INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS AND DISCUSSION REFERENCES

Noninvasive determination of respiratory system resistance (R) in the absence of respiratory muscle pressure output (Pmus), e.g., during controlled mechanical ventilation (CMV), presents little difficulty. Under these conditions the pressure provided by the ventilator is the only applied force and this force is dissipated against resistive and elastic elements which are functions of flow () and volume (V), respectively. Because and V can readily be measured noninvasively, the resistive and elastic properties can readily be determined from two or more points having different values of and V. Calculation of resistance can be made using either regression analysis (1) or, more commonly, using the interrupter technique (2, 3). These techniques, however, cannot be used in the presence of respiratory muscle effort, such as during assisted ventilation (3), because under these conditions the applied force includes a highly variable component generated by the patient (Pmus). At present, therefore, reliable determination of R requires either the insertion of an esophageal catheter, which adds another invasive intervention to already much instrumented patients, or the abolition of respiratory muscle efforts by paralysis or hyperventilation. The latter procedure is disruptive and labor-intensive.

The work described here was motivated by two considerations:

First, there is every reason to expect that resistance may change spontaneously, and substantially, in the course of mechanical ventilation. Large changes may occur as a result of bronchoconstriction, accumulation of secretions, or slippage of the tip of the endotracheal tube into the right mainstem bronchus. Patients on assisted ventilation often develop episodes of respiratory distress at the same level of support that was adequate previously. An increase in resistance is an obvious potential mechanism for such episodes. The availability of continuous estimates of resistance would make it possible to identify such a mechanism and to take appropriate corrective action. Apart from identifying a possible cause for respiratory distress, the occurrence of large swings in resistance during continuous monitoring of this variable, particularly if secretions and tube slippage can be excluded, would establish the presence of bronchial hyperreactivity and would help guide the use of bronchodilators.

Second, proportional assist ventilation (PAV) (4) has several physiologic advantages (4, 5), including automatic changes in level of support (i.e., airway pressure [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 (less barotrauma, less sedation, faster weaning) with implications for 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 elastance (E) and R on an ongoing basis; knowledge of E and R is required for proper and trouble-free implementation of PAV. The availability of a simple, noninvasive way to continuously monitor R, along with a correspondingly simple and noninvasive way to monitor E, would greatly 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. At present, PAV is used only in short-term trials supervised by clinicians with considerable expertise in physiology. Such studies cannot assess the potential of PAV in achieving real clinical benefits (e.g., decreased morbidity and mortality). Availability of online estimates of R would also make it possible to automatically adjust the flow assist component of PAV as R changes.

For a method that estimates resistance to be accurate and suitable for widespread clinical use in the intensive care unit (ICU) environment, it should satisfy certain practical requirements: (1) It should measure inspiratory resistance (RI) because it is RI that is relevant to the work of breathing. Resistance during expiration (RE) may greatly overestimate RI in patients with expiratory flow limitation. In addition, in the latter patients changes in RI need not be reflected in RE. (2) It should be independent of changes in Pmus that occur over the interval of the inspiratory phase during which the measurements are made. (3) Because resistance in intubated patients is flow-dependent (2), the method should ideally define the relation between R and . (4) The method should not include interventions that evoke behavioral responses or, if it does, all relevant measurements must be obtained within the latency of such responses. (5) Any interventions that may be required to obtain the measurements should be well tolerated by patients. (6) It should be possible to obtain measurements at frequent intervals. (7) It should be possible to automate the interventions and measurements. Otherwise, obtaining the information may entail an unacceptable increase in time and expertise of clinical staff. (8) The technical requirements of the pressure and flow measuring systems (i.e., frequency response, phase relation between Paw and signals) should not be so stringent as to require placement of the flow meter and pressure outlets near the patient, far away from the ventilator. To have to do so substantially complicates patient management. (9) The method should ideally be noninvasive.

In this communication, we describe a method that, we believe, meets all of these requirements. To put this approach in context, it is necessary to first review other noninvasive approaches that are potentially suitable during assisted ventilation. Only a very brief review is provided here. A more detailed account of these methods and their limitations is given in Section A of the online data supplement.

A number of approaches have been proposed to estimate RE in patients with spontaneous inspiratory efforts (6, 7). The use of RE as a surrogate for RI has several limitations, chief among which are possible corruption of the measurement by phasic expiratory muscle activity and the fact that RE may greatly exceed RI in patients with expiratory flow limitation.

The forced oscillation technique (FOT) of Dubois and coworkers (8) has been used extensively to measure resistance in spontaneously breathing nonintubated patients. Application of this method to mechanically ventilated patients poses substantial technical and interpretative challenges (see Section A of the online data supplement). Although these difficulties are currently being addressed by several investigators, and progress is being made (9), much additional work remains to be done before FOT can be accepted as a clinical tool.

The interrupter technique, introduced by Neergaard and Wirz (14), has also been extensively used to measure RI in spontaneously breathing, nonintubated patients. It entails very brief occlusions applied during inspiration. To provide reliable results in mechanically ventilated patients, the occluder must be placed very close to the endotracheal (ET) tube and Paw and must also be measured at that location (see Section A of the online data supplement). This is cumbersome and further complicates ventilator and patient management by clinical staff.

The approach we propose entails a transient reduction in during the early part of inspiration. It is not necessary to reduce (at the airway) to zero. Nor is it necessary to reach the new flow level instantly or to maintain it constant at the low level. These features make it possible to apply the technique using the ventilator's valve. The change in Pmus between the points of high and low is taken into account through appropriate utilization of the equation of motion. These same manipulations of the equation of motion make it possible to minimize the impact of the change in V between the two points such that precise knowledge of passive elastance is not necessary.

Theory

During assisted ventilation applied pressure is made up of two components, one provided by the ventilator (Paw) and one provided by the patient (Pmus). According to the equation of motion (15), applied pressure is dissipated against elastic, resistive, and inertial forces. Thus, during assisted ventilation:

(1)

where V is instantaneous volume relative to passive FRC, E is elastance, is instantaneous flow, R is resistance, ..V is flow acceleration (in L · s²), and I is inertance.

To the extent that Pmus at a given instant is not known, accurate elastance values may not be available, and V, relative to passive FRC, is also not known (in view of possible dynamic hyperinflation or active reduction in volume below FRC by expiratory muscles), it is not possible to solve Equation 1 for the resistive terms using a set of measurements made at one point during the inflation phase. For this reason, any approach to measure resistance during inflation in such patients must involve measurements at more than one point, having different flow values.

In the proposed approach Paw (and hence flow) is transiently reduced (pulse) during the inflation phase in the PAV mode (Figure 1) and primary measurements of Paw, , and V are made immediately before Paw and flow begin declining (T₀, Figure 1), at the trough of pressure during the negative pulse (TI, Figure 1), and at a point between T₀ and ventilator triggering (T₁, Figure 1). To facilitate the description of how these measured values are used to derive R, we shall initially treat R as a constant and consider the case where the pulse is designed such that the interval between T₀ and T₁ (i.e., T₁) and the interval between T₀ and T₁ (i.e., T₁) are equal and the average flow rates in these two intervals are also equal.

View larger version (64K): [in this window] [in a new window]   Figure 1.   Tracings illustrating an example of a negative pulse (second breath) and the relevant measurements. The arrows in the Paw tracing denote the three times at which pressure, , and V were measured. P 2 is the highest pressure reached beyond the negative pulse. The external pulse channel illustrates the shape of the negative voltage applied to result in the negative pulse. Pmus is the estimated pressure generated by the respiratory muscles and was calculated from the pressure, , and V channels using the equation of motion. Small arrows in Pmus channel denote calculation artifacts when flow is decreasing rapidly, both during the pulse and at the end of the breath. These artifacts are because we ignored inertial forces in the application of the equation of motion.

Equation 1 can be written for points T₁, T₀, and T_I as follows:

(2)

(3)

(4)

Subtracting Equation 2 from Equation 3, and Equation 4 from Equation 3, assigning X₁ to the difference (X₀  X₁) and X₁ to the difference (X₀  X₁) yields the following two first difference equations:

(5)

(6)

Adding Equations 5 and 6 yields:

(7)

In the PAV mode rarely exceeds 5 L · s², even during the early rising phase of flow, and at T₀ and T₁ are, by design, near zero. As a result, the maximum that ₁ can be is 5 L · s², and ₁ is zero. The bracketed term cannot, therefore, exceed 5 L · s². Inertial pressure losses at these levels of acceleration are negligible (16). Accordingly, the inertial term in Equation 7 can be ignored.

Because volume is rising continuously during the pulse (Figure 1), V₁ and V₁ have opposite signs and tend to cancel out. If the pulse is designed so that the two terms are equal, the entire volume term is reduced to zero and knowledge of E becomes unnecessary.

By contrast, because ₀ is higher than both ₁ and ₁ (Figure 1), both ₁ and ₁ are positive values. The sum of the two flow differences in Equation 7 is larger than either difference, and this provides for a robust term. The same applies to Paw, resulting in a robust Paw term.

The only important assumption in the proposed approach is that the rate of change in Pmus (i.e., Pmus/t) is constant in the interval T₁ to T₁. This seems to be reasonable if the time interval in question is fairly brief (e.g., 0.2 to 0.3 s) and particularly if the first 0.3 s of inspiratory effort is avoided (see Potential Sources of Error). With this assumption it is possible to eliminate the Pmus term, leaving only the flow and Paw terms in the equation. Thus, if Pmus is continuously rising (or continuously falling) in the interval T₁ to T₁, Pmus₁ and Pmus₁ will, as in the case of volume, have opposite signs and tend to cancel out. If, in addition, T₁ and T₁ are made equal, the entire Pmus term drops out, leaving only one unknown (i.e., R), in the equation.

The foregoing analysis entailed some simplifications that, in practice, may not be feasible. Thus, in mechanically ventilated patients, R is flow-dependent (2) and cannot be assumed constant. Furthermore, it may be difficult to arrange it such that T₁ and T₁ are equal. Particularly when the ventilator cannot precisely control the time of occurrence of peak and trough of the negative pulse (i.e., T₀ and T₁), situations may occur where going back from T₀ a distance that equals T₁ places T₁ in the triggering region, where artifacts frequently occur, or too close to the onset of inspiratory effort, where Pmus/t is less likely to be constant. Also, arranging for average flow rates, during T₁ and T₁, to be identical would be technically very demanding, particularly in the PAV mode where flow in the interval T₁ is not predetermined but varies with patient effort. These practical limitations necessitated certain modifications to the mathematical approach. These are described in detail in Section B of the online data supplement. The following equation allows for nonlinearities in the pressure-flow relation and for inequalities in the two time intervals and in average flow rates in these intervals:

(8)

where K₁ and K₂ are the constants for laminar flow and turbulent flow, respectively, in Rohrer's equation and T_r is the ratio T₁/T₁. Note that if T_r is different from unity, or average flow rates in T₁ and T₁ are not the same, the volume term cannot be entirely eliminated. However, as discussed earlier, V₁ and V₁ continue to have opposite signs and tend to cancel out. Unless the time intervals, or the average flow rates in these intervals, are substantially different from each other the net volume term is sufficiently small (a few milliliters) that precise knowledge of E is not essential. Thus, if the value of E is not known, a default value (representing, for example, average E in mechanically ventilated patients) can be used with little risk of important errors.

Derivation of resistance. From each applied pulse an equation of the following form results:

(9)

where X is the flow term (first bracketed term in Equation 8), Y is the ² term (second bracketed term in Equation 8), and Z is the resistive pressure (Pres) term (right side of Equation 8). To obtain Z a known value of E is used or, in the absence of this information, a default value of 28 cm H₂O · L¹, representing average E in mechanically ventilated patients (personal observations), may be used. Resistance can be obtained from Equation 9 in one of two ways: (1) If a range of and ² terms is obtained in successive pulses, either spontaneously or by design (e.g., initiating the pulse at different or reducing by different amounts), KI and K₂ can be obtained by regression analysis. (2) In the absence of reliable, directly determined K₁ and K₂ values, in accordance with approach number 1, K₂ can be assumed to equal K₂ of the ET tube and Equation 9 is solved for K₁. Thus, K₁ = [Z  (Y · K₂ ET)]/X. The K₂ values of clean ET tubes of different sizes are widely available. Resistance can be reported as K₁ + K₂ET, reflecting resistance at a standard flow of 1.0 L · s¹. The resistance so reported may differ from actual resistance at 1 L · s¹ to the extent that actual K₂ may differ from the assumed K₂ of a clean tube and the flow at which R estimates are made is different from 1.0 L · s¹. The error in estimated resistance (at 1 L · s¹) if actual K₂ (K₂ actual) is different from assumed K₂ is given by:

(10)

It can be seen that the error in estimating R at 1 L · s¹ using an assumed K₂ is a fraction of the difference between the actual and assumed K₂ value. The magnitude of this potential error will be assessed experimentally.

What resistance is being measured? Observations of the response to flow interruption during CMV at constant flow have led to the concept that total respiratory system resistance is made up of a strictly flow-related component, produced primarily by airway resistance, and a time-dependent component related to viscoelastic behavior of the lung and, to a smaller extent, the chest wall (2, 17). The flow-related component is expressed by the rapid drop in Paw at the onset of flow interruption. Resistance calculated from this initial P is referred to as R_min. The time-dependent component is expressed by the subsequent gradual decline in Paw, at constant volume, over the following few seconds and is referred to as R. The sum of the two components is referred to as R_max. Because the time interval between peak flow and trough flow (i.e., T₁) in the proposed technique (0.1 to 0.15 s) is small relative to the time required for viscoelastic behavior to be fully expressed (usually several seconds), it would seem reasonable to suggest that RIpulse will be much closer to R_min than to R_max. This was confirmed by model analysis provided in Section C of the online data supplement.

Potential sources of error. These are discussed briefly here. More details are provided in Section D of the online data supplement:

1. Measurement noise: In mechanically ventilated patients the Paw and signals are subject to noise from many sources. These can be reduced by a variety of approaches: (a) ensuring that the change in flow produced by the intervention (i.e., ₁) is large relative to the amplitude of the noise; (b) elimination of sources of noise to the extent possible; (c) critical filtering of the Paw and signals; (d) pulse should not begin when a specified flow is reached; (e) averaging the resistance results obtained from a number of pulses.

2. Difference in response characteristics of Paw and measuring systems: Difference in response characteristics of the measuring systems would cause the peak and trough of the measured pressure to occur at different times relative to the flow signal even if the peaks and troughs of the two signals were, in reality, simultaneous. If T₀ is taken at the time of peak Paw, flow at T₀ will underestimate real flow, and vice versa. To minimize the impact of these differences, the phase lag between the Paw and flow measuring systems should be as short as possible. In addition, the pulse should be designed so that flow is fairly flat over a 30- to 40-ms interval in the vicinity of T₁ and T₀.

3. Errors related to extrapolation of the Pmus trajectory: These are potentially the most serious, particularly when respiratory drive and, hence, rate of rise of Pmus (i.e., Pmus/ t), are high. Our approach involves the assumption that Pmus/ t is constant over the interval T₁ to T₁. This assumption can be in error for a variety of reasons. These, and possible ways to minimize these potential errors, are discussed next.

a. Termination of inspiratory effort (neural T_i) during the pulse: This can potentially produce the largest errors in estimated R. Should it occur, Pmus would actually fall between T₀ and T₁, instead of continuing to rise, with the possibility of greatly overestimating Pmus at T₁ and, hence, greatly underestimating resistance.

Because of the potentially large magnitude of this error, it is necessary to ensure that peak Pmus (end of Ti) does not occur between T₁ and T₁. This is easy to accomplish during PAV (4). In this mode the ventilator provides assist in proportion to instantaneous Pmus and the end of ventilator cycle is automatically synchronized with patient effort and is constrained to occur during the declining phase of Pmus (4). So long as pulses are not delivered in the last fraction (approximately 30%) of ventilator Ti, one is assured that neural Ti termination did not occur within the pulse. With pressure support ventilation (PSV) and assisted volume-cycled ventilation such synchrony is not assured, however, and Ti may terminate at any point within or even beyond the inflation phase.

b. Shape of the rising phase of Pmus: The rate of rise of Pmus during the rising phase in humans is not constant (6). Differences between Pmus/t in the interval T₀ to T₁ (i.e., T₁) and T₁ to T₀ (i.e., T₁) would cause errors in estimated R for the same reasons discussed under (a) above. Such errors can be minimized by delivering the pulse beyond the first 0.3 s of inspiratory effort. The rationale for this conclusion is given in Section D3b of the online data supplement.

c. Behavioral responses: The change in Pmus after the initiation of the pulse may deviate dramatically from that expected from the preceding time interval if the patient perceives the pulse and reacts behaviorally to it. We (18) and others (19) have shown previously that the minimal latency for behavioral responses to changes in Paw and flow is approximately 0.2 s in very alert normal subjects. It follows that errors related to perception of the pulse, with consequent behavioral responses, can be avoided if measurements are restricted to the 0.2-s interval after initiation of the pulse. Behavioral responses can, however, occur without perception if the change is anticipated. The occurrence of anticipatory responses can be minimized by randomizing the order of pulse applications.

d. Nonbehavioral neuromuscular responses to changes in flow: The rapid reduction in flow in the course of an ongoing inspiratory effort may, theoretically, elicit reflex changes in neural output with much shorter latencies than behavioral responses. In addition, the change in flow and, consequently, in time course of volume, may elicit changes in Pmus, independent of changes in electrical activation, through the operation of the intrinsic properties of respiratory muscles (force-length and force-velocity relations). An important contribution from either of these responses after the onset of the pulse (between T₀ and T₁) could alter the time course of Pmus relative to the course predicted from the prepulse interval and introduce errors in estimated Pres. We believe, however, that the impact of these responses on estimated resistance should be minimal, particularly if T₁ is brief (e.g., 0.1 s) and the pulse is delivered fairly early in inspiration, where Pmus is relatively low (see Section D3d in the online data supplement).

In summary, the foregoing considerations led us to believe that transient reductions in flow, with estimation of the change in Pmus during the transient from data obtained in the immediately preceding period, have a reasonable likelihood of providing reliable estimates of airway resistance. These considerations have also pointed out a variety of ways by which systematic and random errors may be minimized. These include the use of the PAV mode, application of the pulse as early as possible during inspiration but avoiding the first 0.3 s of inspiratory effort, ensuring that the interval between T₀ and T₁ is < 0.2 s, randomizing the order of pulse application, maximizing the drop in between T₀ and T₁, and designing the pulse so that there is a nearly flat flow region in the vicinity of T₀ and T₁. In the current study, the proposed approach was applied using, as much as possible, the aforementioned features. Subprotocols were incorporated to test the impact of timing of pulse application, level of assist, and different signal filtering techniques on the results. The results were compared with resistance values obtained during CMV. Testing was done on a large number of patients (71 patients) with a very wide range of clinical problems, mechanical properties, and level of alertness. In addition, in the last 30 patients, the technique was applied using a fully automated system to simulate application under field conditions (suction, patient repositioning, and other ICU interventions and situations), and monitoring was continued for relatively long periods (hours) to permit determination of trends in resistance under typical ICU conditions.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS AND DISCUSSION REFERENCES

Patients

Seventy-one ventilator-dependent patients in the medical ICU were studied. Patients were not preselected on the basis of disease type or acuity. The only inclusion criteria were: (1) ability of patient, or availability of next of kin, to provide informed consent; (2) availability of the respiratory therapist who was trained to carry out the study (K.W.); (3) patient can be ventilated in the PAV mode as demonstrated by adequate gas exchange and comfort in the PSV mode or, if in the volume-cycled mode, by presence of triggering efforts or development of such efforts when assist level was reduced. All patients had been placed on mechanical ventilation because of acute, or acute on chronic, respiratory failure. The specific pathology covered the usual spectrum to be found in a general medical ICU. Duration on mechanical ventilation ranged from 1 d to 6 wk. Ten patients had received a tracheostomy because of > 2 wk of intubation. The remaining 61 patients were ventilated using ET tubes. ET tube size was no. 8 in 53 patients, no. 7 in 12 patients, no. 7.5 in four patients, no. 9 in one patient, and no. 6.5 in one patient. Level of consciousness ranged from totally alert to comatose. The protocol was approved by the University of Manitoba, Faculty of Medicine Ethics Board.

Apparatus

The ventilator used was a custom-built, piston-based, flow-triggered multimode ventilator capable of delivering PAV (The Winnipeg Ventilator; University of Manitoba, Winnipeg, Canada). More details about this ventilator can be found elsewhere (20). Flow was measured by a pneumotachograph, linear to 160 L/min (3600 series; Hans Rudolph, Kansas City, MO), inserted between the Y connector and the ET tube. Volume was obtained by electrical integration of flow. Paw was measured from a side port between the pneumotachograph and the ET tube. Flow and Paw were passed through an 8-Hz low-pass filter. The frequency response of the two signals was identical. In response to a square-wave pressure change the output of the two measuring systems changed with a time constant of 20 ms.

The Winnipeg Ventilator was modified so that its pressure output could be increased or decreased, relative to what is dictated by the PAV algorithm, using an external input. The external input took the form of a brief (0.2-s) negative voltage that was gated to the ventilator after an adjustable delay from the onset of inspiration (zero flow crossing). The shape of the negative external pulse was designed to result in a trough in Paw and flow between 100 and 200 ms from the onset of the pulse. Given the response characteristics of the ventilator this required a trapezoidal external input of the pattern shown in Figure 1. The rate of increase in voltage was adjusted to produce relatively broad Paw and flow troughs. Pulses were gated to the ventilator at random intervals. In the first 41 patients, gating was done by manually activating an external switch. In the remaining 30 patients, gating was carried out by a microcontroller, using a random number generator. Signals corresponding to Paw, , V, and the output of the pulse generator (Figure 1) were digitized at 125 or 200 Hz and stored on computer using a CODAS data acquisition system (DATAQ Instruments, Akron, OH). The signals were also continuously displayed on the computer monitor during the study.

Protocol

Positive end-expiratory pressure (PEEP) and fraction of inspired oxygen (FI_O2) were maintained at the levels used before the study. The first step was to determine passive respiratory mechanics during a period of CMV. The patient was placed in the volume-cycled mode, at a tidal volume slightly greater than the one received before the study. Back-up rate was increased in steps until respiratory efforts ceased, as evidenced by lack of trigger artifacts, a reproducible Paw waveform during inflation, and a monotonic decline in expiratory flow in the deflation phase (21). A series of end-inspiratory plateau maneuvers was performed. Before each plateau maneuver the ventilator rate was reduced to permit expiratory flow to approach zero. In most cases, the flow or volume of the breath to be occluded was temporarily changed to permit determination of passive mechanics over a range of flows and tidal volumes.

Preliminary estimates of passive E and R were then made from the stored CMV data using standard equations (2) [E = (P_platPEEP)/V_t and R = (P_peak  P_plat)/]. The ventilator was then switched to PAV. Volume assist (VA) and flow assist (FA) were set initially at 80% of the preliminary estimates of E and R, respectively. Five to 10 min were allowed for the patient to adjust to the new mode. Thereafter, pulse application began. In the first 41 studies, pulse characteristics were adjusted manually using the monitor display for feedback. Pulse delay was set to initiate the pulse during the rising phase of flow. Pulse amplitude was adjusted so that flow at the trough was near zero but not low enough to reset the ventilator and terminate the inflation phase. These characteristics were subsequently not changed. In the last 30 experiments determination of pulse characteristics was relegated to a microcontroller and these characteristics were adjusted automatically from time to time to obtain the desired pulse characteristics, in particular to ensure that flow decreased close to, but not to, zero during the pulse. This was done because the earlier set of studies demonstrated that the reduction in flow produced by a constant external voltage amplitude varied from time to time during data collection. Thus, a pulse amplitude that resulted in an optimal decrease in flow at one point would, at a later time, result in very small decreases in flow, or too much reduction in flow which reset the ventilator.

In addition to delivery of negative pulses, in all but nine patients brief (0.3-s) end-inspiratory occlusions were applied to some breaths. The sequence of delivering the two interventions was randomized. The average inter-intervention number of breaths was 9 with a range of 4 to 15. In the remaining nine patients end-inspiratory occlusions were done during a separate PAV period preceding pulse delivery. The end-inspiratory occlusions were used to determine elastance on PAV (see ANALYSIS).

In early experiments data collection continued only until 10 to 15 pulses were obtained. Later, the duration of data collection was gradually increased. In the final 30 to 40 studies data collection was routinely continued for more than 1 h (up to 3.5 h). The increase in the duration of the experiment was to permit the assessment of time-dependent changes in resistance and the impact of routine ICU interventions (particularly coughing and suction) on the measurements. With the longer experiments we also determined whether changes in level of PAV support, and hence in Pmus generated by the patient, had an impact on the results. For this purpose, after 20 to 30 pulses were obtained at high level of assist, the assist was reduced, usually to the lowest level deemed by the respiratory therapist to be consistent with patient comfort, and data collection continued. In some experiments pulses were delivered at two specific delays to assess the impact of timing of pulse application on the results. The two delays were not intermingled. Rather, brief delays were applied for a period of time (usually 20 pulses) followed by another set of 20 pulses using the longer delay.

In 24 patients a second period of CMV was carried out at the end of pulse data collection. During this period passive mechanics were assessed again.

Analysis

CMV data. Occluded breaths displaying evidence of respiratory muscle effort were rejected. All remaining breaths were analyzed as follows: Paw and inspired volume were measured at the point where flow reached 0.05 L/s in the early part of the plateau (P_plat, VT). Justification for the use of this point to measure RI in CMV is given in Section E1 of the online data supplement. Peak flow (_peak) and Paw and inspired volume at the time of peak flow were determined (P_peak, V_peak). V_peak is invariably less than VT in view of continued inspiratory flow in the early part of occlusion (22, 23). Passive E was calculated from (P_plat  PEEP)/VT. Passive resistance (Rcmv) at _peak was calculated from: Rcmv = [P_peak  P_plat + (VT  V_peak) E]/_peak. This modification of the standard equation allows for the difference in volume, and hence elastic recoil pressure, between the time of peak flow and the time of P_plat (22, 23).

According to Rohrer's equation: R = K₁ + K₂. In all patients we estimated K_I by assuming that K₂ was similar to the value reported for a clean ET tube. Thus:

(11)

where average Rcmv is the average of Rcmv values obtained from all nonrejected occlusions, average _peak is the average peak flow in the same occlusions, and K₂ET is the K₂ value reported for the ET tube size used in the patient. The latter values were obtained from Wright and coworkers (24) and are as follows: ET no. 9 = 2.0, ET no. 8 = 5.5, ET no. 7.5 = 7.0, ET no. 7 = 9.5, ET no. 6.5 = 15.

When the CMV data included a wide range of _peak (coefficient of variation [CV] > 0.20), linear regression analysis was done between Rcmv and _peak. The slope of this relation provided actual K₂ (as opposed to an assumed K₂ of a clean tube) and the intercept represented actual K₁. K₂ was considered different from K₂ ET if the 95% confidence interval (CI) of the slope was narrow (i.e., < 30% of slope value or < 3 cm H₂O · L² · s²) and did not include the K₂ value of a clean tube.

PULSE data. Analysis was done with a custom program. Each breath receiving a pulse was identified from the gated pulse channel (Figure 1). The point at which Paw changed from rising to falling (Paw/t = 0) was identified (T₀). The point at which Paw reached its lowest value within the pulse period was also identified (T₁). Paw, , and V were measured at T₀, T₁, and at a point 100 ms before to T₀ (T₁, Figure 1), and were tabulated. Measurements were made either from discrete data points at these three times (unfiltered data) or from a moving average of the respective channels. The averaging interval was selected by the user (e.g., ± 8 ms, ± 16 ms, ± 24 ms, etc.). Separate tables were kept for unfiltered data and for data obtained with each averaging interval. The time interval between T₀ and T₁ (i.e., T₁) was also tabulated. The highest Paw value between T₁ and the end of inspiration was also noted (P₂, Figure 1). The pulse was rejected if:

a) (_0  1) < 0.05 L · s¹. This criterion was to ensure that analysis was limited to pulses applied during the rising phase of flow and, hence, Pmus.

b) (P₂  P₁) < 3 cm H₂O. This criterion was to ensure that the inflation phase resumed after the withdrawal of the pulse. Failure of Paw to rise again may be due to ventilator resetting during the pulse (flow decreasing below triggering off level) or, in the PAV mode, to termination of the patient's neural Ti during the pulse. In either case the T₁ values are corrupted.

Equation 8 was solved for each unrejected pulse using an elastance value obtained from end-inspiratory occlusions during PAV [(P_plat  PEEP)/VT]. We have shown earlier that this agrees well with elastance obtained during CMV (25). An equation having the form of Equation 9 was thus generated for each pulse (K₁ · X + K₂ · Y = Z). K₁ was calculated by assuming K₂ = K₂ET and inspiratory resistance at a standard flow of 1 L · s¹ (RI) was calculated from K₁ + K₂ET.

The same analysis was done on data obtained from filtered Paw, , and V signals. Unless otherwise indicated, all resistance values (RI) will be reported as [K₁ + K₂ET].

When all pulses in a given patient were analyzed, the mean, SD, and CV of RI were obtained separately for unfiltered and filtered data. The impact of filtering on average RI and on between-pulse variability could then be assessed. When more than one level of assist was applied in a given patient, the results were also averaged separately for each assist level.

In studies lasting more than 30 min the moving average of RI was obtained to characterize time-dependent trends. For this purpose a 10-pulse moving average was generated for the term, ² term, and Pres. The moving average of K₁ was estimated from these values, assuming K₂ = K₂ ET.

Supplementary analysis. Two supplementary approaches were used to explore possible reasons for the difference between RI obtained during CMV and during PAV:

1. Time-dependent differences in resistance: To the extent that CMV and PAV were applied at different times, we felt that differences in observed RI may, in part, be real (i.e., not due to difference in technique). This possibility was explored by using RE as a surrogate for RI. Accordingly, RE was calculated during CMV and during PAV and, provided certain conditions were met, the difference in RE between the two conditions was used to correct RI. The methods used to estimate RE and the conditions under which RE was accepted as a surrogate for RI are given in Section E2 of the online data supplement.

2. Differences related to the presence and magnitude of inspiratory muscle activity. As indicated earlier, pulse application may actively induce changes in Pmus. Such changes would alter the values of calculated resistance relative to the passive state. To the extent that these responses are a function of intensity of respiratory muscle contraction (26), differences between active and passive resistance are expected to be related to level of Pmus at the time of pulse application. To assess the extent to which level of respiratory muscle contraction results in differences between active and passive resistance, we estimated Pmus as described in Section E3 of the online data supplement. An example of the calculated Pmus tracing is shown in Figure 1. The onset of inspiratory effort was visually identified from the point at which Pmus clearly deviated upwards from its baseline value. Pmus at the onset of the pulse (i.e., T₀) was calculated as the difference in Pmus between T₀ and onset of inspiratory effort. These tracings were also used to determine the delay for pulse application, relative to onset of neural inspiration.

Group averages will be reported as mean ± SD, unless otherwise indicated. Results of comparisons between resistance values under different conditions are given as mean difference with 95% CI (± 1.96 SD), as recommended by Bland and Altman (27). Differences are expressed in absolute resistance units and as percentage of the average of the two values being compared (27). The slope, intercept, and correlation coefficient will be given as well.

	RESULTS AND DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS AND DISCUSSION REFERENCES

CMV Data

In four patients, review of the stored data failed to identify any occluded breaths that met the criteria of passivity. In the absence of a reliable reference with which to compare pulse data, the results of these patients were not analyzed further. In the remaining 67 patients elastance averaged (± SD) 28.6 ± 9.2 cm H₂O · L¹ (range 12.2 to 64.4 cm H₂O · L¹). RI (i.e., K₁ + K₂ ET) averaged 14.8 ± 4.2 (range 7.7 to 25.8 cm H₂O · L¹ · s). In 29 patients a sufficiently wide flow range (CV > 0.20) was tested and the regression between flow and RI was sufficiently tight to estimate K₂ with confidence. In 11 of these patients K₂ was not significantly different from K₂ ET. In the remaining 18 patients, K₂ obtained from the regression was significantly different from K₂ ET. In these patients the average K₂ ET was 6.5 (± 1.7) whereas the actual K₂ was 12.7 (± 2.8) cm H₂O · L² · s². The average (± SD) difference was 6.2 ± 2.5 cm H₂O · L² · s² and the largest difference was 13.4 (K₂ of a clean tube = 5.5, actual K₂ = 18.9 cm H₂O · L² · s²).

Data on the relation between RE and RI during CMV were available in all patients. In 47 patients flow-corrected RE (i.e., K₁ + K₂ ET) was within 25% of RI and did not increase significantly at lower volumes. In the remaining 20 patients, RE was considerably higher than RI or increased significantly at lower volumes. In these patients RE averaged (± SD) 25.6 ± 14.2 cm H₂O · L¹ · s, representing 157 ± 47% of RI (range 115 to 285%).

PAV Data

The average duration of data collection was 73.5 ± 52.0 min (range 4.5 to 222.5 min). Average elastance during PAV was 27.2 ± 9.1 cm H₂O · L¹, not significantly different from the CMV values. The number of pulses examined per patient averaged 63.1 ± 67.2 (range 9 to 227). 10.5 ± 28.6% of pulses were rejected because the ventilator reset during the pulse interval. This occurred usually when the amplitude of the external negative pulse was too large, resulting in flow decreasing below the reset flow limit of the ventilator (0.05 to 0.1 L/s). Another 15.9 ± 20.4% of pulses were rejected because they were delivered too late in the inflation phase and occurred during the declining phase of flow. These events were primarily encountered in early studies in which different delays were used to assess the effect of timing of pulse application on the results.

Pulse characteristics. The average delay between onset of inspiratory flow and pulse application was 0.23 ± 0.07 s. This corresponded to 26 ± 8% of T_i. When measured from the onset of neural Ti, as detected from the Pmus tracing, the average delay was 0.37 ± 0.1 s, corresponding to 35 ± 10% of Ti.

The average flow at which the negative pulse was initiated (i.e., at T₀) was 0.60 ± 0.17 L · s¹. The average flow at the trough of the pulse (i.e., at TI) was 0.29 ± 0.15 L · s¹. The average reduction in flow during the pulse was 0.31 ± 0.11 L · s¹. The average time interval between T₀ and T₁ was 0.12 ± 0.03 s. All measurements relevant to calculation of resistance were, accordingly, completed within the latency of behavioral respiratory responses (18, 19). The pulses were well tolerated by all patients.

Comparison of Pulse and CMV Resistance

The average RI obtained with the pulse technique was 16.2 ± 4.9 cm H₂O · L¹ · s compared with 14.8 ± 4.2 cm H₂O · L¹ · s during CMV (p = 0.0002 by paired t test, n = 67). Figure 2 is a scatter plot of the results obtained with the two techniques. There was a highly significant correlation between the two measurements (p = 5.0 E15). The slope was not significantly different from 1.0 but the intercept was significantly higher than zero (p = 0.02).

View larger version (18K): [in this window] [in a new window]   Figure 2.   Scatter plot and line identity of the relation between passive inspiratory resistance obtained during controlled mechanical ventilation (RIcmv) and using the pulse technique during PAV (RIpulse). Each point represents the average data from a single patient (n = 67 patients).

Notwithstanding the good correlation, there was considerable scatter. The average difference (± SD) was 1.3 ± 3.0 cm H₂O · L¹ · s with a 95% CI of 4.7 to 7.2 cm H₂O · L¹ · s. These values correspond to 7.8 ± 18.2% of average RI with a 95% CI of 27.9 to 43.4%. There are several reasons that may account for the differences between the two measurements:

True differences in passive resistance. The CMV and pulse data were obtained at different times and it is quite possible that resistance changed in the interim. There are reasons to believe that resistance did change, at times substantially, over relatively short periods independent of technique, in these patients. First, in the 24 patients in whom duplicate CMV measurements were made at two different times, once before and once after PAV, a comparison of the results from the two CMV sets also showed considerable scatter (Figure 3). The average (± SD) difference was 1.1 ± 2.8 cm H₂O · L¹ · s (7.8 ± 19.0%) with a CI of 4.4 to 6.6 cm H₂O · L¹ · s (29.5 to 45.1%). These differences are comparable to those of the RI_pulse versus RI_cmv relation (Figure 2). Second, in patients in whom monitoring of resistance with the pulse technique spanned a reasonably long interval, substantial spontaneous changes in measured RIpulse were often observed (e.g., Figure 4; see Time-dependent Changes in RI_pulse).

View larger version (16K): [in this window] [in a new window]   Figure 3.   Scatter plot and line of identity of the relation between passive resistance measured during two periods of controlled ventilation, one before (RIcmv1) and one after (RIcmv2) pulse measurements in 24 patients in whom duplicate CMV measurements were made. Note that the scatter is of the same order as that observed in comparisons between passive data and pulse data (Figure 2).

View larger version (27K): [in this window] [in a new window]   Figure 4.   Moving average of resistance values (K1 + K2ET) obtained over a 2 1/2 hour period of monitoring in one patient. Solid symbols represent data from the pulse technique. Open circles represent data obtained during early expiration after the occlusion maneuvers. Long vertical arrows denote suction. Short arrow denotes spontaneous coughing. Note the substantial fluctuations in the resistance values with both techniques and the rapid decrease in resistance after suction.

To assess the possible contribution of true changes in resistance to the differences observed between RI_pulse and RI_cmv, we used the results of RE in those patients in whom this approach was feasible. RE could not be used in 20 patients because it was deemed not to be representative of RI (see CMV data). The approach could also not be used in nine patients because no inspiratory hold maneuvers were carried out during the collection of the pulse data, and in another three patients because there was no period during which PAV assist was high enough to produce optimal conditions for measuring RE.

Figure 5A shows the relation between RE determined during CMV and RE determined later during PAV. The differences were bidirectional. The average difference (± SD) was 0.7 ± 2.6 cm H₂O · L¹ · s (3.9 ± 16.8%) with a 95% CI of 4.5 to 5.8 cm H₂O · L¹ · s (29.0 to 36.9%). The range of these differences was similar to that observed in comparisons between CMV₁ and CMV₂ (Figure 3) (SD 2.6 versus 2.8). In patients in whom a lengthy period of pulse data collection was available, changes in RE closely mirrored the changes in Rpulse (e.g., Figure 4, open symbols). This suggests that both methods were reflecting true changes in resistance.

View larger version (17K): [in this window] [in a new window]   Figure 5.   (A) Scatter plot of RE measured during controlled ventilation (REcmv) and RE measured later during PAV (REpav). The solid line is the line of identity. (B) The relation between RI measured during CMV and during PAV with the pulse technique after correcting CMV resistance for time-dependent changes indicated by RE (as shown in A). Note the improvement in correlation relative to the data of Figure 2.

To assess the extent to which time-dependent changes in resistance contributed to the differences observed between RI_cmv and RI_pulse (i.e., as shown in Figure 2), the values of RI_cmv were corrected by an amount corresponding to the change in RE in those patients in whom RE data were available (n = 35). The new relation between corrected RI_cmv and RI_pulse is shown in Figure 5B. As can be seen, the correlation improved considerably with the correlation coefficient increasing from 0.78 to 0.90 (r² = 0.82). The mean difference was 1.0 ± 2.0 cm H₂O · L¹ · s (6.1 ± 13.5%) with a 95% CI of 3.0 to 5.0 cm H₂O · L¹ · s (20.3 to 32.4%). This improvement indicates that the differences observed initially between RI_cmv and RI_pulse were, to a considerable extent, related to true changes in resistance.

Differences between actual and assumed K₂. In comparisons between RI_cmv and RI_pulse (e.g., Figure 2 and 5B) all resistance values were adjusted to a standard flow of 1.0 L · s¹ assuming that K₂ was similar to that of a clean, in vitro, ET tube. In the event actual K₂ was different, owing to secretions or kinking, for example, false differences between RI_cmv and RI_pulse can result with a magnitude that depends on differences between actual K₂ and K₂ET and on differences between the average flow during CMV and pulse measurements (Equation 10). It was possible to assess the impact of this potential error in 18 patients in whom actual K₂ was significantly higher than K₂ET (see CMV results). Figure 6 shows the impact of using actual K₂ (versus K₂ ET) on estimated RI_pulse in the 18 patients in whom actual K₂ was different from K₂ ET. In 15 patients the impact of using an assumed K₂ ET, instead of the actual K₂, was minimal (< 15%). In three patients, however, the use of an assumed K₂ ET resulted in important errors; R amounted to 3.5, 5.0, and 6.1 cm H₂O · L¹ · s, corresponding to 22%, 25%, and 30% error, respectively. These relatively large errors reflect large differences between actual K₂ and K₂ ET and, also, a large difference in the flow at which CMV and pulse measurements were made. From these observations it may be concluded that the use of an assumed K₂ ET results in important errors in only a minority of patients (3 of 29 patients in whom actual K₂ was known with certainty; see CMV results). In the great majority of patients the error is negligible because differences between actual and assumed K₂ are small or the flow at which pulse measurements are made is not sufficiently different from 1.0 L · s¹ to result in an important error.

View larger version (19K): [in this window] [in a new window]   Figure 6.   Scatter plot and line of identity of the relation between resistance calculated using an assumed K2 (K1 + K2ET) and using the actual K2 obtained by regression analysis (K1 + K2reg) in 18 patients in whom the assumed and actual K2 values were significantly different from each other.

The impact of errors related to the use of the wrong K₂ on the relation between RI_pulse and RI_cmv (Figure 5B) was assessed by recomputing both RI_pulse and RI_cmv using actual K₂, as opposed to K₂ ET, in the 29 patients in whom actual K₂ was known. As may be expected from the foregoing analysis, the effect of this correction was small; r increased from 0.90 to 0.92 (r² = 0.85, p = 8E28). The slope (1.01) and intercept (0.77) of the relation were not significantly different from 1.0 and zero, respectively. The average difference was 0.9 ± 2.0 cm H₂O · L¹ · s (5.4 ± 12.6%) with a 95% CI of 2.9 to 4.7 cm H₂O · L¹ · s (19.3 to 30.1%). This plot is not shown because it is not visibly different from that of Figure 5B.

Active versus passive resistance. The calculations used to estimate RI_pulse do not take into consideration the possible effects of changes in flow on the pressure output of the active inspiratory muscles. Thus, it is assumed that Pmus will continue rising during the flow perturbation at the same rate obtaining before the perturbation. As discussed elsewhere (see THEORY), changes in flow may alter Pmus by reflexly changing muscle activity or through the intrinsic properties of respiratory muscles. In the presence of such effects the true change in Pmus during the pulse may differ from the value estimated from extrapolation of an earlier segment, thereby resulting in differences between calculated RI_pulse and passive R. These differences need not be considered as technical errors. To the extent that intrinsic muscle properties and reflex responses operate during natural changes in flow, and not only during artificial flow changes, resistance of the active respiratory system may be different from that of the passive system. We believed, therefore, that some of the remaining differences between RI_pulse and RI_cmv may be due to the fact that RI_pulse reflects active resistance whereas RI_cmv reflects passive resistance. This possibility was addressed by determining whether the difference between RI_cmv and RI_pulse was related to the magnitude of active pressure generation (Pmus) at the onset of the pulse (i.e., T₀). Thus, we reasoned that the more Pmus the patient is generating the more pronounced the effects of changes in flow on Pmus will be, resulting in greater difference between RI_cmv and RI_pulse. This reasoning is valid at least in the case of intrinsic muscle responses (26).

The impact of Pmus at the onset of the pulse was assessed in two ways: (1) Interindividual differences in Pmus: The 67 patients studied were subjected to a wide range of assist and also displayed a wide range of respiratory drive; minute ventilation on PAV ranged from 6.2 to 24.0 L · min¹. As a result, the extent of Pmus generated at the onset of the pulse varied considerably among patients (0.5 to 16.2 cm H₂O, x (± SD) = 5.8 ± 3.2 cm H₂O). We determined the relation between individual Pmus values and the individual differences between RI_pulse and RI_cmv (n = 67). There was no significant relation between the two variables (r = 0.19, p = 0.13). (2) In 17 patients there were two or more steady-state periods at different levels of assist. As a result different levels of Pmus were obtained in the same patients. We determined RI_pulse and Pmus during the highest and lowest assist in each of these patients and the results were compared by the paired t test. RI_pulse was not different (17.4 ± 6.1 versus 16.3 ± 6.3 cm H₂O · L¹ · s in high and low assist, respectively; p < 0.1) even though Pmus was significantly higher during low assist (6.0 ± 3.1 versus 9.2 ± 2.7 cm H₂O, in high and low assist, p < 0.00001).

Collectively, these observations indicate that any effect prevailing Pmus has on RI_pulse estimates is too small to be of concern. This is of relevance in the clinical setting where respiratory drive and level of assist change frequently. This finding should facilitate interpretation of changes in RI that would, otherwise, have to be corrected for possible changes in Pmus. We believe that a number of technical features in the current approach have contributed to this insensitivity to Pmus. First, the pulse was applied early in inspiration where Pmus is relatively low (mean Pmus = 5.8 cm H₂O). Second, the change in flow (approximately 0.3 L · s¹) was small relative to the gain of the force-velocity relation [0.24 ± 63 cm H₂O · L¹ · s (28)]. Third, the pulse was designed so that V in the intervals T₁ and T₁ are small and roughly equal in magnitude. This minimized the impact differences in V between the two intervals may have had on the Pmus trajectory through the force- length relation; the volume effect on Pmus is considerably greater than the effect of flow [11.2 ± 2.5 cm H₂O · L¹ (28)]. Fourth, the short interval between T₀ and T₁ (0.12 ± 0.07 s) ensured that no important, intervention-related changes in muscle activity occurred before the end of data collection. Thus, the minimal latency for behavioral responses is 0.2 s (18, 19) whereas the reflex responses to similar perturbations have a latency of 0.12 s, and the response evolves gradually (28).

Other factors. There are several factors that may account for the differences between RI_pulse and RI_cmv which remain after the previously discussed considerations (i.e., difference from line of identity in Figure 5B). (1) It may be recalled that allowance for true changes in RI between the CMV and pulse measurements, and for differences between actual K₂ and K₂ ET, was not possible in all patients. These two factors may account for some of the residual variance in the remaining patients. (2) The pulse measurements were made near the beginning of inspiration, where volume is low, whereas the CMV measurements were made at end inspiration where volume is considerably higher. In some patients RI is volume-dependent (2). (3) The current approach assumes that Pmus increases linearly in the interval T₁ to T₁. This may not be true in some patients, introducing potentially bidirectional errors related to extrapolation.

Given the good agreement (between RI_pulse and RI_cmv) that exists without consideration of the aforementioned three factors (Figure 5B) and the certainty that some of the residual variance represents true differences in resistance [point (1), above], it can be safely concluded that possible nonlinear behavior of Pmus [point (3)] and volume dependence [point (2)] do not importantly influence the results of this approach, at least when the average of multiple observations is considered.

Effect of Timing of Pulse Delivery

In early experiments pulses were delivered with two different delays to assess the impact of timing of pulse delivery. We com