In both panels of Figure 5, the diagonal line marked zero represents the relationship between peak Pmus and Vt in the absence of assist. With PAV, as percent of assist increased, the slope of the relation increased. In all cases a stable solution was reached with RRpat:RRvent ratio of 1:1 and the ventilator cycle terminating within 0.1 s of the end of inspiratory effort.

Fig. 5. Results of computer simulation of the relationship between peak inspiratory muscle effort (peak Pmus) and tidal volume (Vt) at different levels of support with pressure support (PSV) and proportional assist (PAV). Patient's characteristics (resistance and elastance) were similar to those of the average patient in this study (Table 1). Numbers to the right of different isopleths in the PSV panel denote PSV level, while corresponding numbers in the PAV panel denote the percent of elastance and resistance used for volume assist and flow assist, respectively. Dashed line marked “N” is the relation in a subject with normal mechanics (R = 4.0 cm H₂O/L/s, E =11.0 cm H₂O/L). Solid dots: relation pertaining to results at PSVmin and PAVmin observed in the current study. Open square: expected Vt as assist level is increased to PSVmax or PAVmax without downregulation of patient's effort. Open circle: final steady-state response to the change from minimum to maximum assist. Ratios in the PSV panel denote the relation between patient's rate and ventilator rate at the relevant points. When no ratio is indicated in either panel, it is 1:1. To generate data at peak Pmus = 0 (i.e., Y intercepts), very brief efforts (0.1 s) were simulated.

Download Figure | Download Powerpoint

[More] [Minimize]

In the PSV mode the relation was variable and was affected by PSV level and intensity of effort. At PSV 5, the assist simply caused a parallel upward shift and the RRpat:RRvent ratio remained 1:1. At higher levels the response displayed three phases. At high peak Pmus there was a vertical shift and a continued RRpat:RRvent ratio of 1:1. As peak Pmus decreased to a critical level, which was higher at higher PSV levels, a transitional zone developed where the relation between Pmus and Vt was biphasic; Vt decreased sharply as Pmus decreased, followed by a rapid increase as Pmus decreased further. At the beginning of this transitional zone, ventilator cycle began extending progressively more into neural expiration, reducing the time available for emptying and resulting in progressively more dynamic hyperinflation (DH). Initially, efforts continued to be able to trigger the Paw cycle, and RRpat:RRvent ratio remained 1:1. However, Vt decreased sharply as DH increased. At slightly lower levels of peak Pmus in this transitional phase a steady state was not reached; with successive neural cycles, ventilator cycle extended progressively more into neural Te and end-expiratory volume progressively increased. After a number of cycles, one effort fails to trigger PSV, allowing volume to return to passive FRC. The sequence then repeats (notice 23:22 ratio at the trough in the PSV 20 line). As Pmus decreased further, within this transitional zone, mechanical Ti increased further and a larger fraction of breaths failed to trigger PSV. In this region average Vt increased (as Pmus decreased), in part because more breaths started near passive FRC (less DH) as a result of preceding nontriggered efforts and in part because of the longer mechanical Ti. Beyond this transitional zone, Vt reached a constant level that was independent of Pmus with a stable 2:1 rhythm. Here, every triggered breath begins near passive FRC and mechanical Ti is long enough to permit equilibration between applied pressure and elastic recoil. Vt equals PSV/Ers.

Whereas Figure 5 shows the result for the case where the patient's RR and mechanics are similar to the average patient in this study, other simulations (not shown) using different patient characteristics show that a higher RRpat and a longer time constant (Rrs/Ers) cause the transitional zone to occur at relatively higher Pmus and the RRpat:RRvent ratio at minimal Pmus to be greater (e.g., 3:1, 4:1, etc.), and vice versa.

Use of flow to estimate ineffective efforts. Our use of flow and Paw to identify ineffective efforts requires justification. The criteria used were based on straightforward physical principles and practical consideration of the features of other possible causes of such flow transients.² We therefore feel that with the criteria used here there is little possibility that ineffective effort will be wrongly identified. This was borne out by the results in the three patients in whom efforts were also directly identified from Pes; in no case did the number of ineffective efforts identified from flow exceed the number identified from Pes. Difficulties may arise where the magnitude of the flow transient is very small or its duration is borderline (e.g., transient marked by downward pointing arrow, Figure 1A). These features, however, reflect feeble and/or very brief negative transients in intrapleural pressure (Ppl) (see corresponding Pes tracing, Figure 1A). Such transients would be questionable regardless of whether one uses flow or Ppl.

Whereas overestimation of ineffective efforts using the current criteria would be quite unlikely, underestimation is possible under two circumstances. First, if an inspiratory effort occurs at the very beginning of the ventilator's expiratory phase, before a clear expiratory flow trajectory has been established, it may simply truncate the peak of the expiratory flow and not be clearly recognized as an independent flow transient (e.g. effort marked by open arrow, Figure 1A). Second, where expiratory flow limitation exists and Ppl is in excess of the value required to generate maximum flow, reduction in Ppl may not result in reduction in expiratory flow. The three patients in whom ineffective efforts were identified independently by the two techniques represent extreme examples of dyssynchrony with PSV. The results in these patients (Figure 2B) suggest that underestimation of ineffective efforts by the flow method is acceptably small.

Comparability of level of assist with PSV and PAV. The choice of the lowest level of assist was based on a clinical judgment by the experimenters (E.G. and K.W.) that the patient could not sustain a further reduction in support. This is admittedly subjective. However, since no difference was identified in any variable at minimum assist, the issue of whether the assist level was comparable is moot. The situation is, however, different at the higher levels of assist where substantial differences in breathing pattern and ΔRR were observed. It may be argued that the observed differences in breathing pattern and ΔRR were related to a relatively lower level of assist with PAV. We believe this to be highly unlikely for the following reasons:

1. Arterial Po ₂, Pco ₂, and pH were virtually identical at level 4 (Table 3). To the extent that these are important (if not the most important) determinants of inspiratory effort, there is no reason to believe that patients had a higher respiratory drive with PAV.

2. Although the operational target for PAVmax was 80% of E and R, the actual assist delivered, determined posthoc, ranged from 65 to 95%. There was no evidence that increasing PAV in the range from 80 to 95% resulted in a different response to PAV. In fact, in three patients in whom percent of assist exceeded 90%, the differences in Vt between PAVmax and PAVmin were zero, 0.06, and 0.05 L, not any greater than the average increase in Vt (0.06) (Figure 3) and in each of these patients ΔRR was zero. This suggests that using a higher PAVmax would not have altered the results.

3. Although ΔPaw at PAVmax was only 80% of ΔPaw at PSVmax (16.2 ± 4.8 versus 20.4 ± 3.2 cm H₂O) (Table 2), the pressure assist with PAV was delivered more than twice as often (RRvent = 27.6 ± 10.5 with PAVmax versus 13.2 ± 3.9 with PSVmax) (Figure 3). The product ΔPaw RRvent at PAVmax was 167% of the corresponding value at PSVmax.

4. Breathing pattern at PSV 3 remained different from PAVmax where ΔPaw was similar (16.2 ± 4.8 versus 16.4 ± 2.7 cm H₂O) (Table 2). Thus, at PSV 3, Vt was larger (p < 0.02), RRvent was lower (p < 0.005) and ΔRR was greater (p < 0.005) than at PAVmax.

The results observed during PAV are very similar to those reported previously (15); the changes in Vt and respiratory rate are quite small over a wide range of assist in the tolerable range. The response to different levels of PSV in the current study is also similar to earlier findings (1, 16-19); Vt progressively increases and ventilator rate progressively decreases as PSV level is increased. The advantage of the present study is that both modes were used in the same patients, thereby allowing us to determine whether the different responses are artifactual or represent differences in respiratory control with the two modes.

Differences in respiratory rate. The responses of ventilator rate to different levels of PSV and PAV were substantially different; the decrease in RRvent with PSV was nearly four times greater than with PAV (29.6 to 13.7 with PSV versus 31.9 to 27.6 with PAV) (Figure 3). These differences, however, were largely artifactual; the responses of RRpat were substantially smaller in absolute terms and not significantly different between the two modes (Figure 3). The decrease in RRvent with PSV was predominantly related to a progressively greater number of ineffective efforts (ΔRR) (Figure 3).

That discrepancies between RRvent and RRpat are likely to occur with PSV was theoretically predicted earlier (6, 7) and was well demonstrated in practice in several recent studies (20-22). It was also predicted that PAV would be less likely to result in such discrepancies (6, 7). The present study confirms this prediction. As shown by computer simulation of the two modes (6, 7, current study), the main reason for this to occur with PSV is the tendency for ventilator Ti to outlast neural Ti, particularly when the mechanical time constant (R/E) is long. This results in reduction in the time available for expiration, and hence inadequate emptying prior to the onset of the next effort. Computer simulations have also shown that the extent to which ventilator Ti exceeds neural Ti, and hence the frequency of unsuccessful efforts, should increase as the level of PSV increases (6, 7). This prediction is also borne out by the present results: ΔRR progressively increased with PSV level.

In a previous study using different levels of PAV, we observed that RRpat covered a wide range (18 to 33 min⁻¹) despite high levels of assist (15). The present results confirm these previous findings and extend them in two respects. First, the same conclusion applies during PSV if one considers the patient's rate as opposed to RRvent. Second, the range of RRpat observed in the current study is considerably larger. Thus, at the highest assist level, RRpat ranged from 12 to 46 min⁻¹ with PSV and 8 to 46 min⁻¹ with PAV (open circles, Figure 4). More importantly, in several patients with marked tachypnea at low levels of support, RRpat changed little over the entire range of assist (Figure 4). This suggests that in some patients tachypnea may be largely due to abnormal tachypneic influences unrelated to respiratory load.

Differences in tidal volume response. Although Vt increased with level of assist in both modes, the increase was substantially more in the case of PSV (Figure 3). The large increase in Vt with PSV cannot be attributed to unusually high levels of PSV; even at PSVmax, PSV level was well within the range used in practice both in terms of absolute pressure (20.4 ± 3.6 cm H₂O) (Table 2) and in terms of associated Vt (10.2 ± 2.2 ml/kg). Likewise, the small increase in Vt with PAV was not due to low levels of assist (see Technical Considerations).

We believe that the difference in Vt response with the two modes is largely related to different patient–ventilator interactions. While with both modes inspiratory effort decreases progressively as assist level increases (16-19, 23-25), the response of the ventilator to this downregulation of Pmus is very different in the two modes. The results of computer simulation carried out here (Figure 5), where patient characteristics were matched to those of the average patient in this study and ventilator characteristics were similar, provide a framework for interpreting the experimental results. For simplicity, only the changes between minimum and maximum assist will be discussed. At PAVmin average Vt was 0.45 L at an average assist level of approximately 45%. This is represented by the solid circle on the 45% line (Figure 5, right panel ). Increasing the assist to an average of 75% (actual increase was to 77%) would, in the absence of downregulation of Pmus, immediately result in doubling of Vt with continued 1:1 synchrony (open square, Figure 5, right panel ). Pa_CO2 would rapidly decrease as a result of the increased V˙e and decreased Vd/Vt (due to larger Vt). Peak Pmus decreases as a result, causing Vt to decrease along the 75% line. A new steady state is reached (open circle) at a point where the decrease in Pa_CO2 is commensurate with the decrease in Pmus (as governed by CO₂ responsiveness) and the change in V˙e is commensurate with the change in Pa_CO2 . It can be seen from this analysis that a substantial increase in Vt with PAV is possible only if RR decreases substantially, thereby minimizing the decrease in Pco ₂, or Pmus fails to downregulate as Pco ₂ decreases (e.g., in the presence of important nonchemical sources of drive).

In our patients, Vd/Vt decreased slightly, while metabolic rate did not change as a result of increasing PAV support from minimum to maximum (Table 3). As a result of the small decrease in Vd/Vt, Pa_CO2 decreased by 2.8 mm Hg, despite a small reduction in V˙e (13.4 to 12.6 L/min) (Table 3). Alveolar V˙e actually increased from 5.6 to 5.9 L/min, which accounts for the decrease in Pa_CO2 . RRpat decreased from 30 to 26.4 min⁻¹, likely the result of the decrease in mechanical load and/ or the small decrease in Pa_CO2 . Had RRpat decreased less, Vt would have increased even less than it did. The opposite, of course, would occur if RRpat decreased more. The decrease in estimated Pmus between PAVmin and PAVmax (14.0 cm H₂O to 8.2 cm H₂O, Figure 5, right panel ) is likely related to the decrease in Pa_CO2 , although a reduction in other sources of respiratory drive cannot be ruled out.

The solid circle in the left panel of Figure 5 represents the case at PSVmin (average Vt = 0.44 L at PSV of approximately 10 cm H₂O) (Table 2). Increasing PSV from 10 to 20 cm H₂O would have, in the absence of Pmus downregulation, resulted in an increase in Vt to 0.73 L with maintained 1:1 synchrony (open square, Figure 5, left panel ). Pa_CO2 would decrease sharply and Pmus would decrease accordingly. As a result, Vt would initially decrease sharply. However, if Vt at the trough of the transitional phase is not commensurate with a steady state, Pmus continues to decrease. Vt rises again, along with development of increasing degrees of nonsynchrony and prolongation of mechanical Ti. The final steady state would again be reached when the decrease in Pa_CO2 is commensurate with the decrease in Pmus and the increase in alveolar volume is commensurate with the decrease in Pa_CO2 . In the average patient represented in Figure 5, the trough of the Vt response at PSV 20 would not be consistent with a steady state. Vt at this point was 0.58, 26% greater than Vt at PSVmin. Even ignoring the lower Vd/Vt, this would have resulted in a 10 mm Hg reduction in Pa_CO2 . Pmus could not stabilize at the trough level and must decrease further, forcing the nonsynchrony and the secondary increase in Vt. The equilibrium point thus occurs at a much higher Vt. On average, in our patients at this equilibrium point (PSVmax), approximately one-half of inspiratory efforts failed to trigger (ΔRR = 12.1, RRpat = 25.3 min⁻¹) (Figure 3). It is of interest to note that according to the simulation presented here, mechanical Ti would increase from 0.9 s to 1.9 s between PSVmin and PSVmax, despite a constant neural Ti in the simulation. This is quite similar to what was actually observed (0.9 to 2.1 s) (Table 2).

The simulation data shown in Figure 5, left panel, also provide an explanation for the occasional occurrence of rapid dramatic changes in ventilator output during PSV. Figure 6 provides an example. In this case, breathing pattern, as judged by ventilator output, suddenly and spontaneously changed from slow, deep breathing to rapid, shallow breathing. The increase in ventilator rate was substantially greater than the increase in RRpat (arrows), reflecting better synchrony. The improved synchrony was, in turn, likely related to an increase in patient effort. Note that with improved synchrony, Vt decreased. The transitional zone demonstrated by the computer simulation implies that small changes in Pmus may cause large changes in ventilator Vt and RRvent when they occur in a critical Pmus range.

Fig. 6. An example of a rapid spontaneous change in Vt and ventilator rate. Arrows indicate the patient's efforts and reveal that much of the increase in RRvent was due to improved synchrony (3:1 → 1:1). Vt decreased markedly as synchrony improved. The change in ventilator output from a slow, deep pattern to a rapid, shallow pattern was likely the result of a small increase in the patient's inspiratory effort.

Download Figure | Download Powerpoint

[More] [Minimize]

The above analysis indicates that the differences in the experimental results can be readily explained by the different patient–ventilator interactions in the two modes. It is to be noted that the expected increase in Vt upon transition from PAVmin to PAVmax in the absence of Pmus downregulation (difference between solid circle and open square, Figure 5, right panel ) is substantially larger than the corresponding expected increase upon transition from PSVmin to PSVmax (0.43 L in right panel versus 0.26 L in the left panel, Figure 5). This further supports our contention that the smaller Vt response with PAV was not due to a relatively smaller range of assist.

1. The present study, along with earlier theoretical (6, 7) and experimental (20-22) findings, indicates that RRvent should not be used as a surrogate for RRpat in the PSV mode, whereas RRvent is a reliable surrogate of RRpat during PAV support.

2. A high RRpat, as high as 46/min, on either PAV or PSV need not reflect inadequate support. In some patients respiratory rate is fairly insensitive to level of support. This conclusion is further supported by recent findings that respiratory rate is not a good predictor of work of breathing or pressure–time product during PSV (16, 26).

3. During PSV, particularly in patients with high resistance, breathing pattern, as judged by ventilator output, can undergo dramatic changes in response to minor changes in the patient's respiratory motor output or PSV level. These changes may be artifactual and should not be interpreted as a major change in the patient's inherent pattern of breathing or level of distress without independent confirmation.

4. Although synchrony between patient and ventilator is much better preserved with PAV, the clinical significance of this feature is not clear. There is very little impact on Pa_CO2 and it does not appear that synchrony, or lack thereof, materially affects gas exchange as judged by the Pa_O2 /Fi _O2 ratio.

In summary, the breathing pattern response to changes in assist level with PSV and PAV are substantially different. These differences are, however, largely artifactual and have little impact on gas exchange. The present study also confirms that breathing pattern at high levels of assist is highly variable with some patients displaying respiratory rates in the forties with very small Vt (approximately 0.3 L). Such patterns in the course of weaning need not signify excessive work.