INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

In the lungs of patients with acute respiratory distress syndrome (ARDS), a large portion of alveoli are atelectatic or fluid-filled (1), whereas other lung regions are aerated and relatively unaffected by the disease process (2). Traditional techniques of mechanical ventilatory assistance involve a generous tidal volume (VT) and positive end-expiratory pressure (PEEP). However, many animal studies suggest that this approach can perpetuate or exacerbate lung injury from high pressures and overdistention of the aerated lung regions. Consequently, there has been considerable interest in alternative ventilatory methods to improve oxygenation in these critically ill patients.

In a pig oleic acid (OA) model of ARDS, a novel mechanical ventilatory approach utilized a computer programmed to vary the breath-to-breath respiratory frequency (f) and VT with natural biologic variability (3). Compared with conventional mechanical ventilation (CV) at the same mean f and VT, such biologically variable ventilation (BV) was associated with a higher Pa_O2 and static compliance (Cst), and with reduced shunt and lung water content over a 4-h observation period. These salutary changes could be explained by recruitment of small airways and alveoli, which the investigators who conducted the study attributed to an intrinsic characteristic of the biologic variability in f and VT. For example, increased recruitment could have been induced with the larger BV breaths, but in some way that was different than that with regularly spaced large tidal volumes (sighs), which have had inconsistent effects on arterial oxygenation in previous studies (4). Decreased lung water in the BV group was interpreted to represent decreased ventilation-associated lung injury, which could also be explained by better airways recruitment.

The results of this initial experience with BV are potentially important because improvements in oxygenation appear to have been achieved without increased risk of volutrauma to the lungs. If validated, this novel strategy could represent a significant advance in patient management. However, there were significant differences in cardiac output (CO), Pa_CO2, pH, and actual delivered VT between the CV and BV groups, which could have influenced the results of the study. Because of the potential implications of these findings for patient care, we have attempted to confirm these results in a dog OA model of ARDS. In contrast to the results of Lefevre and colleagues, we found no improvement in Pa_O2, Cst or shunt with BV in our experiments.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Experimental Preparation

The study was approved by the Animal Care and Use Committee of The Johns Hopkins University. Eighteen mongrel dogs weighing 9 to 31 kg (mean: 15.9 kg) were anesthetized with pentobarbital sodium (30 mg/kg intravenously), orally intubated, and placed in the supine position. Pancuronium (3 mg bolus, followed by 1 to 1.5 mg/h intravenously) was administered for muscle relaxation. Isoflurane was administered at 1 to 1.5% end-tidal in 50% O₂ for maintenance of anesthesia. A catheter was placed in the femoral artery for blood gas sampling and monitoring of arterial pressure (Pa). A size 7.5-French Edwards Swan-Ganz catheter was inserted into the pulmonary artery via the external jugular vein. Mixed venous blood was sampled from the distal end of the pulmonary artery catheter. An esophageal balloon was inserted into the esophagus, and located at the point of maximum swings in tidal pressure, and its pressure (Peso) was continuously recorded. All pressure transducers were zero-referenced to the midchest level. Lactated Ringer's solution was given as a bolus until the animal's initial pulmonary artery occlusion pressure (Ppao) was 8 mm Hg, and was then infused continuously at 10 ml/kg/h for the duration of the experiment. At the end of the experiment the animal was killed by intravenous injection of saturated potassium chloride, after supplementation of anesthesia with additional barbiturate.

Mechanical ventilation was provided via a portable piston ventilator (PLV-102; Lifecare, Lafayette, CO), which was modified by the manufacturer to allow remote computer control. The ventilator was set at 15 breaths/min, with the VT adjusted to maintain the end-tidal PCO₂ (PET_CO2) at 30 to 35 mm Hg. Two sighs (each equal to three tidal breaths) were induced in each animal by occluding the expiratory limb of the ventilatory circuit before beginning control measurements. Compliance of the ventilator circuit was 0.57 ml/cm H₂O.

After administration of the initial fluid bolus was completed, baseline measurements were obtained of hemodynamic, respiratory, and lung mechanics parameters. These included heart rate, Pa, right atrial pressure (Pra), pulmonary artery pressure (Ppa), and Ppao. All were recorded on a Gould 2600S oscillograph (Gould, Inc., Cleveland, OH). CO was measured by thermodilution, using a 5 ml injection of saline at room temperature (mean of three measurements). Measurements of arterial and mixed venous blood gases (Model 248 pH/Blood Gas Analyzer; Ciba Corning Diagnostics, Medfield, MA) and hemoglobin concentrations (OSM3 Hemoximeter; Radiometer, Copenhagen, Denmark) were also made. The O₂ saturations of arterial and mixed venous blood gas samples were calculated with the equations for the pH-corrected O₂ dissociation curve (7). Pulmonary vascular resistance (Rpv) and shunt fraction (venous admixture, S/T) were calculated with standard formulas (8). Static compliance was obtained by clamping the expiratory limb of the ventilatory circuit at end-inspiration for 1 to 2 s to obtain a plateau pressure. The VT used for calculation of compliance was that set on the ventilator. FRC, measured in triplicate with the helium dilution method, and a quasistatic pressure- volume (PV) curve (three cycles at 2 L/min at a Paw between 2 and 30 cm H₂O) were obtained at baseline and after OA injury. For the last five animals in the CV group and last six animals in the BV group, inspiratory flow was measured with a pneumotachograph (Hans Rudolph Inc., Kansas City, MO), and instantaneous flow and airway pressures (Paw and Peso) were digitized at 10 Hz and stored in a computer (Macintosh PM7100/80; Apple Computer, Cupertino, CA) using a Superscope data-acquisition system (GW Instruments, Inc., Somerville, MA). These acquired data were used for the breath-by-breath analysis of mean Paw, peak inspiratory pressure (PIP), and delivered VT, as subsequently described.

Lung Injury

Lung injury was induced by infusion of 0.06 ml/kg pure OA (Sigma Chemical Co., St. Louis, MO), diluted 1:2 with absolute ethanol, over a period of 30 min, into the right atrial port of the Swan-Ganz catheter with the animal in the supine position. This dose was chosen on the basis of preliminary experiments designed to identify the OA dose necessary to achieve a level of lung injury similar to that in the recently reported study of BV in pigs (3). At the end of OA infusion, animals were randomly assigned to one of two ventilatory modes: CV, fixed at the baseline VT and f of 15 breaths/min, or BV, with the same mean f and VT as subsequently described. As in the previous study of BV (3), no PEEP was used in either group. Ventilation was continued with either CV or BV for the duration of the experiment, and hemodynamic, respiratory, and lung mechanics data were obtained every 30 min for 4 h. Four animals in the CV group and three animals in the BV group did not survive to 4 h, but all animals lasted at least through the 180-min measurement period.

BV Ventilation System

The Superscope data-acquisition system used to digitize the Paw, Peso, and flow signals also directed the PLV-102 ventilator to deliver biologic variations in f and VT according to a previously generated file of interbreath periods from an awake, spontaneously breathing, quiescent dog (provided by Dr. Paul Murray and Jim Palazzo of the Department of Anesthesia, Cleveland Clinic). The original breathing file contained 784 breaths with an f value of 37 ± 9.7 breaths/min (mean ± SD) (coefficient of variation [CV]: 26.2%). This data set was scaled to a mean f of 15 breaths/min for use in this protocol. It was necessary to edit the file such that the f and VT differences between any two adjacent breaths were within limits imposed by the mechanical capability of the ventilator. This resulted in minor changes in the frequency spectrum (Figures 1 and 2). The edited and scaled file contained 784 breaths with an f of 15 ± 3.6 breaths/min (CV: 24%). The minimum and maximum f of the file were 6.5 and 30 breaths/min, respectively (Figure 1A). Histograms of the f distribution before and after editing are shown in Figure 2. A corresponding VT file was generated from this f file on the basis of the control VT for a given animal: the VT of each breath was calculated such that minute ventilation (E = f × VT) remained constant (Figure 1B). The mean f and VT of these files were thus equal to the f and VT of the baseline ventilation settings. Because inspiratory flow and the I:E ratio were kept constant, larger breaths had longer inspiratory times followed by longer expiratory times. Table 1 presents the characteristics of scaled f and normalized VT files after editing. For example, 50% of breaths (25th to 75th percentile) were within the f range of 12.2 to 17.1 breaths/min with VT between 88% and 123% of the control value. At a mean f of 15 breaths/min, the BV file lasted about 52 min, and thus was repeated several times during the 4-h experiments. Ventilator settings were transiently returned to the baseline settings for 3 to 5 min while hemodynamic and compliance measurements were made, after which BV was resumed.

View larger version (29K): [in this window] [in a new window]   Figure 1.   (A) BV f file and (B) corresponding VT file before (open symbols) and after (closed symbols) editing, plotted against breath number. Breaths with too high or too low frequencies were adjusted to meet mechanical constraints of the BV ventilator system.

View larger version (13K): [in this window] [in a new window]   Figure 2.   Histograms of BV f file before (open symbols) and after (closed symbols) editing. Editing maintained the overall shape of the distribution. Both distributions have a mean f of 15 breaths/min.

PV curves were obtained before and after OA-induced lung injury, with examples shown in Figure 3. A computer-controlled system measured quasistatic PV curves during inflation and deflation of the lungs at a constant flow of 2 L/min from values of Paw of 2 to 30 cm H₂O for three cycles. Paw, Peso, and airway opening flow signals were digitized and stored on the computer for subsequent analysis. The PV curve of the OA-injured lung often exhibits a change in slope (Pflex) on the early-portion of its inspiratory limb, as has been described in many patients with ARDS (9). Pflex was determined from the PV curves as the intersection of the linear projections of the early and midportions of the first cycle inspiratory curve.

View larger version (30K): [in this window] [in a new window]   Figure 3.   PV curves before (darker line) and 240 min after (lighter line) OA-induced lung injury. PV curves were acquired from Paw values of 2 to 30 cm H2O for three cycles each, at a constant air flow of 2 L/min. Upward shift with subsequent cycles reflects lung volume recruitment. The noisy appearance of the curves is caused by cardiac oscillations in Peso. Note Pflex on the inspiratory limb of the PV curve of the OA-injured lung.

Statistical Analysis

Data were subjected to longitudinal analysis, using the STATA statistical software package (Stata Co., College Station, TX). This analysis takes into account the missing data for animals that did not survive for the full 4 h of the experimental period, and fits linear models to compare the time courses of different parameters, with the assumption that data values depend upon each other over time. The data values at the time of development of initial injury were used as the "baseline" values for the longitudinal analysis. Student's t test was used for single-time-point comparisons between groups. Survival differences between groups were tested with a log-rank test (Stata). A value of p < 0.05 was considered significant in both longitudinal analysis and with Student's t test. Data are presented as mean ± SEM unless otherwise noted.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

OA infusion caused significant lung injury, with decreased Pa_O2 and Cst and increased airway pressures observed at 30 min after OA infusion in both the CV and BV groups. Comparable levels of injury were achieved in the two groups, with Pa_O2 values of 89 ± 9.7 mm Hg versus 77 ± 6.4 mm Hg and shunt values of 0.36 ± 0.09 versus 0.42 ± 0.05 for the CV and BV groups, respectively, at 30 min after lung injury. In the CV group, nine, six, and five animals survived to 180, 210, and 240 min, respectively, compared with nine, seven, and seven animals in the BV group (p = 0.90).

Hemodynamic data are shown in Table 2. Some animals died before 4 h, and the means at 210 and 240 min therefore include fewer subjects, as indicated. Mean Pa, Ppao, and hemoglobin (h) were constant over the 4-h experimental period and showed no significant differences in the CV and BV groups. Ppa and Rpv increased with time in both groups. However, the two groups were not statistically different except at baseline and at 240 min. CO decreased significantly after OA injury in both groups, and remained relatively unchanged thereafter; the two groups were not statistically different in CO at any time point. Longitudinal analysis showed no differences in the time courses of hemodynamics in the BV and CV groups (Ppa: p = 0.62; Ppao: p = 0.98; Rpv: p = 0.18; CO: p = 0.99; Pa: p = 0.61).

Respiratory gas data are summarized in Table 3 and Figure 4. In both groups, Pa_O2 and Pv_O2 decreased significantly after OA injury, and were not statistically different in the two groups at any time points or over time. Pa_CO2 increased and pH decreased with time in both groups, but were not statistically different in the two groups. S/T increased significantly after OA injury, but was again not statistically different in the two groups. By longitudinal analysis, significance values were Pa_O2: p = 0.77; Pv_O2: p = 0.29; Pa_CO2: p = 0.93; pH: p = 0.92; and shunt: p = 0.86. Figure 5 shows the Pa_O2 data versus time for the nine individual animals in each group.

View larger version (17K): [in this window] [in a new window]   Figure 4.   Time course of mean Paw, PaO2, and S/ T for BV and CV groups. OA was infused from time = 0 to time = 30 min. Values are means ± SEM.

View larger version (30K): [in this window] [in a new window]   Figure 5.   Time course of PaO2 for the nine individual animals in each of the CV and BV groups.

Airway pressures, VT, and Cst are shown in Table 4 and Figure 4. PIP and mean Paw increased significantly after OA injury and continued to increase with time, but the CV and BV groups did not differ significantly (PIP: p = 0.86; mean Paw: p = 0.68). PIP and mean Paw of individual breaths were calculated from the digitized airway pressure data and averaged over 30-min epochs after lung injury (CV group, n = 5; BV group, n = 6). Cst decreased significantly after OA injury and continued to decrease with time in both the CV and BV groups. The two groups did not differ significantly in Cst (p = 0.82). Delivered VT was well matched between groups, and the baseline mean VT values were maintained throughout the experiment (p = 0.18 for VT versus time by longitudinal analysis).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The management of mechanical ventilation in patients with ARDS and ALI remains a difficult challenge because of the tradeoff between maintaining adequate gas exchange and furthering lung injury by overdistention. Some innovative approaches and adjuncts to mechanical ventilation, such as extracorporeal gas exchange (12, 13), inverse-ratio ventilation (14, 15), high-frequency ventilation (16, 17), and liquid ventilation (18) have been suggested to support gas exchange while reducing potentially injurious overdistention. The recent study of BV (3) described earlier suggested that mechanical ventilation with natural biologic variability in f and VT could reduce lung injury and improve gas exchange without increases in mean Paw or PIP. In that study, VT gradually decreased in both CV and BV treatment groups. It decreased more in the CV group, and this was associated with a higher Pa_CO2, lower pH, and higher CO. The higher CO could have affected the key outcome variable, oxygenation. Our experiments were designed to assess effects of BV with constant tidal volumes, Pa_CO2, and CO. With this approach, we found no differences between our BV and CV groups in hemodynamics, respiratory gas tensions, airway pressures, or lung mechanics. Other possible explanations for the differences in our findings and those of the study described earlier (3) include the animal models used and the BV pattern utilized.

Hemodynamic and Ventilatory Considerations

In the previous study (3), there were significantly higher mean Pa_O2, lower shunt, and higher Cst values in the BV group than in the CV group. However, the CV group had higher mean Pa_CO2 and lower pH values, and CO was significantly higher starting at 30 min after OA injury. Although the baseline values of VT were comparable in the CV and BV groups, mean VT decreased with time in both groups and, more importantly, the values of VT in the CV group became significantly smaller than those in the BV group.

Some of these results may be due in part to important differences between the ventilators used in this previous study and our study. The ventilator used in the previous study appears to have been load sensitive, so that the delivered VT progressively decreased as compliance decreased, whereas our ventilator delivered a VT independent of compliance. As a result, airway pressures in our animals progressively increased. Although our baseline oxygenation was similar to that in the previous study, gas exchange deteriorated during our study, whereas it stabilized or improved in the previous study. Perhaps this was caused by the higher airway pressures in the face of falling compliance in our study, leading to further ventilator induced lung injury. However, despite similar mean Paw and PIP values in our two study groups, there was no apparent salutary effect of BV. Moreover, the peak and mean pressures were lower than those known to cause ventilator-associated lung injury, and were lower than those obtained in the prior study. Nonetheless, the possibility of ventilator-induced lung injury cannot be excluded.

Differences in CO between treatment groups may also have had an effect on the shunts in the groups, a possibility recognized in the previous study. In an isolated blood-perfused dog lung lobe preparation, Sandoval and coworkers (19) showed that with a fixed Pv_O2, a changing CO had no immediate effect on shunt, and that with a fixed CO, shunt was linearly related to Pv_O2. They postulated that decreases in pulmonary arterial PO₂ could contribute to hypoxic pulmonary vasoconstriction, altering the distribution of pulmonary blood flow at constant CO and changing the extent of shunt. In light of this, the investigators in the previous BV study reasoned that the differences in CO observed in their experiments did not cause the differences that occurred in shunt, since the Pv_O2 values were similar in the CV and BV groups. Thus, they attributed the improved gas exchange with BV to enhanced airway recruitment.

However, aside from the immediate influence of CO and PvO₂ on shunt, the development of alveolar edema after OA-induced changes in lung vascular permeability depends strongly on the level of perfusion of the injured lung regions (20, 21); a higher CO can increase shunt through further worsening of lung edema (22, 23). Moreover, decreased ventilation in the CV group in the previous study of CV and BV could have contributed to the increased shunt in that group owing to the smaller delivered VT and decreased end-inspiratory recruitment (24). Hypercarbia and acidosis may have further contributed to the higher CO found in the CV group, through reflex sympathetic stimulation. The differences in Cst in the previous study are also difficult to interpret. They may represent a beneficial effect of BV on lung recruitment, and the hypoventilation in the CV animals may have resulted from their falling compliance. Alternatively, the lower Cst may merely reflect the greater edema resulting from the higher CO, and/or decreased end-inspiratory recruitment from the decreased VT in the CV group. Thus, smaller values of VT, decreased ventilation, and higher CO are confounding factors that may have influenced the results of the previous study. In our experiments, VT, Pa_CO2, Pv_O2, and CO were well matched in the CV and BV groups, and no differences were found in the resultant Pa_O2, shunt, or Cst.

BV Data Sources

A potentially important difference between the previous study of BV and our study was the source of the data used to generate the f variability files. The previous study used the peak-to-peak variability in systolic blood pressure of an anesthetized pig as the basis for the breathing f file, based on the previous demonstration that variability in heart rate, peak-to-peak changes in systolic blood pressure, and respiratory rate were similar and shared a common centering frequency equal to the respiratory f (25). In the present study, a file of actual interbreath periods, obtained from an awake, spontaneously breathing, quiescent dog was used to generate the BV breathing pattern. The frequency distributions of the f files for the two studies were similar in shape (Figure 2; [3]: Figure 2), but our file was longer (784 versus 368 breaths), and its coefficient of variation was considerably higher (24% versus 11.5%). It is possible that BV showed no advantage in our study because of differences in these f distributions. For example, improvement in alveolar recruitment implies that the net effect of volume gained during larger breaths must exceed the effect of any volume lost during smaller breaths. The larger coefficient of variation indicates that our file had more breaths below as well as above the mean, and the more numerous smaller breaths may have counteracted the benefits of the larger breaths.

Species Differences

Another potentially significant difference between the previous study and our study of BV is the animal model of pig versus that of dog. Although we adjusted our OA dose to provide a similar initial Pa_O2 as Lefevre and colleagues, the higher mortality (six of 18 animals) and progressive deterioration in oxygenation in both of our study groups suggests a species difference that resulted in a more severe injury that was not responsive to BV.

Perhaps the most noteworthy anatomic difference between the pig and the dog is that the dog has collateral ventilation channels in the distal lung, which in the pig are virtually absent (26). These channels, which primarily connect at the level of the alveolar duct (26), provide an alternate pathway for inflation of distal lung units. Human lungs also contain collateral channels. In pathologic conditions in which airway obstruction may occur at any level of branching, collateral channels may provide a partial bypass to the obstruction and increase the number of pathways available for opening distal airways. In addition, collateral channels improve mechanical interdependence, which further supports uniform expansion of air spaces (27). When surrounding airways are partly inflated, obstructed alveolar ducts may open because tethering forces act downstream of the obstructed region. Thus, one would expect that collateral channels would facilitate the recruitment of airways and alveoli. However, since BV appeared to be more effective in the pig than the dog, the presence of collateral channels does not explain the difference in findings in the studies of BV with these two species.

However, Kuriyama and associates (28) have hypothesized that species with well-developed collateral channels (e.g., dogs) have less need for a vasoconstrictor response to hypoxia, whereas species with poorly developed collateral channels (e.g., pigs) have a greater dependence on vascular responses to maintain Pa_O2 when airways are obstructed. As they point out, this may explain why dogs exhibit feeble pulmonary vascular responses to hypoxia as compared with pigs, and why shunts are greater in dogs than in pigs, as seen in our study and that of Lefevre and colleagues (3). Although multiple airway pathways are present and mechanical interdependence would be expected to be improved with collateral channels, a poor pulmonary vascular response to regional hypoxia in dogs may help explain the difference in findings in the two studies.

Stochastic Resonance Hypothesis

Suki and colleagues proposed a mechanism invoking the statistical phenomenon of stochastic resonance to explain the effects of BV seen in Lefevre and colleagues' study of BV (29). Because of the nonlinear shape of the PV curve of the injured lung, which typically exhibits a change in slope about the Pflex point on its inspiratory limb (Figure 3), an increment in VT above a mean value that extends above Pflex will cause a smaller increase in Paw than the decrease in Paw caused by a corresponding decrement in VT. As a result, when values VT are distributed above and below Pflex, the resulting mean Paw is lower than it would be for the same average VT delivered constantly. Through this mechanism, therefore, BV could produce an increased average end-inspiratory volume and consequently increased recruitment without increases in mean Paw. It's essential, however, that the values of PIP be distributed about Pflex in order to take advantage of the nonlinearity of the PV curve.

Although our study was not specifically designed to investigate this hypothesis, we were able to examine the relationship between PIP and Pflex for our last 10 animals, using an online system in which Paw and flow of every breath were digitized for subsequent analysis. Table 5 presents the relationship between the distribution of breath PIP values relative to Pflex and the final Pa_O2. As required by the stochastic resonance hypothesis, the majority of these animals did have their PIP values distributed about Pflex. Consequently, our data do not support the stochastic resonance explanation for how BV could improve recruitment and oxygenation. As intriguing as the hypothesis of Suki and colleagues appears, a ventilation scheme with a variable VT designed to distribute PIP about Pflex may be of limited clinical relevance, since some clinicians and investigators recommend raising PEEP above Pflex to prevent ventilator-associated lung injury from repeated opening and closing of small bronchioles and alveoli (27, 30).

Variability or BV?

Although we used a BV file of dog respiratory breathing patterns, we know of no advantage of "biologic" variability over other forms of variability with respect to the explanations offered for how BV ventilation could improve recruitment and oxygenation. Both the "avalanche" model of recruitment (31) and the stochastic resonance mechanism (29) would work equally well with an arbitrarily imposed variation of appropriate size, amplitude, and distribution. Alternatively, one could speculate that, in addition to these mechanical explanations, there is a requisite reflex component to the phenomenon of recruitment that is activated only with "biologic" breathing patterns. Thus, although we did not find a benefit to BV, it could be that our pattern lacked some essential component of variability.

Conclusion

In summary, in a canine model of lung injury, we found no advantage to mechanical ventilation with f and VT varied in a pattern derived from normal canine breathing as compared with ventilation with a constant f and VT at equal minute volumes. The previously reported benefit of BV in pigs may reflect technical or species differences, or may be due to the lower CO in animals allocated to BV than in those given CV in the study in which this was reported. Frequent large sighs have shown transient improvement of oxygenation in ARDS (4). However, we found no empiric advantage to either a smaller variation in breath size and frequency or to a "natural" breathing pattern. These data suggest that BV neither aids lung recruitment nor ameliorates lung injury.