INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS AND DISCUSSION REFERENCES

Assisted mechanical ventilation is currently the most common form of ventilatory support in critically ill, ventilator-dependent patients. Unlike the case of controlled mechanical (passive) ventilation (CMV), where airway pressure (Paw) is the only distending force, with assisted ventilation the total applied pressure at any instant is made up of Paw and a highly variable and unpredictable contribution from the patient's own muscles (Pmus). Consequently, the various methods that employ the relation between Paw and flow and volume to noninvasively estimate passive mechanics during CMV [interrupter technique (1), linear regression analysis (2, 3)] cannot be used with any confidence during assisted ventilation. For reliable estimates of passive mechanics, it is currently necessary to temporarily place the patient on CMV, or to use invasive methods (4). Both of these approaches have practical limitations. As a result, potentially useful information regarding changes in respiratory mechanics in the course of mechanical ventilation is not obtained, or at least not reliably obtained, in most patients.

Proportional assist ventilation (PAV) (5, 6) has certain features that may circumvent the problems associated with other methods of assisted ventilation and may, theoretically, make it possible to estimate, continuously and noninvasively, passive elastic properties of the respiratory system without the need to silence the respiratory muscles. This offers two potential advantages:

First, there is every reason to expect that elastance (E) may change spontaneously in the course of mechanical ventilation. Changes may occur as a result of alteration in lung water and blood content, abdominal volume, body position or when the tip of the endotracheal tube slips into the right mainstem bronchus. Patients on assisted ventilation often develop episodes of respiratory distress at a level of support that was adequate previously. An increase in E is an obvious potential mechanism for such episodes. The availability of continuous estimates of E would make it possible to identify such a mechanism and to take appropriate action.

Second, PAV (5) has several physiologic advantages (5, 6), including automatic changes in level of support (i.e., Paw) as ventilatory demands change, lower distending pressures, and better synchrony between patient and ventilator, which should enhance comfort and provide more accurate estimates of changes in patient's respiratory rate. These physiologic advantages may translate into clinical benefits with implications to mortality, morbidity, and length of ventilatory support. To date, widespread use of PAV, and hence testing for these potential clinical benefits, has been seriously hampered by lack of simple ways to estimate E and resistance (R) on an ongoing basis; knowledge of E and R is required for proper, trouble-free implementation of PAV. The availability of a simple, noninvasive way to continuously monitor E would facilitate the clinical use of PAV and would also make it possible to apply PAV for extended periods in order to assess its potential clinical benefits, if any.

The purpose of this communication is to describe the theoretical basis for a modified inspiratory hold technique to assess passive elastic properties during PAV and to report the results obtained with this approach in a large number of ventilator-dependent patients.

Theory

Figure 1 is a schematic illustration of the main features of respiratory muscle output during spontaneous efforts. This general pattern is based on decades of observations made by numerous investigators using electrical activity or pressure output of respiratory muscles under a variety of experimental circumstances (7, 8). The onset of a breathing cycle is defined by the point at which inspiratory muscle activity or pressure output begins increasing from baseline. Inspiratory muscle output typically rises in a ramplike fashion, reaching a peak after a variable duration. There is often a period of fairly stable Pmus near the peak (Figure 2). Pmus then begins declining toward baseline. Under most experimental circumstances, both inspiratory and expiratory muscles are silent, and generate no pressure, until the onset of the next inspiration (horizontal dashed line, Figure 1). Expiratory muscles may, however, show spontaneous phasic activity particularly when respiratory drive is high or when lung emptying is delayed (7). Under these conditions, Pmus changes in an opposite direction (from inspiratory pressure) during the course of neural expiration (Figure 1). When present, phasic expiratory activity rises gradually, beginning at some point during neural expiration, and reaches a plateau which is sustained until the next inspiration. When phasic expiratory activity is present, it usually begins when the declining phase of inspiratory activity ends (7). We have shown in animals, however (9), that if the airway is occluded at end-inspiration, the onset of phasic expiratory activity is delayed so that a period during which there is neither inspiratory nor expiratory activity is produced (gap, Figure 1). The duration of this "quiet" period is a function of lung volume (during the occlusion) and of chemical drive [9; see also Figure 41 in (7)]. If Paw, flow (), and volume (V) are measured during this interval, the results should reflect passive properties of the respiratory system.

View larger version (13K): [in this window] [in a new window]   Figure 1.   Schematic representation of the general pattern of respiratory motor output. Positive Pmus denotes inspiratory pressure and vice versa. Inspiratory pressure rises in ramplike fashion, usually leading to a brief plateau, followed by a decrease in inspiratory pressure. In most cases there is no expiratory pressure development and Pmus remains zero until the next inspiration (dashed line). When there is active expiratory effort, it usually also increases as a ramp leading to a plateau toward the end of neural expiration. It is postulated that with end- inspiratory occlusions, there will be a time period between the end of inspiratory pressure and the beginning of expiratory pressure during which pressure output is zero. Time and Pmus are in relative units.

View larger version (60K): [in this window] [in a new window]   Figure 2.   Tracings of , V, Paw, and transdiaphragmatic pressure (Pdi) during a PAV breath subjected to a brief end-inspiratory occlusion. Note that flow crosses zero (at cursor) during the declining phase of Pdi and that Paw during the occlusion rises in a manner that is the mirror image of the decline in Pdi (the scales of Paw and Pdi tracings are equivalent). Note also the slight expiratory flow at the beginning of occlusion, which represents transfer of gas from patient to the tubing as tubing pressure increases during the occlusion.

In the proposed approach, we take advantage of the fact that with PAV the end of the inflation phase is constrained to occur during the declining of phase inspiratory activity (5; see also section A of online data supplement). Under these conditions, the onset of an end-inspiratory occlusion should coincide with the latter part of the declining phase of inspiratory Pmus (Figure 2). Because lung volume during the occlusion is nearly constant (section B of online data supplement), and there is virtually no flow (section B, online data supplement), Paw should essentially be the mirror image of Pmus (Figure 2). The continued decrease in Pmus should result in a corresponding increase in Paw, and, when inspiratory Pmus becomes zero, a plateau should appear in Paw. During PAV, this plateau should coincide with the period of no inspiratory or expiratory Pmus (gap, Figure 1), and Paw during this plateau (Pplat) should reflect passive elastic recoil (Pel). With other modes of mechanical ventilation, where end-inspiratory occlusion may coincide with any part of the patient's respiratory cycle, a plateau in Paw is less likely to occur, and, if it does, it need not reflect Pel. A plateau in occlusion pressure simply means that Pmus is constant (see section B of online data supplement). Pmus may be temporarily stable at the height of inspiratory or expiratory Pmus, resulting in a stable Paw. Pplat under these conditions would underestimate or overestimate Pel, respectively, by amounts corresponding to peak inspiratory or expiratory Pmus.

There are two major uncertainties with the proposed approach. These relate to the potential impact of phasic expiratory activity and the possible occurrence of behavioral responses to the occlusion:

Phasic expiratory activity. Success of the proposed approach is predicated on the presence of a period, immediately after the end of the declining phase of inspiratory Pmus, during which there is no expiratory activity (gap, Figure 1). In the absence of phasic expiratory activity, this condition will clearly be met. If phasic expiratory activity is present, a plateau will appear only if there is a reasonably long gap between the end of inspiratory activity and onset of expiratory activity. In the absence of such a gap, the point at which net Pmus (i.e., Pinsp  Pexp) crosses zero would be difficult to identify because the increase in Paw due to the declining inspiratory Pmus would merge directly with the increase in Paw due to increasing Pexp.

The frequency of occurrence of phasic expiratory activity in the general intensive care unit (ICU) patient population has not, to our knowledge, been reported. Phasic expiratory activity is very common in patients with severe chronic obstructive pulmonary disease (COPD), even at rest (10, 11), and has been observed in several patients with COPD during mechanical ventilation (12, 13). The frequency of occurrence of this activity in ventilated non-COPD patients is not known. However, it may be expected that patients will, as do normals, display phasic expiratory activity when respiratory drive is high or lung emptying is slowed. It may, therefore, be expected that at least a fraction of patients may be affected by this uncertainty. The duration of the "gap" between end of inspiratory activity and onset of phasic expiratory activity, when the latter is present, has also not been systematically examined. Published recordings in unintubated patients with COPD (10) and in normal subjects with experimental flow limitation (14) suggest that the gap, if any, is very brief during unoccluded expiration. However, as indicated earlier, in animals, end-inspiratory occlusions tend to delay the onset of expiratory activity relative to end of inspiratory activity, in effect creating a gap (9). If this reflex is operative in mechanically ventilated patients, it would provide an opportunity for obtaining passive data even in patients with strong phasic expiratory activity.

Behavioral responses. A major potentially confounding factor is the occurrence of behavioral responses to the occlusion before the appearance of a plateau in Paw. The potential for this to occur depends on the time required to reach a plateau and the latency for behavioral responses. In highly alert normal subjects who were anticipating a respiratory intervention, the minimum latency for a response is approximately 200 ms (15, 16). Lack of anticipation, as may be produced by randomly applying the occlusions, and a somewhat attenuated level of vigilance, as in most mechanically ventilated patients, may be expected to increase the latency. Nonetheless, it would seem imprudent to discount behavioral responses contributing to Paw beyond 250 to 300 ms of occlusion time except in totally unresponsive patients. Determination of the frequency of occurrence and latency of behavioral responses in mechanically ventilated patients is a primary objective of the present study.

The magnitude of Pmus remaining at zero flow crossing (and hence at onset of occlusion) and the time required for pressure to reach a plateau beyond this point may be expected to vary considerably (see section A of online data supplement). Where respiratory motor output is low or flow reversal occurs near the end of the declining phase, Paw should become fairly stable well within the latency for behavioral responses. Problems may arise, however, if the patient generates large inspiratory Pmus and flow reversal occurs relatively early in the declining phase (section A, online data supplement). Here, the time required to reach a plateau may, theoretically, be the total duration of the declining phase. Little is known about the duration of the declining phase in mechanically ventilated patients. In alert normal subjects, the declining phase is generally fairly long and may occupy most of expiratory time (17). The proposed approach would clearly not be suitable under these conditions because a plateau will not be reached, or even approached, within the latency of behavioral responses. Measurement of Paw at a time within this latency (e.g., at 200 ms from occlusion) is likely to include an important inspiratory Pmus component leading to underestimation of Pel. In unconscious (20) and sleeping (personal observations) humans, the duration of the declining phase appears shorter, with Pmus reaching a small fraction of peak value within 0.3 to 0.4 s from the onset of decline in most subjects. If this applies to mechanically ventilated patients on PAV, then it is likely that very little Pmus will be left by 0.2 to 0.3 s from the onset of occlusion, particularly because a finite, and in many cases a large, fraction of the declining phase will have already elapsed before the onset of occlusion. Another important objective of the present study is to determine the pattern of change in Paw, and hence the pattern of decline in Pmus, over the initial part of occlusion, that corresponds to the latency for behavioral responses.

In summary, we propose that in the PAV mode a Paw plateau will appear soon after the onset of the inspiratory hold maneuver and that Paw at this point (Pplat) will accurately reflect passive elastic recoil pressure. In the current study, we will define the time course of Paw during the first 300 ms after the onset of inspiratory hold and will compare the E calculated from this Pplat with that obtained during a period of CMV.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS AND DISCUSSION REFERENCES

Patients

The study was done over a 2-yr period (7/96 to 9/98) in the medical ICU (MICU) of the Health Sciences Center. The unit is the only general MICU in this large tertiary care facility and attends to all medical patients requiring mechanical ventilation. Patients were not preselected on the basis of age, sex, length of intubation, level of vigilance, or severity or type of illness. Any patient who was already on an assisted mode of ventilation and displayed triggering efforts and who was hemodynamically stable, was a candidate. Candidates were studied whenever the respiratory therapist trained in this procedure (K.W.) was available and an informed consent could be obtained from the patient or next of kin. The study was approved by the institutional committee for research on human subjects.

Seventy-four patients were studied in total (37 female, 37 male). Their age ranged from 19 to 83 yr (63.4 ± 15.5). Level of vigilance spanned the usual spectrum observed in a general MICU, from highly alert to comatose. The duration of intubation before the study ranged from 2 to 65 d. Ten patients had undergone tracheostomy because of intubation > 2 wk. The mechanical abnormalities also covered a wide range. Mechanics measured using the interrupter technique (1) during a period of CMV before the study showed E values ranging from 10.6 to 96.0 (27.1 ± 12.0) cm H₂O · L¹ and R values from 5.6 to 21.4 (13.2 ± 4.0) cm H₂O · L¹ · s. Positive end-expiratory pressure (PEEP) before the study was 5.8 ± 2.1 cm H₂O.

Apparatus

All studies were done using the Winnipeg Ventilator, a locally constructed, piston-based ventilator capable of delivering all common modes, in addition to PAV (21). Tubing was standard. A standard exhalation valve was used (Instrumentation Ind. Inc., Bethal Park, PA) and PEEP was applied by attaching PEEP valves (Life Design Systems, Carollton, TX) to the exhalation valve. An end-inspiratory occlusion was done by turning on a solenoid valve connected to the tubing that pressurizes the exhalation valve. When activated, this solenoid valve prevented the opening of the exhalation valve that normally occurred at the end of inspiration.

A heated pneumotachograph (linear to 160 L/min; Hans Rudolph, Inc., Kansas City, MO) and a pressure port were attached between the Y connector and the endotracheal (ET) tube to measure flow and Paw. , V (integrated ), and Paw signals were digitized at 125 Hz and stored on computer using a Windaq data acquisition system (DATAQ Instruments, Akron, OH).

Protocol

The patient was switched to the Winnipeg Ventilator. Initial settings (mode, level, PEEP, fraction of inspired oxygen [FI_O2]) were identical to those in use before the study. After a few minutes to allow the patient to adjust to the new ventilator, the ventilator was placed in the volume-cycled mode with a tidal volume (VT) setting slightly higher than that received by the patient earlier, and a square flow pattern. Back-up rate was gradually increased until inspiratory efforts ceased (CMV). This was evident from lack of triggering efforts, reproducible Paw waveforms, and a reproducible, smooth decline of expiratory flow during the expiratory phase (22). At these CMV settings, expiratory flow often did not reach zero before the onset of the next breath, indicating dynamic hyperinflation (DH). At intervals, the back-up rate was decreased for one cycle to permit expiratory flow to return to near zero and the next inflation was subjected to an inspiratory hold. In a few patients, notably those with severe flow limitation, it was not possible to wait until expiratory flow reached zero. This required an excessive delay during which the patient began inspiratory efforts. In such cases, the longest expiratory time (TE) consistent with continued passivity was used. The VT and flow rate of the breath to be occluded after the long expiration were varied, generally over a range of 0.5 to 1.5 of the CMV settings, unless the higher settings resulted in Pplat > 35 cm H₂O. Approximate calculations of E and R were made by the bedside by noting peak Paw (Ppeak), plateau pressure (Pplat), PEEP, flow, and VT of some occluded breaths. Standard formulas were used (i.e., E = [Pplat  PEEP]/VT and R = [Ppeak  Pplat]/).

The ventilator was then switched to the PAV mode with volume assist (VA) and flow assist (FA) initially set at 80% of E and R determined during CMV. End-inspiratory occlusions were done at random intervals with between 3 and 30 breaths separating successive occlusions. In the first 46 patients, the respiratory therapist decided when to perform an occlusion and the occlusions were initiated and terminated by manually controlling the exhalation valve control switch. In these experiments, occlusions were generally done only during periods of relative stability (i.e., no nursing interventions, agitation, cough, etc.). In the last 28 patients, in order to assess results when occlusions were done in a truly random fashion under field conditions, the implementation of occlusions was relegated to a microcontroller. A random number generator determined when an occlusion was to be performed, and this took place regardless of nursing interventions or patient status. When occlusions were done manually (first 46 patients), the duration of occlusion was not controlled and varied between 0.5 and 1.5 s (0.76 ± 0.31 s). With the automated procedure, the duration of occlusion was fixed at 0.30 s.

In the first several experiments, monitoring on PAV was limited to the time required to obtain 10 to 20 observations, usually 10 to 30 min. Later on, monitoring time was increased to permit measurements at different levels of assist and to assess time-dependent changes in the measurements.

In 24 patients, a second period of CMV was obtained, after the PAV study, and passive mechanics were measured again. During this second study, end-inspiratory occlusions were done without prior slowing of the ventilator. Pplat, therefore, included a component related to any DH that may have been present during baseline ventilation.

Analysis

CMV data were analyzed manually from the Windaq files. All occlusions were inspected for evidence of respiratory efforts and those showing efforts were deleted. For all others, Paw was measured 0.25 s from the onset of occlusion (P_0.25). This was to match measurement time during PAV. The onset of occlusion (T₀) was the point at which flow reached zero. In cases where there was slow equilibration between patient and tubing during the occlusion, with a long tail of low flow, T₀ was defined as the point at which flow reached 0.05 L · s¹. For each occluded breath we measured Ppeak, flow at Ppeak (peak), PEEP, and VT, in addition to P_0.25.

The relation between VT and (P_0.25  PEEP) was plotted for each patient. Inspiratory R at peak (Rpeak) was calculated from [(Ppeak  P@ T₀) /peak]. R was corrected to 1 L · s¹ using known values of K₂ (of Rohrer's equation) of ET tubes (23). Thus R @ 1 L · s¹ = K₁ + K₂ET, where K₂ET is K₂ of the ET tube and K₁ was computed from: K₁ = Rpeak  (peak · K₂ET).

The PAV data were analyzed using a macro program specially written for this purpose. Data were obtained from each occluded breath. T₀ was defined as the point of transition from decreasing flow (near the end of inflation phase) to the nearly flat flow (at near zero) during the occlusion. Paw was sampled (and stored) at 0.05-s intervals between T₀ and T₀ + 0.3 s. VT and duration of the inflation phase (TI) corresponding to each occlusion were also stored. Occlusions were rejected if any of the following conditions occurred: (1) The inflation was terminated by a pressure or volume limit set on the ventilator. In the first 46 studies, where the occlusions were done manually, this situation was simply avoided; occlusions were not carried out until ventilator limits were set such that they exceeded the range of Paw and VT observed on PAV. In the last 28 patients, a signal was gated to the microprocessor, from the ventilator, signifying a ventilator-terminated cycle. Occluded breaths during which such signals were received by the microcontroller were rejected. (2) The flow during the inflation phase preceding the occlusion did not exceed 0.1 L · s¹ continuously for > 0.3 s. (3) VT or TI of the occluded breath fell outside ± 3 SD from the mean values of these two variables. (4) The relation between P_0.25 and VT of a given occlusion fell outside ± 3 SD of the relation established from all occlusions in the same patient. (5) They occurred during periods in which unoccluded breaths showed a bipeaked early expiratory flow pattern. The rationale for these rejection criteria is discussed in section C of the online data supplement.

Comparison between CMV and PAV E. E is often VT-dependent in mechanically ventilated patients (1). In the current study, the VT range available during CMV occlusions was wide enough in 69 of the 74 patients to establish if E was VT-dependent. In 13 patients, the relation between VT and Pel [i.e., (P_0.25  PEEP)] was convex to the VT axis (i.e., E increasing with VT). In 17 patients it was concave to the VT axis, whereas in the remaining 39 patients it was independent of VT over the range studied (i.e., the intercept of the relation was not significantly different from zero). To determine the CMV elastance value (ECMV) to be used in comparisons with EPAV, we averaged the VT of all occlusions used during PAV in each patient (VTPAV) and determined Pel during CMV, at this volume, from the VT  Pel relation of CMV. The relevant ECMV was then estimated from this Pel value and VTPAV. In the five patients in whom a reliable VT  Pel relation was not available in CMV (VT range was too small), this correction was not possible and average ECMV and EPAV were directly compared.

The other confounding factor in comparisons between ECMV and EPAV is the potential for end-expiratory volume (EEV) to be different, relative to passive FRC, during the two measurements. During PAV, EEV is dynamically determined and may be higher than passive FRC (in the presence of DH) or lower than FRC (when active expiration is present in patients who are not flow-limited). To the extent that occlusion pressure (Pocc) during PAV includes the Pel that exists at the beginning of the breath, EPAV may be higher or lower than ECMV (when the latter is determined from passive FRC, as was the case here). Differences between ECMV and EPAV need not, therefore, reflect differences in methodology but may be, in part or in total, related to differences in EEV. Allowance for differences in EEV was only possible in patients with severe flow limitation where flow near the end of expiration can be used as a marker of EEV, relative to passive FRC. In the remaining patients, we assumed that EEV during PAV was not different from passive FRC. Although this correction was possible only in a minority of patients (10 patients), these were the patients most likely to show large increases in EEV in the dynamic state. The methods used to identify patients with severe flow limitation and to infer the extent of DH during PAV in these patients are detailed in section D of the online data supplement.

Determination of assist level. The assist levels dialed in during the study were based on preliminary, lumped R and E values obtained by the bedside. These values may not have been accurate and did not allow for the effect of differences in V and on E and R. We determined the actual assist level delivered after detailed determination of respiratory mechanics. The approach used to estimate actual assist level is described online in section E.

	RESULTS AND DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS AND DISCUSSION REFERENCES

A total of 6,812 occlusions were performed, during PAV, in 74 patients (x ± SD = 92 ± 68, range 12 to 497 per patient). The average duration of monitoring was 70.8 ± 46.8 min (range 8 to 220 min). The average frequency of performing occlusions was 1.6 ± 1.4 min¹. The procedure was well tolerated by all patients.

Time Course of Paw during Occlusions

Figure 3 illustrates the spectrum of Paw patterns observed during occlusions in PAV. Paw during the early part of occlusion either showed a small decline (Figure 3A), was flat (Figure 3B), or increased to different extents (Figures 3C to 3H). The pattern was consistent in a given patient at the same assist level. As indicated previously in "THEORY," these patterns reflect the time course of Pmus remaining at the beginning of occlusion (Pmus0), modified to a small extent by viscoelastic behavior. Thus, the pattern shown in Figure 3A indicates that Pmus0 was extremely small, so that the net response was dominated by viscoelastic behavior. The pattern of Figure 3B indicates that Pmus0 and viscoelastic behavior canceled each other out. To the extent that the change due to viscoelastic behavior is small (section B of online data supplement), Pmus0 in these cases was also small. The patterns in Figures 3C to 3H indicate more substantial Pmus0 that overwhelmed viscoelastic behavior. The magnitude of increase in Paw varied considerably with the largest change being in the 8 to 10 cm H₂O range (e.g., Figure 3F).

View larger version (69K): [in this window] [in a new window]   Figure 3.   Tracings of Paw showing the range of patterns observed with end-inspiratory occlusions during PAV. The beginning of occlusion (zero flow crossing) is indicated in each case by the event marker.

Figure 4 shows the relation between assist level and the magnitude of change in Paw in the interval between 0.05 and 0.30 s from the onset of occlusion. The outer limit of this interval was chosen because 0.3 s was the lowest common denominator of the duration of occlusions in all patients. Each point is the average of all observations at a given assist level in a given patient. The number of assist levels per patient ranged from one to four. It can be seen that occlusions associated with a decline in Paw (e.g., Figure 3A) occurred primarily at high levels of assist whereas those associated with large increases in Paw occurred during lower levels of assist. There was considerable scatter, however (r² = 0.49), particularly in the mid assist range. This scatter may be explained by the fact that Pmus0 is not only related to percent assist but also to E and volume above passive FRC (see section A of online data supplement), and these varied considerably among patients. Furthermore, the percent assist determined here reflects overall assist [i.e., Paw/(Pmus + Paw)] at a specific point during the inflation phase (section E online) and need not reflect exactly the percent assist of the elastic component at the end of the inflation phase. It is VA as percentage of E that is important in determining Pmus0 (section A online).

View larger version (16K): [in this window] [in a new window]   Figure 4.   Magnitude of increase in occlusion pressure at different levels of assist. The ordinate indicates the increase in Paw from 0.05 s to 0.30 s after the onset of occlusion. Each point is the average of all occlusions done in a given patient at a given level of assist. A decrease in Paw during the occlusion (negative values) was observed primarily at high levels of assist and that there is a significant negative correlation between the change in Paw and percent assist (r < 0.00001).

Figure 5A shows the average time course of Paw during the first 0.3 s of occlusion at different levels of assist. Nearly all patients had data in the 50 to 75% assist range whereas data for the very low assist range (0 to 25%) were available in only 14 patients. Figure 5A shows the actual Paw values relative to pressure at 0.05 s from the onset of occlusion. As also evident from Figure 4, the total amplitude (P.3-P.05) was inversely related to percent assist. The time course was, on average, flat when assist was 75 to 100%. At lower levels of assist, where P was positive, the rate of change in Paw ( i.e. Paw/t) progressively decreased with occlusion time. At the lowest assist, the average increase in Paw at 0.3 s was only 3.3 cm H₂O but there was considerable variability (0.2 to 6.9 cm H₂O).

View larger version (16K): [in this window] [in a new window]   Figure 5.   (A) Average change in Paw over the interval 0.05 to 0.30 s of occlusion, at different levels of assist. Between 75 and 100% assist, the occlusion pressure is essentially flat from the beginning of occlusion. (B) Relation between Paw during occlusion and the rate of change in pressure (Paw/t). Different symbols refer to different levels of assist (see inset). Diagonal isopleths indicate the time, during occlusion, at which values are reported. At each level of assist, there is a negative linear relation between actual pressure and the rate of change in pressure, indicating an exponential relationship. The Y intercept of these lines is the asymptote of pressure at the different levels of assist. At the highest level of assist, pressure was flat; this is denoted by the open circle at the origin.

A fundamental requirement of the proposed approach is that inspiratory Pmus should decline to near zero before the onset of behavioral responses or of phasic respiratory muscle contractions (e.g., onset of phasic expiratory muscle contraction or onset of the next inspiration). A critical question, therefore, is how long it takes Paw to reach a plateau when an important amount of inspiratory Pmus is present at the beginning of occlusion (e.g., Pmus0 > 2.0 cm H₂O). In most cases, when there was an important increase in Paw during the occlusion, the time course appeared to be exponential (e.g., Figures 3D and 3E). This is also reflected in the average time course of Paw shown in Figure 5A. To test for the goodness of an exponential fit, and if confirmed, to determine the time constant and Paw asymptote, we plotted the relation between instantaneous Paw and instantaneous Paw/t, as shown for the average data in Figure 5B. At the highest level of assist, there was no increase in Paw, and Paw/t was zero throughout (open circle). At lower levels, there was an inverse linear relation between Paw and Paw/t. The correlation coefficient was > 0.99 at all three levels, indicating that an exponential fit adequately describes the time course. With such plots, the intercept is an estimate of the asymptote and the slope is the time constant. The slope, representing the time constant of averaged data, was 0.15 s at all three levels. The intercept was higher than Paw @ 0.25 s by 0.3, 0.6, and 1.1 cm H₂O in the 50- 75, 25-50, and 0-25% assist, respectively. These results indicate that, on average, Paw @ 0.25 s may underestimate passive recoil but that the error is fairly small even at the lowest assist.

The same analysis was done on data from individual patients in whom a sufficient number of occlusions demonstrated an important increase in Paw during the occlusions. To do that we visually inspected the tracings of all occlusions associated with a Paw > 2 cm H₂O. Those in which there was a discontinuity in the trajectory of Paw time course, indicating appearance of new respiratory muscle contraction (see the next subsection, LATENCY OF EXTRANEOUS Paw CHANGES), were rejected. Paw values were measured at discrete points (every 0.05 s) during all remaining occlusions. These values were averaged for each patient. The relation between instantaneous Paw and instantaneous Paw/t was examined by linear regression. The time course of Paw was assumed to be exponential if the correlation coefficient was > 0.75 and the SE of the intercept was < 1.0 cm H₂O. Thirty-eight patients had an adequate number (> 10) of occlusions with Paw > 2.0 cm H₂O. In 32 of these patients, the time course of Paw fit an exponential function (r = 0.96 ± 0.06, SE of intercept = 0.29 ± 0.24 cm H₂O). The average time constant was 0.21 ± 0.06 s with a range of 0.09 to 0.34 s. Figure 3D is an example from the patient with an average time constant (0.20 s); Figures 3E and 3F represent the patients with the longest (0.34 s) and shortest (0.09 s) time constants. In these 32 patients, the average difference between the intercept (estimated asymptote) and actual pressure at 0.3 s was (± SD) 1.17 ± 0.38 cm H₂O and between the intercept and Paw at 0.25 s was 1.53 ± 0.92 cm H₂O. In the remaining six patients, the time course was not exponential but was nearly linear. In two of these, the linear phase ended rather abruptly in a plateau at approximately 0.5 s (e.g., Figure 3G). In the remaining four patients, the occlusion was not long enough to establish the time of appearance of the plateau; the occlusion durations were 0.30, 0.42, 0.50, and 0.63 s. Figure 3H is an example from the patient with linear behavior over the longest duration of occlusion. A linear increase in Paw over a relatively long period (Figure 3H) may represent a particularly long Pmus declining phase. Alternatively, in such patients the linear behavior of Paw may be a result of the decay in inspiratory Pmus merging imperceptibly with the onset of a phasic expiratory pressure ramp.

Latency of Extraneous Paw Changes

Estimates of passive recoil may be corrupted if extraneous forces cause Paw to deviate significantly from the course dictated by the natural decay of inspiratory Pmus. Such forces may be nonrespiratory (e.g., cardiac artifacts) or may be the result of "new" contractions involving the respiratory muscles. Several types of "new" contractions may occur in the course of end-inspiratory occlusions. Phasic activation of expiratory muscles may begin at some point after the occlusion causing Paw to rise and E to be overestimated. Conversely, if TE is short, the next inspiratory effort may begin soon after the onset of occlusion, resulting in underestimation of E. Apart from these scheduled events, unscheduled contractions may occur either as a behavioral response to the occlusion or as a component of motor acts in which the respiratory muscles participate (e.g., cough, hiccups, retching, body movements). Such acts may, fortuitously and at random, occur during the occlusion.

An assessment of the potential of these extraneous forces to corrupt the measurement was undertaken. Analysis was limited to the first 46 patients in whom the occlusion was relatively long (0.76 ± 0.31 s), thereby permitting assessment of the likelihood of occurrence of these events as a function of occlusion time. All occlusions from these patients were visually inspected (n = 3,425). An extraneous event was deemed to have occurred if the trajectory of Paw clearly deviated from the course established in the preceding segment, with this deviation resulting in at least a 1.0 cm H₂O difference from the extrapolated prior trajectory at any point in the next 0.25 s. An event was also called if the pattern of Paw was clearly different from that observed in the majority of neighboring occlusions. Whenever an event was identified, its latency, relative to the onset of occlusion, was measured.

Figure 6 illustrates the range of extraneous events observed. Cardiac artifacts were identified by their brevity ( 0.2 s) and regularity, with an interevent interval consistent with cardiac rhythm, and the occurrence of similar artifacts in the Paw or flow tracing during nonoccluded periods (Figure 6A). Cardiac artifacts were visible in only 20 of 46 patients. In all but one patient, their amplitude was < 1.0 cm H₂O. The tracing in Figure 6A is from the single patient with artifacts > 1.0 cm H₂O. As may be expected, their timing during the occlusion was random. Because of their random timing and their very small amplitude, their effect would, at most, be a small increase in the variability of estimated E. 

View larger version (74K): [in this window] [in a new window]   Figure 6.   Examples of different extraneous events identified during the study. In each panel the two event markers indicate the beginning and end of occlusion, as ascertained from the flow tracing. (A) Cardiac artifacts. (B) Onset of expiratory pressure ramp at arrow. (C1) Development of a large positive dome-shaped artifact during the occlusion. This is identified as an artifact because, in most other occlusions, the pattern was indicative of high level of assist. This is shown in Panel C2, which shows Paw during the occlusion immediately after the one shown in Panel C1. (D) Occurrence of a new inspiratory effort, at vertical arrow, during the occlusion. This is identified as the onset of a natural inspiratory effort because the latency for its appearance is similar to expiratory time in unoccluded breaths (note the width of the horizontal arrows). (E ) Occurrence of a very large, broad negative artifact during the occlusion. (F1) A brief negative pressure spike is observed early during an occlusion (first breath in the panel). This is believed to be a hiccup. Note that two breaths later, there is an inspiratory spike in an unoccluded breath. Several breaths later (F2), there is another sharp inspiratory spike during early expiration in an unobstructed breath.

Events related to "new" expiratory muscle contraction took two forms. In the most common form (Figure 6B) Paw/t increased relative to the preceding segment with this increase lasting to the end of the occlusion, resulting in progressively greater deviation from the expected Paw trajectory. The latency of these events was never less than 0.2 s. They either represent behavioral responses or the onset of scheduled phasic expiratory activity.

The other type of expiratory recruitment is shown in Figure 6C1. These were dome- or cone-shaped increases in Paw with a base of 0.4 to 1.5 s. They were distinguished from early decay of inspiratory Pmus followed by a new inspiratory effort (Figure 6E) by the fact that the Paw pattern in neighboring occlusions was consistent with minimal residual Pmus at the onset of occlusion (Figure 6C2), or by their appearance after Paw had reached a plateau. Their latency varied and a few had a very short latency (Figure 6C1). Only seven of these (out of 3,425 occlusions) began in the first 0.25 s, however. The mechanism responsible for this pattern is not clear. We suspect that they represent either behavioral responses (particularly when latency is > 0.2 s) or random motor acts (cough efforts, trunk movement, retching, or gagging).

Figure 6D shows a scheduled inspiratory effort beginning during the occlusion. It was deemed to be scheduled because its latency was similar to the interval between end of inflation phase and onset of next spontaneous inspiratory effort in unoccluded breaths (e.g., see preceding breath). The latency for this response depended on the duration of spontaneous expiration (and hence on respiratory rate). Two patients had average latencies of 0.36 ± 0.08 s (Figure 6D). Both patients had a high respiratory rate (approximately 40 to 45 min¹) when these events were observed and the occlusion began near the end of the declining phase of Pmus (as evidenced by minimal increase in Paw early in the occlusion), thereby leaving little time between the end of the inflation phase and the beginning of the next effort.

Figure 6E is an example of an inspiratory effort that was deemed to be unscheduled because its latency was much shorter than TE of unoccluded breaths. These appeared as an inverted dome or cone with a broad base (0.4 to 0.8 s). Their latency varied but none occurred in the first 0.25 s. These likely represent behavioral responses or random nonrespiratory acts.

Finally, Figure 6F1 shows a brief (< 0.4 s) negative transient in occluded Paw, indicating brief recruitment of inspiratory muscles. The latency of this artifact was variable, including extremely short latencies in a few cases (Figure 6F1). Patients with this artifact invariably displayed brief inspiratory transients during unoccluded inspirations and expirations (see last breath in Figure 6F1 and first breath of Figure 6F2, obtained, in the same patient, several breaths away from the occluded breath). We believe these are hiccups. Only six such events occurred during the first 0.25 s.

Figure 7 shows the cumulative probability of extraneous events (excluding cardiac artifacts) as a function of occlusion time. It is clear that in the majority of occlusions no events occurred even up to 1 s. This probably reflects the fact that in most patients, TE is longer than 1 s, there is little phasic expiratory activity, and behavioral responses are uncommon in view of the obtunded state of most patients. Nonetheless, the frequency of these events is not insignificant. The figure also shows a sharp increase in the rate of occurrence of extraneous events at approximately 0.25 s. In fact, only 34 events began in the first 0.25 s (from a total of 3,425 occlusions;  1.0%). Of these, 19 were gentle expiratory ramps (as in Figure 6B) that began between 0.20 and 0.25 s and attained insignificant amplitude by 0.25 s (< 1.0 cm H₂O). Only 15 (< 0.5%) were important in that they were associated with large changes in Paw (Figures 6C and 6F), and no more than two occurred in any given patient throughout the study (three patients had 2 and nine had 1).

View larger version (17K): [in this window] [in a new window]   Figure 7.   Kaplan-Meier plot of the cumulative probability of extraneous events as a function of occlusion time. Extraneous events are extremely rare in the first 0.25 s.

Optimal Time for Measuring Paw during Occlusion

Ideally, the optimal time is as far away from the onset of occlusion as possible, to permit full decay of inspiratory Pmus, but before the onset of extraneous events. Because of the wide range of latencies of extraneous events within and between patients, such an approach would require identification of the onset of such events in individual occlusions. We have attempted several algorithms to do so. None was successful, reflecting the great variety of patterns in uncontaminated occlusions (Figure 3) and of extraneous events (Figure 6). The more successful an algorithm was in detecting extraneous events, the more "good" occlusions it also rejected. Extraneous events can be readily identified by an experienced eye. However, this would preclude the use of this approach in the clinical setting.

The results of the present study point to a time window that should offer a satisfactory compromise in virtually all patients. We feel that sampling Paw at between 0.25 and 0.30 s offers a reasonable compromise, where the probability of important artifacts is sufficiently small (< 1%) that they can be easily filtered out, while underestimation of E should be negligible in most cases and relatively small (< 25%) in the worse case (for additional justification of this choice, see section F of online data supplement). As a result, we adopted a uniform sampling time of 0.25 s in all patients and assist levels. A comparison of E values obtained at high and low levels of assist, in the same patients, will be given next.

Comparison of E Values at Low and High Assist

In 25 patients, there were two periods, each with an adequate number of occlusions (> 10), separated by a wide margin in percent assist (> 25% of E and R). On average, the percent assist was 41 ± 14 and 78 ± 14% in the low and high assist, respectively. In these patients, as with all others, E was estimated from P_0.25 s in both the low and high assist periods. Figure 8 is a scatter plot comparing E values under the two conditions. There was an excellent correlation between E at low and at high assist (r = 0.98). Average E was 27.6 ± 10.7 and 30.8 ± 13.6 in the two groups with the difference being highly significant (p < 0.001). The magnitude of the difference was, on average (± SD), 3.2 ± 3.7 cm H₂O · ¹ (range 16.8 to +1.8 cm H₂O · L¹). Expressed as percentage of E at high assist, the difference was 9.3 ± 8.0% (range 21 to +7%). The greatest underestimation was, therefore, 21% (point with highest E in Figure 8).

View larger version (13K): [in this window] [in a new window]   Figure 8.   Relation between estimated E at high assist and low assist during PAV. Each point is a different patient.

The difference in E need not be entirely due to incomplete decay of Pmus with consequent underestimation of passive elastic recoil. There are other reasons that may have contributed to the observed difference. Thus, the two sets of measurement were obtained at different times. E may have changed in the interim. Likewise, the difference may in part be the result of a smaller VT during the low assist in patients in whom E is volume-dependent. VT during low assist was significantly smaller (0.47 ± 0.16 versus 0.51 ± 0.17 l, p < 0.001). Furthermore, there may be greater use of expiratory muscles during low assist causing elastic recoil at the beginning of the breath to be lower. That these factors played a role is supported by the fact that in the three patients with the largest difference between the two E values, a plateau had been reached by 0.25 s (e.g., Figure 3F, which belongs to the patient with the highest E, and largest difference, in Figure 8). Accordingly, incomplete decay of Pmus accounts for only a fraction of the differences observed in this comparison.

We were particularly interested in the results (of this comparison) in patients in whom Paw increased linearly during the first 0.25 s because in these patients the actual value of the plateau could, theoretically, be considerably higher than P_0.25. Five of these patients had two periods with substantially different assist levels and were, accordingly, included in this group. The average difference in E in these five patients was 3.4 ± 2.8 cm H₂O · L¹ (9.6 ± 6.2% of E at high assist) with the largest difference being 6.8 cm H₂O · L¹ (16.9%). Considering that other factors likely contributed to these differences, the error remains acceptably small, even in these patients. In section G of the online data supplement, we describe some measures that could further reduce the magnitude by which E is underestimated during low assist.

Comparison of E during CMV (ECMV) and PAV (EPAV)

Figure 9 is a scatter plot of the relation between ECMV and EPAV, after matching VT (all patients) and correcting for DH in 10 patients with flow limitation (see METHODS). The average (± SD) values were 28.8 ± 12.2 and 28.5 ± 10.1 cm H₂O · L¹ for ECMV and EPAV, respectively (not significant). There was an excellent correlation (r = 0.92) but there was a small positive intercept (6.5 cm H₂O · L¹) and the slope was significantly less than 1.0 (0.76). These results indicate that EPAV tended to exceed ECMV in the low E range. The average (± SD) difference between ECMV and EPAV was 0.3 ± 4.9 cm H₂O · L²¹, representing 0.9 ± 16.4% of average E (i.e., average of EPAV and ECMV).

View larger version (14K): [in this window] [in a new window]   Figure 9.   Relation between E estimated during PAV from brief end-inspiratory occlusions and E determined during CMV.

The differences observed between ECMV and EPAV need not be entirely related to incomplete relaxation at 0.25 s of the occlusion in the PAV measurements. The two sets of measurements were separated by 0.5 to 2.0 h and E may have changed in the interval. Furthermore, correction for differences in EEV was only done in 10 patients who had flow limitation (FL). EEV may have been different in some of the remaining patients owing to some DH (albeit in the absence of FL) or to lowering of EEV below passive FRC through active use of expiratory muscles.

The possible contribution of time-dependent changes in E and of differences in dynamic EEV was assessed in 24 patients who underwent two sets of CMV measurements but where the occlusions in the second set (i.e., post-PAV) were not preceded by slowing of the ventilator to return V to passive FRC. Two of these patients were included in the FL group. In them, ECMV in the second study (ECMV2) was 139% and 390% of ECMV1, emphasizing the large effect DH may have on estimated E in patients with FL. These two patients were eliminated from the current comparison. Figure 10 shows the relation between ECMV2 and ECMV1 in the remaining 22 patients. In 16 of these (solid symbols), there was no evidence of DH in CMV2 (expiratory flow before the onset of inflation < 0.10 L · s¹ without prior slowing of the ventilator). In the absence of DH in either study, these patients serve to illustrate time- dependent differences. The solid line in Figure 10 is the regression for these 16 patients only. As can be seen, although there was excellent agreement between the two sets of measurements (r = 0.95), there were some differences. The average difference was 0.1 ± 1.6 cm H₂O · L¹ (0.4 ± 5.9%), representing approximately one-third of the differences observed between ECMV1 and EPAV (Figure 9).

View larger version (14K): [in this window] [in a new window]   Figure 10.   Relation between E obtained during two periods of controlled ventilation, one before and one after PAV, in patients who had two such determinations. Solid circles are the results of patients who did not have DH during either measurement; open circles, relation in patients who did not have DH in the first study (because of slowing of the ventilator before the occluded breaths) but had DH in the second study. The solid line is the regression for the first group of patients. The dashed line is the regression for all patients.

In six patients, there was clear evidence of DH before the occluded breaths in CMV 2; preinspiratory flow ranged from 0.24 to 0.51 L · s¹ (0.34 ± 0.12). These patients are indicated by the open symbols in Figure 10. DH occurred principally in patients with low E. This is understandable because, for the same resistance, patients with low E have a longer expiratory time constant (i.e., R/E) and, all else being the same, are more likely to have incomplete emptying before the next inspiration. The impact of these few patients on the relation between ECMV1 and ECMV2 was pronounced. The dashed line (Figure 10) is the regression for all 22 patients, including the six patients with DH. The correlation deteriorated substantially (r = 0.77 versus 0.95) and, because of the preferential location of these patients in the low E range, there was a substantial intercept (11.5 cm H₂O · L¹) and the slope was much less than 1.0 (0.59). These changes in slope and intercept are in the same direction as observed in the relation between ECMV1 and EPAV (Figure 9). In fact, they are more pronounced. This may be attributed to the fact that spontaneously breathing patients (as in PAV) are better equipped to deal with DH through appropriate adjustment of TI/TE ratio and, in the absence of FL, through use of expiratory muscles.

The comparison between ECMV1 and ECMV2 indicates that the differences observed between ECMV1 and EPAV were, at least in part, due to differences in time of measurement and in EEV. To the extent that the total differences between ECMV1 and EPAV were small, we believe that any contribution incomplete relaxation (during occlusions in PAV) may have had to these differences must be very small.

In summary, we believe that measurement of Paw 0.25 s from the onset of end-inspiratory occlusion in the PAV mode provides a reliable estimate of passive elastic recoil at the prevailing volume. This conclusion is based on theoretical considerations of the phase relation between ventilator cycle and patient effort in the PAV mode and of the latencies for behavioral responses and phasic expiratory muscle activity. The validity of this approach was supported in the current study by demonstrating a fairly fast decay of inspiratory pressure in the early phase of occlusion, by virtual absence of extraneous events in the first 0.25 s, by insensitivity of E values to level of assist, and by good agreement with E measurement during CMV. Because success of this approach is predicated on synchrony between end of ventilator cycle and end of inspiratory effort, and on a fairly short decay time of inspiratory effort, this approach cannot be recommended in other modes of assisted ventilation, where such synchrony is not assured, or in unintubated patients, where decay of inspiratory effort may be more protracted than in the typical intubated ICU patient studied here.