INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Keywords: acute lung injury; mechanical ventilation; variable ventilation

Positive pressure ventilation is an essential component of supportive therapy among patients with acute lung injury (ALI). However, it has also been shown to contribute to worsening of pre-existing injury when administered by protocols that impart high levels of stress to lung parenchyma (1). Numerous studies have examined novel ventilation protocols designed to prevent ventilator-induced lung injury (VILI) while improving gas exchange. Available data suggest that injury can be propagated when ventilation is administered by using pressures at either extreme of the pressure-volume (P-V) curve of the lung (2). The current accepted approach used to manage patients with ALI applies sufficient positive end-expiratory pressure (PEEP) to prevent airway collapse, in conjunction with small tidal volumes (VT) to minimize alveolar "overdistention." Several clinical trials suggest that this "open lung ventilation" strategy decreases mortality in patients with acute respiratory distress syndrome (ARDS) compared with previously accepted large tidal volume ventilation strategies (8).

Alternatives to open lung ventilation that seek to limit alveolar overdistention have been introduced with promising results. Lefevre and coworkers (9) observed that randomly varying the delivered breath rate and VT with each breath significantly increased arterial blood oxygen tension (Pa_O2) and lung compliance without causing an increase in mean airway pressures in a porcine model of ALI. Mutch and coworkers (10, 11) later demonstrated similar observations in porcine models of lung injury. These investigators speculated that the apparent improvements in Pa_O2 may be a result of their ventilation mode mimicking the spontaneous variability in physiologic rhythms that is invariably removed during mechanical ventilation.

Suki and coworkers (12) have developed a mathematical model that suggests that the physiological improvements following application of ventilator support with a randomly varying breath pattern are a consequence of augmented alveolar recruitment. Intermittent application of larger breaths during variable ventilation opens collapsed alveoli and increases net lung volume by exploiting the nonlinear nature of the P-V curve of the atelectatic lung. In addition, Suki and coworkers (12) also predicted that the amount of variability or "noise" placed about the peak inspiratory pressure would act on blood oxygenation in a manner analogous to a phenomenon called stochastic resonance (13). The essence of stochastic resonance is that when noise is applied as input to a nonlinear system the response is first improved, but then, as the amplitude of the noise increases further, it produces a detrimental effect (13). With regard to mechanical ventilation, this observation specifically suggests that noise added to mechanical ventilation can be adjusted, or "tuned," in such a way as to provide an optimal level of physiological improvement.

Experimental studies have suggested that application of variability during mechanical ventilation is not uniformly advantageous, however. In an oleic acid model of ALI, Nam and coworkers (14) demonstrated that variability had no beneficial effects over conventional ventilation. This observation suggests that factors other than variability per se can be important for conferring beneficial effects on lung physiology in this setting. Accordingly, we hypothesized that the ability of variable ventilation (VV) to improve lung mechanics and oxygenation in the setting of acute lung injury depends on the amount of variability added to VT as well as how peak inspiratory pressure is positioned relative to the lower and upper knees of the P-V curve. To test this hypothesis we compared lung mechanics as well as blood oxygenation in a rodent model of ALI during conventional volume-cycled ventilation (CVV) and VV.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Animal Preparation

Male Hartley guinea pigs (n = 14; weight, 500-600 g; Charles River Laboratories, Needham, MA) housed in a virus-free facility were studied within 2 wk of arrival under an institutionally approved protocol. To induce injury, each guinea pig was treated with endotoxin (15) as follows: 2.5 mg of Pseudomonas serotype 10 endotoxin (Sigma, St. Louis, MO) was dissolved in 5 ml of normal saline, suspended in saline by vortexing, and then sonicated for 25, 3-s cycles with a microtip sonicator (Heat Systems-Ultrasonics, Farmingdale, NY) at maximal microprobe tip power. The solution was administered to the animals via a DeVilbiss acorn nebulizer in a sealed chamber, using continuous flow at 6 L/min over a 15-min interval.

Twelve hours after the exposure, guinea pigs were anesthetized with intraperitoneal xylazine (5 mg/kg) and sodium pentobarbital (40 mg/kg), tracheally cannulated, and had a carotid arterial line placed for blood pressure and blood gas measurements. Animals underwent a partial midsternal thoracotomy to facilitate measurement of lung mechanics. Animals were placed on a FlexiVent rodent mechanical ventilator (SCIREQ, Montreal, Quebec, Canada) and initially ventilated on room air at 60 breaths/min, constant flow inspiration, VT = 5.1 ml/kg, inspiratory:expiratory (I:E) ratio of 1, PEEP = 3 cm H₂O, and fractional inspiratory oxygen (FI_O2) of 0.22.

Protocol

After surgery, 10 animals (Group I) underwent a 30-min control period with CVV at the above-described settings, followed by 5 test periods of 30 min of VV with different distributions of VT. At the conclusion of each control period, a blood sample of 0.2 ml was taken for blood gas analysis and lung mechanics were measured. An additional four animals (Group II) were ventilated for 3 h, using only CVV at the above-described settings. Lung mechanics were measured at 30-min intervals, whereas blood gases were taken at 1-h intervals. In both groups, blood volume removed for measurement of arterial oxygenation was replaced by phosphate-buffered saline, and airway and blood pressure were continuously monitored.

The order of administration of the different VV distributions was randomized to eliminate any systematic errors secondary to protocol design. Three animals from the group had quasistatic P-V curves recorded at the conclusion of each experiment.

Variable Ventilation Design

Each period of VV was designed to provide a different range of VT adjusted to achieve the desired target variability level. The widths of the VT distributions were set by assigning a random sequence of VT values taken from a uniform probability distribution falling between ± 0, 10, 20, 40, or 60% of the mean VT. Thus, a ventilation period with 0% variability represents an additional period of CVV, whereas a 60% level of variability means that, had the animal been receiving a mean VT of 1 ml, it would have an equal probability of receiving a VT anywhere between 0.4 and 1.6 ml with each delivered cycle. The mean VT and I:E ratio were set equal to those used during the control period and breath rate was adjusted with each cycle to ensure that the animal received a constant minute ventilation throughout each test period. Any changes in gas exchange or lung mechanics could thus be attributed solely to the mode of ventilation. Figure 1 shows one example of VT and frequency for the first 60 cycles of VV scaled to 60% variation in VT with a mean of 3.8 ml. The total number of VT delivered during a 30-min ventilation period was 1,800 cycles, ensuring a sufficient sampling of the uniform distribution of VT.

View larger version (34K): [in this window] [in a new window]   Figure 1.   Delivered VT (top) and frequency (bottom) during the first 60 cycles of V V scaled to 60% variability and mean VT of 3.8 ml. In this example, each delivered VT fell between 1.52 and 6.08 ml and had an equal probability of having any values within the specified range.

Lung Mechanics

Lung mechanics were assessed by the optimal ventilator waveform technique, a forced oscillation method that uses an input volume consisting of six sine waves with frequencies between 0.5 and 14.25 Hz (15). These have been selected to minimize the nonlinear interactions at the input frequencies (16). The amplitude of the waveform was adjusted to match the VT delivered during CVV.

The fast Fourier transforms of pressure and flow were calculated on overlapping segments of data, and lung impedance was determined as the ratio of the cross-power spectrum of pressure and flow and the auto-power spectrum of flow. Using an optimization algorithm (17), the impedance spectra were fit to a model to estimate airway resistance, airway inertance, tissue damping, and tissue elastance (H) as proposed by Hantos and coworkers (18).

Statistical Analysis

All data are presented as means ± standard deviation. Data were analyzed by one-way analysis of variance (ANOVA) for repeated measure (SigmaStat, San Rafael, CA), and differences between groups were considered statistically significant at p < 0.05.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Table 1 shows the population averages of mean airway pressures, mean peak airway pressures, and mean arterial blood pressures of Group I during testing. There were no significant differences between ventilation modes for any measured parameters. Mean airway pressures resided near 6 cm H₂O for each level of variability, and mean peak inspiratory pressures were between 11 and 13 cm H₂O. Mean arterial blood pressures ranged between 41 and 49 mm Hg.

Figure 2 shows a representative example of measured data and the model fit of resistance (real{impedance}) and elastance (2 × frequency × imag{impedance}) for one animal after a ventilation period with CVV and 60% VV. During both conventional and variable ventilation, resistance decreased with increasing frequency until it reached a plateau value at airway resistance. Elastance values increased at low frequencies until reaching a peak near 5 Hz, at which point values began to decline due to inertial effects. Elastance is approximately 30% higher during CVV than during 60% VV. Low-frequency elastance is closer to static elastance and, therefore, should be inversely related to recruited lung volume. Hence, any decrease in low-frequency elastance during VV, compared with CVV, reflects recruitment of collapsed alveoli.

View larger version (15K): [in this window] [in a new window]   Figure 2.   Examples of resistance (real{impedance}) (top) and elastance (2 × frequency × imag{impedance}) (bottom) of the measured lung impedance (symbols) and model fits (lines) during CV V (open circles, dashed lines) and 60% V V (closed circles, solid lines) as a function of frequency.

Figure 3 shows the tissue elastance parameter (H), arterial oxygenation (Pa_O2), and carbon dioxide tension (Pa_CO2) as a function of the amplitude of variability for two representative Group I animals. Responses among Group II animals ventilated with CVV are also summarized. Previously, we have shown that healthy guinea pigs receiving conventional ventilation have H values of 1.5 ± 0.4 cm H₂O/ml, Pa_O2 values of 82.1 ± 5.5 mm Hg, and Pa_CO2 values of 46.2 ± 1.6 mm Hg (15). By contrast, Group II animals with ALI had baseline H values of 2.4 ± 0.3 cm H₂O/ml, Pa_O2 values of 57 ± 9.8 mm Hg, and Pa_CO2 values of 35 ± 5.5 mm Hg. After 4 h of CVV, H increased to 4.9 ± 1.3 cm H₂O/ml (p < 0.05), Pa_O2 declined to 27 ± 3.8 mm Hg (p < 0.05), and Pa_CO2 rose to 51 ± 9.5 mm Hg (p < 0.01), clearly demonstrating a progressive worsening of lung mechanics and gas exchange.

View larger version (57K): [in this window] [in a new window]   Figure 3.   (A and B) Measured tissue elastance parameter (H), PaO2, and PaCO2 as a function of percent variability in VT, and equivalent time scale, in two different guinea pigs from Group I. Each column represents a measurement taken at the conclusion of a 30-min ventilation period with the indicated level of variability. The 0% variability represents an additional period of CV V. Solid lines represent the population averages of H, PaO2, and PaCO2 of Group II over the same time course.

The protocol used in Group I animals allows a direct comparison between CVV and VV. Figure 3A shows that after 30 min of conventional ventilation in a representative animal with ALI, H was 3.7 cm H₂O/ml, Pa_O2 was 36.3 mm Hg, and Pa_CO2 was 38.8 mm Hg. Application of VV using a 40% variability in VT resulted in a decrease in H to 3.3 cm H₂O/ml and an increase in Pa_O2 by nearly 15 mm Hg. After reducing variability to 20%, there was a subsequent increase in H to 3.9 cm H₂O/ml, and reduction in Pa_O2 to 32.5 mm Hg. The beneficial effects of 60% VV were considerably more dramatic. H decreased by more than 30% to 2.5 cm H₂O/ml (about half the average H from Group II), and Pa_O2 increased by 25% to 46 mm Hg. These beneficial effects were subsequently lost with resumption of CVV.

Figure 4 summarizes responses for H, Pa_O2, and Pa_CO2 as a function of percent variability in VT. After the initial period of CVV, H was 2.8 cm H₂O/ml and decreased to 2.7 cm H₂O/ml after ventilation at 10% variation in VT. Increasing variability to 20% resulted in little change in H, but increasing the amount of variability to 40 and 60% significantly decreased tissue elastance to 2.4 and 2.0 cm H₂O/ml, respectively (p < 0.05). After a second period of CVV, H increased to 2.7 cm H₂O/ml, convincingly demonstrating a beneficial response to variability.

View larger version (22K): [in this window] [in a new window]   Figure 4.   Population averages (mean ± SD) for tissue elastance H, PaO2, and PaCO2 as a function of percent variability in VT. *,#Statistically significant difference (p < 0.05) between the measured values after ventilation at a given level of variability and after the initial or the second period of CV V, respectively.

During CVV, the mean Pa_O2 was 42 mm Hg. Application of 40% variation in VT caused a significant increase in mean Pa_O2 to 54 mm Hg (p < 0.05). Increasing variability further to 60% had no additional beneficial effect. A return to CVV lead to a significant decline in Pa_O2 to values similar to those observed during the first CVV test period. Mean Pa_CO2 remained relatively unchanged at each level of variability throughout the 3-h experiment.

Figure 5 summarizes the distributions of peak inspiratory pressures for the animal shown in Figure 3A during each 30-min test ventilation period. The arrows denote the average peak airway pressure for each ventilation period. Peak airway pressures during CVV ranged from 9 to 13 cm H₂O with an average value of 11.1 cm H₂O. Use of 10% VV produced a wider distribution of pressures with a mean peak pressure of 12.9 cm H₂O. Ventilation at 20, 40, and 60% VV resulted in progressively wider distributions of peak inspiratory pressures with means of 12.7, 11.4, and 11.0 cm H₂O, respectively.

View larger version (41K): [in this window] [in a new window]   Figure 5.   Histograms of peak pressure distributions for each ventilation period shown for the animal represented in Figure 3A. The y axis represents the percentage of peak inspiratory pressures that fell within each bin, and the arrows indicate the mean peak inspiratory pressure delivered during that ventilation period. Each histogram on average represents greater than 1,000 cycles.

Figure 6 compares the inflation limbs of the P-V curves obtained in three endotoxin-exposed animals, (A, B, and C) after completion of the experimental protocol. In the normal guinea pig lung, lung volume tends to increase in a linear fashion until the curve plateaus near the upper knee of the P-V curve. After endotoxin treatment, the P-V curves display a different pattern. Volume tends to increase linearly until an inflation pressure of approximately 19 cm H₂O is reached. At that point, the slope of the P-V relationship changes abruptly, such that volume tends to increase more rapidly as a function of pressure, indicating an increase in compliance above a specific threshold. This "threshold recruitment pressure" was readily distinguishable in each of the ALI animals tested.

View larger version (18K): [in this window] [in a new window]   Figure 6.   Inspiratory pressure-volume relationship of an intact, healthy guinea pig lung inflated from 0 to 12 ml lung volume (top) and three endotoxin-exposed guinea pig lungs (A, B, and C ) inflated from 0 to 12 ml lung volume.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The primary findings of this study are that (1) VV improved lung mechanics over CVV at nearly every degree of variability and (2) the most significant increase in oxygenation over CVV occurred when variability in VT was 40%. Comparison of the lung function parameters obtained during VV in Group I animals with those obtained in Group II animals ventilated with CVV (Figures 3 and 4) demonstrates that changes in lung mechanics and blood oxygenation varied with the specific ventilation mode and not with the duration of ventilation. Thus, VV was able to considerably slow down the deterioration of lung function with time.

The beneficial effects of variability reported in this study are similar to those previously described by Lefevre and coworkers (9) and Mutch and coworkers (10). In these studies, breath rate was randomly varied about a given value and VT was adjusted to keep minute ventilation constant during oleic acid-induced lung injury (9, 11) and unilateral lung collapse (10). In both cases, a single level of variability was applied and resulted in a significant increase in lung compliance and Pa_O2 in comparison with CVV. The present study was designed to be more mechanistic, however, in that it seeks to characterize not only the magnitude of physiological benefit resulting from application of VV during ALI, but also the physiological basis for benefit. The animal model employed in this study causes biological and physiological changes that mimic those of human ARDS. As previously reported by Mora and coworkers (15), nebulized endotoxin results in neutrophil influx into the alveolar space, capillary leakage, increased tissue elastance, impaired gas exchange, and surfactant dysfunction, all characteristics observed in human patients during respiratory distress. We used volume-cycled ventilation applied to simulate open lung ventilation (VT = 5.1 ml/kg) in accordance with the current convention that large-volume excursions are a probable source of ventilator-induced lung injury (19). To minimize the risk of barotrauma, we applied a relatively small PEEP (3 cm H₂O), which ensured that the larger tidal excursions during VV would not result in abnormally high airway pressures.

To examine how variability in ventilation acts to recruit atelectatic lung and improve oxygenation, we explored the relationship between the lower knee of the P-V curve measured during quasistatic inflation, and the peak airway pressures imparted at different levels of variability. Table 2 summarizes the percentage of breaths that exceeded the lower knee during each 30-min ventilation test period and the effect that each ventilation period had on H and Pa_O2. In nearly every test period of VV where none of the delivered peak inspiratory pressures surpassed the lower knee of the P-V curve (10%, 20%, and the second period of CVV), there was an increase in H and a decrease in Pa_O2. Conversely, when peak pressures during VV exceeded the lower knee of the P-V curve there was a decrease in H and an increase in Pa_O2 in every instance. These findings strongly support the notion that application of pressures that exceed those present at this lower knee can, in fact, result in significant and sustainable recruitment during mechanical ventilation without requiring high airway pressures or large VT. It is also interesting to note that even during the periods of 40 and 60% VV, the lower knee was not exceeded in more than 8% of the cycles. This suggests that these lung regions, once recruited, remain open.

Although these observations support the analytical model of Suki and coworkers (12), other findings clearly indicate that the physiological impact of VV on the lung is more complex than can be explained by this simple model. For example, the degree of variability associated with the largest improvement in lung elastance did not always correspond to that at which gas exchange was optimized. Every Group I animal demonstrated maximal reductions in H during 60% variability, whereas many animals demonstrated optimal oxygenation with 40% variability. This may be a consequence of several factors that could result in a decoupling of improvement in H from that of Pa_O2. Whereas larger airway pressures tend to open collapsed regions of the lung, and lower H, the presence of extensive heterogeneities may promote simultaneous alveolar overdistention in less damaged regions, and consequent hypoventilation with ventilation-perfusion (/) mismatching. Intermittent large pressures could also adversely affect venous return, lowering mixed venous Pa_O2 and thus decreasing Pa_O2 in the setting of pre-existing shunt. This effect may be a "physiological" consequence of the application of large amounts of variability. Specifically, while application of 60% VV increases the percentage of large breaths, it also increases the percentage of small, and potentially ineffective breaths. This may further worsen / mismatch despite improving intrapulmonary shunt. Independent of the precise mechanism, these data suggest that a maximum Pa_O2 and hence an optimal level of variability in VT exists at approximately 40 to 60%. These results are in agreement with the prediction of Suki and coworkers (12) based on the stochastic resonance hypothesis.

Nam and coworkers (14) failed to see any significant improvements in oxygenation or lung compliance when they ventilated dogs after oleic acid-induced lung injury with a ventilation mode similar to that of Lefevre and coworkers (9). This may be related to species variability because dogs contain collateral airways while pigs, used by Lefevre and coworkers (9), do not. Thus, dogs may have a more severe response to oleic acid injury than pigs. Variations in hemodynamics and ventilatory system, and inherent differences in the design of VV patterns, were also suggested as possible factors (14).

Our results also allow us to speculate that other factors may have also contributed to the lack of efficacy of VV reported by Nam and coworkers (14). These authors used a mode of VV such that VT varied between approximately 50 and 230% of the mean. In contrast, Lefevre and coworkers (9) and Mutch and coworkers (10) applied a distribution of VT between 75 and 135%. Our data suggest an optimal level of variability with a distribution of VT at 40 to 60% of the mean, with possible deterioration in oxygenation when this level of variability is exceeded. Thus, it is likely that the wide distribution of VT used in the study by Nam and coworkers may have contributed to the lack of improvement in blood oxygenation (14). In addition, Nam and coworkers (14) generated a more severe ALI model, using oleic acid injury, than either Lefevre and coworkers (9) or the present study. During our studies, extremely sick guinea pigs, displaying abnormally large tissue elastance, low Pa_O2 values (< 25 mm Hg on initial measurement), and P-V curves that were flat and considerably linear until extremely high airway pressures, failed to improve with any mode of ventilation (our unpublished observation). Indeed, on the basis of the stochastic resonance hypothesis of Suki and coworkers (12), improvement in Pa_O2 will occur only if the P-V curve is nonlinear. Thus, we conclude that if the P-V curve is essentially flat, linear, and steeply sloped as in extreme cases of acute lung injury, VV using any settings is unlikely to improve lung function.

Applications of intermittent deep inflations or sighs have been suggested as a possible method for lung volume recruitment. However, there has been little clinical evidence to indicate that sighs provide lasting beneficial effects during mechanical ventilation (20, 21). In addition, Mutch and coworkers (10) demonstrated that their biologic variable mode of ventilation proved superior to applications of regular sighs delivered at the same frequency and magnitude than their most extreme VT. Regardless, we were able to show consistent improvements in both lung function and gas exchange by using a ventilation mode that, at the greatest extreme, delivered a maximum VT of 7.1 ml/kg, significantly less than a vital capacity maneuver.

Further work is required to verify that VV may also reduce the risk of ventilator-induced lung injury during long-term mechanical ventilation, and experimentation with a larger animal model and at a variety of PEEPs is warranted. In conclusion, VV at moderate levels of variability may serve as a simple, inexpensive therapy for patients undergoing mechanical ventilation during acute lung injury.