. Although PSV provides greater comfort to patients than volume-cycled ventilators [32, 33], patient-ventilator dyssynchrony is a frequent cause of failure in both modes of ventilatory support. Studies in patients treated with conventional invasive mechanical ventilation delivered by PAV have shown that this mode is able to reduce work of breathing [4, 5, 6, 7, 8, 9], with the potential of enhancing patient-ventilator synchrony, since PAV provides inspiratory flow and pressure in proportion to the patient's breathing effort, so that, when properly adjusted, delivery of inspiratory assistance terminates with cessation of inspiratory effort. In addition, PAV is able to adjust to changing patient ventilatory demands [2, 4, 5, 7, 8]. Considering its potential advantages, we performed this study to test the hypothesis that noninvasive ventilation in patients with acute respiratory failure delivered by PAV would be associated with a greater tolerance and patient comfort (in relation to enhanced patient-ventilator synchrony) than noninvasive ventilation delivered by conventional PSV. The comparison between the two modes of ventilation was performed in a population of ICU adult patients with acute respiratory insufficiency.

The study population included 117 patients, who were randomised to receive PAV or PSV via the Vision ventilator. In accordance with the study of Gay et al. [18], no differences in intubation or mortality rates for PAV or PSV were observed, although the number of patients randomised to PAV or PSV in that study was half that of our study, in which both modes of ventilatory support were delivered using different devices.

In our study, like others [16, 18, 34], noninvasive PAV improved patient comfort more effectively than PSV. In addition, patients in the PAV group rated their level of breathing comfort constantly higher over the 24-h period, as opposed to patients in the PSV group who rated level of comfort significantly and progressively lower. On the other hand, intolerance was greater in the PSV group and occurred earlier than in the PAV group, although the intubation rates were virtually identical between both groups. This apparently contradictory finding may be explained by the distribution of patients at the time of randomisation, with initially higher mean respiratory rate—an important predictor of failure of noninvasive ventilation [21]—in the PAV group. Despite the fact that more patients in a severe condition were allocated to PAV, the tolerance and comfort rating was better in the PAV group as compared to PSV.

Non-comparative studies of noninvasive ventilation with PAV, although in a small number of patients, showed improvements in blood gases and in the load of inspiratory muscles [12, 13, 14, 15]. In a recent study, Gay et al. [18] also reported that respiratory rate dropped more in the PAV group than in the PSV group suggesting that PAV may have lowered work of breathing more rapidly than with PSV. However, in a group of 12 patients with chronic obstructive pulmonary disease and hypercapnic acute respiratory failure [34], noninvasive ventilation with PAV was able to unload inspiratory muscles similarly to PSV but breathing comfort was significantly improved in the PAV mode. In the present study, decrease of respiratory rate was similar in both groups, which may indicate that PAV can be as effective as PSV for improving work of breathing. The improvements obtained in the remaining respiratory parameters and degree of dyspnoea were also similar in both groups; however, end-expiratory pressure required to achieve adequate oxygenation was significantly lower in the PAV group compared to the PSV group.

The present findings should be interpreted taking into account some limitations of the study. With respect to the sample size, to detect a 5% difference in major outcome parameters (rate of endotracheal intubation and mortality) with an alpha error of 5% and a statistical power of 80%, 1,000 patients in each arm would have been needed. Therefore, calculations were based on the number of patients required to detect a difference of 2 points in the dyspnoea and comfort scores. A possible information bias may occur in studies that are not masked, but the design of a comparison of two modes of noninvasive ventilation prevented blinding of the intervention. Measurement of dyspnoea and comfort scores as well as ventilator settings were performed by ICU staff physicians at different shifts. Although strict criteria for adjusting ventilatory parameters and decision-making regarding endotracheal intubation had been previously established, some degree of variability amongst physicians might be possible. Finally, patients with acute respiratory failure due to different causes were included, but assessment of the usefulness of PAV in a larger number of patients stratified by pathological conditions is an interesting proposal for further prospective studies.

In summary, in patients with acute respiratory failure that should be mechanically ventilated, noninvasive ventilation delivered by PAV was able to achieve respiratory stabilisation. Although PAV seems more comfortable and intolerance occurred less frequently, no major differences exist in terms of physiological improvement or in terms of outcomes when comparing PSV and PAV.