INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

MONITORING OF RESPIRATORY mechanics is important in mechanically ventilated patients to diagnose the disease underlying acute respiratory failure (ARF), to assess the status and progress of the disease, and to measure the effects of treatment such as drugs and application of positive end-expiratory pressure (PEEP) (27, 32). Abnormal respiratory mechanics is due not only to increased flow resistance and elastance but also to intrinsic PEEP (PEEP_i) (26). The latter reflects the end-expiratory elastic recoil of the total respiratory system due to incomplete expiration and dynamic pulmonary hyperinflation. PEEP_i is determined predominantly by the patient's alterations in respiratory mechanics (increased compliance and resistance), but it is also influenced by ventilator settings, for example, a short expiratory time or a large tidal volume (26). An unrecognized PEEP_i may cause 1) a misinterpretation of hemodynamic data, leading to erroneous evaluation of the patient's volemic status; 2) an erroneous evaluation of respiratory mechanics (26); and 3) a severe patient-ventilatory asynchrony (27). Detection and measurement of PEEP_i is paramount in ventilator-dependent patients, not only to prevent its adverse effects, but also to implement therapeutic strategies such as changes in the ventilator settings, aggressive administration of bronchodilatators, and application of PEEP.

Whereas the assessment of PEEP_i may present some problems in patients actively triggering the ventilator, this is relatively simple during controlled mechanical ventilation, when the respiratory muscles are relaxed, for example, by means of the end-expiratory occlusion technique (27). However, this technique requires either ventilators equipped with an end-expiratory occlusion button (26) or some additional equipment and skill that may not be routinely available in clinical settings (14). In addition, measurements obtained by the ventilator's facilities require adequate correction (30), whereas the end-expiratory occlusion maneuver interferes with the ventilator settings and may not be suitable for continuous monitoring of PEEP_i. In fact, continuous monitoring enables the early detection of changes in patient status, thus allowing a rapid therapeutic response as well as the evaluation of its effectiveness.

Mathematical models for on-line monitoring of respiratory mechanics during mechanical ventilation (3, 18) offer an attractive tool to assess PEEP_i continuously in the intensive care unit without interfering with the ventilator settings and also overcome some technical issues associated with the end-expiratory occlusion technique. However, only one group of investigators (12) has specifically addressed monitoring of PEEP_i in ventilated acute respiratory distress syndrome patients in whom the levels of PEEP_i are, in general, relatively low.

The hypothesis of this study was that on-line monitoring of PEEP_i in ventilator-dependent patients is possible when using a recursive least square approach. The purpose here is to present this new approach and to try to interpret our measurements by comparing them with the values of PEEP_i obtained by means of two more commonly used techniques, i.e., 1) end-expiratory airway occlusion and 2) the change in airway pressure preceding the onset of inspiratory airflow (26, 27).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

This study was approved by the ethics committee of the institution, and informed consent was obtained from each patient or the next of kin. Eight patients admitted into the intensive care unit of the Ospedale Maggiore of Parma (Italy) who required mechanical ventilation because of respiratory failure of various etiologies were enrolled into this study. All patients were stable hemodynamically with a mean arterial pressure of >70 mmHg without inotropic drugs. None had chest wall abnormalities, hemothorax, pneumothorax, or high intracranial pressure. The patients' characteristics and ventilator patterns at the time of the inclusion into the study are shown in Tables 1 and 2, respectively. All patients were mechanically ventilated with Dräger Evita II (Drägerwerk, Lubeck, Germany).

Measurements. Flow () was measured with a heated pneumotachograph (Fleisch no. 2, Lausanne, Switzerland) connected to a differential pressure transducer (Validyne MP 45, ±2 cmH₂O, Validyne, Northridge, CA) that was inserted through cones between the Y piece of the circuit and the proximal tip of the endotracheal or tracheostomy tube. The pneumotachograph was calibrated with a supersyringe using the gas mixture in use and was linear over the experimental range of flow. The instrumental dead space was 120 ml. Volume (V) was obtained by numerical integration of the flow signal. Because of the inspiration-by-inspiration analysis performed (see Estimation algorithm), we reset the integration of airflow at the beginning of each breath. Pressure was measured at the airway opening between the pneumotachograph and the artificial airway. The side ports were connected through air-filled noncompliant catheters 50 cm long and 1.5 mm of internal diameter to two differential pressure transducers (Validyne MP 45 ± 80 cmH₂O). Calibration of the transducers was done with a water manometer. To reduce the effects of the compliance and resistance of the circuit on the mechanics measurements, a single length of standard noncompliant tubing (2 cm ID and 60 cm long) was used, and the humidifier was omitted from the inspiratory line. Special care was taken to avoid air leaks within the equipment and around the cuff of the tube. All signals were recorded on a personal computer (Macintosh II CI; Apple Computer, Cupertino, CA) via an analog-to-digital converter (MacLab Analog Digital Instrument, Castle Hill, Australia) at a sample rate of 100 Hz and were stored in diskettes for subsequent computer analysis. During the measurements, a physician not involved in the study was always present for patient care.

Procedure. The investigation was performed with the patient in the supine position, sedated with a continuous infusion of phentanyl (0.02 to 0.03 µg · kg¹ · min¹) and diazepam (0.6 µg · kg¹ · min¹) and paralyzed with vecuronium (0.1 mg/kg, followed by 0.05 mg/kg if necessary). Measurements were taken during four different ventilatory modalities applied in random order, each characterized by a distinct ventilatory pattern: pattern 1, volume-controlled ventilatory mode with an inspiratory flow of ~1 l/s and without PEEP (zero end-expiratory pressure, ZEEP); pattern 2, identical to pattern 1 except for the addition of an external PEEP equal to PEEP_i on ZEEP increased by 5 cmH₂O; pattern 3, volume-controlled ventilatory mode on ZEEP, with an inspiratory flow of ~0.5 l/s; and pattern 4, pressure-controlled ventilatory mode on ZEEP.

Data analysis. For each ventilatory pattern the following parameters were determined: 1) static PEEP_i (PEEP_i,st), i.e., the value of airway pressure recorded at the end of the expiratory occlusion maneuver as the mean value of airway pressure during the last 0.2 s of the occlusion period (Fig. 1); reported values are means of three measurements; 2) dynamic PEEP_i (PEEP_i,dyn), i.e., the increase in airway pressure preceding the onset of inspiratory flow (Fig. 2); and 3) model-based, on-line estimates of PEEP_i (PEEP_i,mod) that were obtained as described below.

View larger version (19K): [in this window] [in a new window]   Fig. 1.   Measurement of static intrinsic positive end-expiratory pressure (PEEPi,st) by the end-expiratory occlusion maneuver in patient 7 with ventilatory pattern 1. The arrow in B indicates the value of PEEPi,st.

View larger version (19K): [in this window] [in a new window]   Fig. 2.   A: airway pressure (Paw) vs. flow diagram in patient 3 with ventilatory pattern 1. Twenty-two consecutive ventilatory cycles are superimposed. Inspiration is rightward. Note that the start of inspiration precedes the zero flow (dotted line). The target portion (dashed box) is magnified in B. Dynamic PEEPi (PEEPi,dyn) is measured as the difference between the value of Paw at zero flow and end-expiratory pressure. The small end-expiratory pressure reflects the resistive pressure due to the end-expiratory flow and is included in the value of PEEPi,dyn.

Model of lung mechanics. The model used in literature (3, 18) for on-line monitoring of respiratory mechanics is the first-order lumped viscoelastic model. It is described by

(1)

where Paw is the airway pressure, is airflow, V is lung volume, R accounts for total inspiratory resistance, E accounts for respiratory system elastance, P₀ is the value of the total positive pressure at the end of expiration, and t is continuous time.

Estimation algorithm. The RLS algorithm with an exponential weighting factor, , has been used to track changes in the mechanical properties of the respiratory system in real time (3, 18). To do so, one can express the measured samples of the Paw in discrete time form

(2)

where T is the sampling time, (kT) = [R(kT),E(kT),P₀(kT)]' is the parameter vector, (kT) = [(kT),V(kT),1]' is the data vector, and (kT) is the error term representing both noise measurements and model prediction errors.

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

Statistical analysis. Statistical analysis was performed on the measured and estimated PEEP_i by use of Student's t-test for paired data. Linear regression analysis was made by using the least square method. Bland-Altman plots were constructed after regression analysis (10).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

For each ventilatory modality, 2 min of respiratory signal measurements were analyzed by the proposed algorithm. Figure 3A shows a representative (t), Paw(t), and V(t) trace for one of the subjects in our database in the ZEEP ventilatory pattern 1. The recursive procedure identified, breath by breath, the portion of inspiratory data on which to perform real-time PEEP_i estimation (Fig. 3B). This allowed us to obtain the time course of PEEP_i estimates. The tracking algorithm was tuned according to the criteria previously described, leading to a choice of  = 0.95 that is a weighted data window of ~0.2 s. Figure 3B also depicts the on-line estimates during the inspiratory cycle provided by the method (±SD). The estimated PEEP_i exhibits a modest increase during inspiration. Parameter SDs are relatively low except for the end-inspiratory data, in which the effect of transient phenomena and/or higher order behavior of the respiratory system are more marked and lead to a spike in PEEP_i estimates. The initial decrease of PEEP_i during the initial portion of inspiration is probably due to this spike and to the memory effect of the algorithm (see Estimation algorithm).

View larger version (35K): [in this window] [in a new window]   Fig. 3.   A: representative record of ventilatory flow, Paw, and volume (by numerical integration of flow) in patient 3 with ventilatory pattern 1. B: inspiratory portion of data as identified by the estimation algorithm in a breathing cycle for patient 3 with ventilatory pattern 1 (first 3 panels). The last panel illustrates the time course of estimated PEEPi (thick line) ± SD (thin line) obtained from the records shown.

Figure 4 shows the time course of the estimated PEEP_i [µ_PEEPi(t) ± _PEEPi(t)] for the same subject during 2 min of ventilation with ventilatory pattern 1. The interval ±_PEEPi(t) is the real-time measure of parameter variation during inspiration. The breath-by-breath estimates, as well as parameter variability, are very stable, and this enables us to evaluate the algorithm performance and to test the reliability of the estimates using the mean of the on-line estimates during all the measurement intervals. In this way, we were able to compare the individual and mean values of PEEP_i derived from PEEP_i,mod, PEEP_i,st, and PEEP_i,dyn for the three ZEEP patterns (patterns 1, 3, and 4). Data are shown in Table 4. At any given ventilatory modality, the mean values of PEEP_i obtained with the same technique were not significantly different (P > 0.21). By contrast, the values of PEEP_i obtained with the three techniques were very different, regardless of the given ventilator setting (Table 4). It can be noted that PEEP_i,dyn (2.7 ± 1.9 cmH₂O) and PEEP_i,st (8.1 ± 5.5 cmH₂O) provided the lowest and highest values, respectively. The pooled mean PEEP_i,mod (5.6 ± 3.4 cmH₂O) was significantly lower than PEEP_i,st and greater than PEEP_i,dyn. However, Fig. 5 shows that the values of PEEP_i obtained from the three techniques were significantly correlated. PEEP_i,dyn measurements were significantly correlated with PEEP_i,st (r = 0.65, P < 0.001), although linear regression analysis gave a slope m = 0.23 (significantly different from 1, P < 0.001) and an intercept q = 0.86 cmH₂O (not significantly different from 0). PEEP_i,mod exhibited a higher correlation with PEEP_i,st (r = 0.90, P < 0.001); the slope of the regression equation was m = 0.560 (significantly different from 1, P < 0.001), whereas q = 1.05 cmH₂O (not significantly different from 0). Comparison between PEEP_i,dyn and model estimates also gave a significant correlation (r = 0.768, P < 0.001), and linear extrapolation yielded m = 1.359 (not significantly different from 1) and q = 1.908 cmH₂O (significantly different from 0, P < 0.05). The findings of the correlation analysis are better evidenced in the Bland-Altman diagrams (Fig. 5), in which it is clear that PEEP_i,mod is systematically lower than PEEP_i,st and systematically higher than PEEP_i,dyn. Consequently, the model estimates are in better agreement with PEEP_i,st than with PEEP_i,dyn (having notably lower 95% confidence limits), particularly in patients with PEEP_i >5 cmH₂O.

View larger version (17K): [in this window] [in a new window]   Fig. 4.   The thick line shows the time course of model PEEPi (PEEPi,mod) estimated on 22 consecutive ventilatory cycles on an inspiration-by-inspiration basis. The thin lines illustrate ±1 parameter variability within each inspiratory cycle. The model estimates are quite stable, and dispersion within the inspiratory cycle is limited. Data are from patient 3 ventilated with pattern 1.

View larger version (20K): [in this window] [in a new window]   Fig. 5.   Identity plots (A) and Bland-Altman plots (B) comparing values of PEEPi obtained at zero end-expiratory pressure (ZEEP) with the 3 methods. In A, dashed lines are the identity lines, and solid lines are computed according to regression analysis. In B, the dashed lines represent the ±2-SD interval.

Table 5 shows that, when PEEP was set by the ventilator (ventilator setting 2) at a value 5 cmH₂O greater than PEEP_i,st on ZEEP, the values of PEEP_t were essentially identical, without any significant difference. This is confirmed by Fig. 6, in which it can be seen that data points lie virtually on the identity lines with slopes not significantly different from 1 and intercepts not significantly different from 0, the correlation coefficients ranging from 0.97 to 0.99 (P < 0.001). The Bland-Altman analysis does not show any appreciable discrepancy; the data were well under the 95% limit of agreement, and the small underestimation (~0.5 cmH₂O) is not statistically significant (Fig. 6).

View larger version (17K): [in this window] [in a new window]   Fig. 6.   Identity plots (A) and Bland-Altman plots (B) comparing values of PEEPi obtained at the high PEEP ventilatory pattern with the 3 methods. In A, dashed lines are the identity lines, and solid lines are computed according to regression analysis. In B, the dashed lines represent the ±2-SD interval.

Finally, we compared the performance of the two methods suitable for a breath-by-breath PEEP_i estimation in chronic obstructive pulmonary disease (COPD) patients ventilated at ZEEP, i.e., the condition with the highest PEEP_i, by taking the values of PEEP_i,st as a reference. As illustrated in Fig. 7, cycle-by-cycle measurements of both PEEP_i,dyn and PEEP_i,mod were always lower than occlusion PEEP_i. However, whereas values of PEEP_i,dyn were not significantly correlated to PEEP_i,st (r = 0.225), PEEP_i,mod exhibited a strong correlation with PEEP_i,st (r = 0.707, P < 0.01), linear regression giving a slope (m = 0.707) not significantly different from 1 and an intercept (1.571 cmH₂O) not significantly different from 0. 

View larger version (20K): [in this window] [in a new window]   Fig. 7.   Identity plots (A) and Bland-Altman plots (B) comparing PEEPi obtained in the COPD patients ventilated with the ZEEP patterns. In A, dashed lines are the identity lines, and solid lines are computed according to regression analysis. In B, the dashed lines represent the ±2-SD interval.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Since the recognition of Auto-PEEP, almost 20 years ago (23), clinicians in critical care settings have become increasingly conscious that PEEP_i is a fundamental physiological parameter to be measured in ventilator-dependent patients and that it is important to have a simple technique that is reliable for measurement and even monitoring in the clinical environment (17). In this study, we have presented a new approach for on-line monitoring of PEEP_i based on the simple first-order model of respiratory mechanics and an RLS estimation approach. We then critically evaluated our measurements by comparing them with two well-established experimental techniques, i.e., end-expiratory occlusion and change in airway pressure preceding the onset of inspiratory airflow (26). As mentioned previously, we are well aware that our model estimates, as well as the other two techniques for measuring PEEP_i, i.e., the zero-flow technique for PEEP_i,dyn and the end-expiratory occlusion technique for PEEP_i,st, provide values of total PEEP (PEEP_t). The latter is determined by PEEP_i, reflecting the end-expiratory elastic recoil pressure of the respiratory system and any positive pressure either set intentionally by the ventilator or due to the resistance of the ventilator expiratory circuits, tubing, and valves. However, it is a reasonable assumption that without PEEP intentionally set, PEEP_t essentially reflects PEEP_i. In fact, in patients with expiratory flow limitation, such as COPD patients (5 out of 8 in this study), external positive pressure lower than PEEP_i does not alter the value of PEEP_i and does not add to it (27). Under these circumstances, the values of PEEP_i and PEEP_t are virtually identical. By contrast, in patients without expiratory flow limitation, external positive pressure may affect the value of PEEP_i, and hence the term PEEP_t may also be more correct in the ZEEP condition. In our three patients with ARF without COPD, the values of PEEP_t (or PEEP_i on ZEEP) were small and <3 cmH₂O in all cases. On one hand, it cannot be excluded that a slight PEEP_i was detectable in these patients, because it has been shown that PEEP_i can also be present in patients without COPD (11). On the other hand, part of PEEP_t could be due to the low resistive pressure at end-expiration as a result of the expiratory ventilator circuits, as shown by comparing Tables 3 and 4. However, it could be argued that values of PEEP_i or PEEP_t < 3 cmH₂O might not be of significant clinical relevance in ventilator-dependent patients. By contrast, with PEEP set by the ventilator, all three techniques provide values of PEEP_t that, as shown in Table 5, were very similar. Under these circumstances, PEEP_i may be computed according to the simple equation PEEP_i = PEEP_t  PEEP (27).

A number of studies (3-7, 9, 16, 18, 19) involving the RLS technique have recently been designed to track changes in respiratory system resistance and elastance in real time. However, no study addressed the issue of monitoring PEEP_i using a mathematical model of lung mechanics together with recursive parameter estimation techniques. The only previous publication that presented a method suitable to estimate PEEP_i (12) on a breath-by-breath basis and without the need of a flow interruption maneuver was based on the off-line least square fit of the first-order model (Eq. 1). In addition, that method (12) was tested only in mechanically ventilated patients with acute respiratory distress syndrome and with high levels of PEEP (11 ± 2 cmH₂O). In this paper, we present a new algorithm for continuous, on-line monitoring of PEEP_i in ventilator-dependent patients, but we also include patients with COPD in our analysis, namely, the patients with the highest levels of PEEP_i and in whom PEEP_i may have the worst adverse effects (26).

Our results indicate that the algorithm provides stable and repeatable PEEP_i measurements during the 2 min of controlled mechanical ventilation (see Fig. 4) in all 32 traces analyzed. The adequacy of our model estimates is also supported by two other considerations. First, PEEP_i,mod did not exhibit any different behavior from PEEP_i,dyn and PEEP_i,st when the inspiratory flow rate or profile was modified: there was no significant difference in the values of PEEP_i provided by the same method at any given ventilatory pattern. Second, values of PEEP_i,mod were always correlated with PEEP_i obtained by means of other techniques, in particular PEEP_i,st, which represents a more reliable value than PEEP_i,dyn. However, the three methods to measure PEEP_i yielded different results, although correlated.

The simplest way to validate our measurements would have been the comparison with a gold standard. This has been done, for example, when different techniques to measure PEEP_i,st have been proposed (33). However, this kind of comparison was not possible to validate our model because the estimated value of PEEP_i was obtained under dynamic conditions, i.e., during the ventilatory cycle, and from the analysis of almost the whole mechanical inflation. By contrast, PEEP_i,st and PEEP_i,dyn are obtained respectively during end-expiratory airway occlusion, i.e., a static maneuver, and from a single point (zero flow) at the beginning of the inspiration. Literature consistently shows that PEEP_i,dyn is systematically and significantly lower than PEEP_i,st during spontaneous breathing (24) as well as during assisted (1, 33) and controlled (22) mechanical ventilation. Whereas the physiological meaning of PEEP_i,st is acceptably clear, PEEP_i,dyn is a matter of more debate (2). PEEP_i,st represents the average end-expiratory elastic recoil of the total respiratory system at the lung volume at which the airway occlusion occurs. In fact, during the 3-5 s of airway occlusion, both stress adaptation phenomena occur, and lung units with different regional time constants, and hence different PEEP_i, can equalize (pendelluft). The achieved equilibration between regional alveolar and airway pressure is indicated by the plateau in airway pressure (26). By contrast, PEEP_i,dyn represents the lowest regional PEEP_i that has to be counterbalanced by the positive pressure of the ventilator to start inspiratory flow (26). The difference between PEEP_i,dyn and PEEP_i,st during controlled mechanical ventilation can be very great, PEEP_i,dyn amounting only to 25-30% of PEEP_i,st (22). In our patients, the PEEP_i,dyn-PEEP_i,st ratio averaged 30%. This result concurs with the fact that 63% of our patients (5 out of 8) were COPD patients, who are well known to have a large degree of lung inhomogeneity (11) that substantially affects the PEEP_i,dyn-PEEP_i,st ratio by decreasing it (22). It should be said that a recent experimental study, in mechanically ventilated anesthetized rabbits, found that PEEP_i,dyn could be greater than PEEP_i,st (13). However, some technical issues suggest that this finding should be treated with caution and requires further confirmation (2).

Our PEEP_i,mod estimates were smaller than PEEP_i,st, although the values from the two methods were significantly correlated in all instances (Figs. 5-7). This might not be surprising in line with the results obtained by the other authors who showed that PEEP_i measured under dynamic conditions, i.e., before a sufficient equilibration time between regional alveolar and airway opening pressure had elapsed, is consistently smaller than PEEP_i,st. Our PEEP_i,mod is constructed from time-variant estimation of P₀ in the first-order model during inspiration, i.e., under dynamic conditions. However, our PEEP_i,mod was systematically greater than PEEP_i,dyn measured from the positive value of airway pressure at the point of zero flow, i.e., when inspiration starts, although the two values were correlated. Our interpretation of the difference between PEEP_i,dyn and PEEP_i,mod is as follows. The positive airway pressure at the point of zero flow, i.e., PEEP_i,dyn according to common terminology, reflects the minimum pressure needed to start inspiration, which is the pressure that counterbalances the lowest PEEP_i. In fact, as soon as the lowest regional value of PEEP_i is counterbalanced, mechanical lung inflation begins. From that point on, the inspiratory flow continues, and lung regions with higher values of PEEP_i can be recruited to inspiration by the increasing positive pressure. This progressive recruitment of alveolar regions with higher PEEP_i cannot be appreciated because inspiration had already started. However, if inspiratory flow is delivered by a preset constant profile, the endowment of a constant flow indicates that all lung units are filling at the same rate even in the presence of lung heterogeneity and hence that most of the regional PEEP_i has been counterbalanced (8). It has been shown that inspiratory flow becomes constant after a variable time from the beginning of inspiration (25). With constant pressure ventilation (pattern 4), the flow profile exhibits an initial rapid rise and a subsequent slow decay (25). The lung units with a fast time constant receive the initial rapid flow, whereas the units with a longer time constant fill later with the slower flow. Clearly, the fast time constant units are those with the lowest PEEP_i, and airway pressure at zero flow reflects the lowest regional PEEP_i. Therefore, with both constant and decelerating flow, PEEP_i,dyn reflects the minimum PEEP_i. By contrast, the PEEP_i,mod cycle-by-cycle estimates are obtained as the weighted mean of the time course of intrinsic PEEP during inspiration. Hence, the value of PEEP_i,mod is affected not only by the initial lowest PEEP_i but also by the subsequent higher values of PEEP_i in the lung units with a longer time constant, which are recruited either by the increasing airway pressure or by the decreasing flow rate with the progress of inspiration.

Our interpretation of this physiological meaning of PEEP_i,mod is supported by three findings in this work. First, PEEP_i,mod is always significantly correlated with PEEP_i,st, the latter being the best gold standard available because of its safe physiological interpretation, even in conditions in which PEEP_i,dyn loses the significant correlation (Fig. 7). Second, the coefficients of correlation between PEEP_i,mod and PEEP_i,st are always greater than between PEEP_i,dyn and PEEP_i,st. This is in line with the fact that PEEP_i,dyn comes from a single point, whereas both PEEP_i,mod and PEEP_i,st are influenced by values of PEEP_i from different lung units. The algorithm is likely to estimate the PEEP_i,st lower bound. Finally, when PEEP_i is abolished by PEEP set 5 cmH₂O above PEEP_i, all the techniques provide virtually identical results (Table 5 and Fig. 6) for PEEP_t. A reasonable explanation of the excellent agreement among all the techniques to measure PEEP_i can be found in the more homogeneous lungs when PEEP_i is replaced by PEEP. In fact, whereas in the nonhomogeneous lungs the fast time constant units can start filling while the long time constant units are still emptying, in the more homogeneous lungs, most if not all the lung units can start filling almost simultaneously. In the most homogeneous condition, all the lung units can start filling from their elastic equilibrium point. This can be the case when external PEEP has abolished PEEP_i. Under these circumstances, the differences between any PEEP_i,dyn and PEEP_i,st disappears. This interpretation is supported by some results obtained by Rossi and colleagues (28), who showed that ventilation-perfusion mismatching diminishes and Pa_O2 and Pa_CO2 improve when PEEP replaces PEEP_i, suggesting more homogeneous lungs. As shown in Fig. 6, in our patients, when PEEP was set by the ventilator at a value greater than PEEP_i, no difference was detectable among PEEP_t measured by means of different techniques. The agreement shown in Fig. 6 further supports the validity of our on-line method to estimate PEEP_i and PEEP_t.

The estimate obtained with our model gave us a better approximation of PEEP_i,st than that given by the other methods not requiring any maneuver or additional equipment and explained the differences found between our method and the standard measurement of PEEP_i,dyn, which does not adequately describe the heterogeneity of alveolar pressure distributions.

To be thorough, we have also applied to our ZEEP-COPD data the other possible methods proposed in Refs. 5 and 12 for performing continuous monitoring of PEEP_i. However, as specified by the authors in Ref. 5, the term P₀ of Eq. 1 represents essentially the value of Paw when both V and are 0 (i.e., PEEP). This is confirmed by comparing the real-time estimates obtained by this method with the end-expiratory occlusion measurements: we found a poor correlation with PEEP_i,st [r = 0.12, P = nonsignificant (NS)]. Repeating the same procedure, we have compared the results of Eberhard and coworkers' (12) method with PEEP_i,st. Here, the mathematical method fails if the database is restricted to the case of COPD patients ventilated at ZEEP. In fact, in this case, the correlation with PEEP_i,st is lost (r = 0.44, P = NS).

Finally, we would like to point out that our on-line method to estimate PEEP_i provides a value obtained from the whole mechanical breath analysis whereas the so-called PEEP_i,dyn comes from only one point of airway pressure at zero flow. Because, at the beginning of the breath, the rise in airway pressure is very fast and steep, the single-point PEEP_i,dyn is likely to be less accurate than our weighted value from the whole breath. Furthermore, the end-expiratory airway occlusion technique not only requires a slight intervention in the ventilator setting from the caregiver but also needs either additional equipment if the patient is ventilated with ventilators without the end-expiratory occlusion facility (14) or adequate correction if the facilities available in the ventilator are used (15, 30). Therefore, our on-line estimation of PEEP_i provides some technical advantages over the other available techniques for monitoring PEEP_i in ventilator-dependent patients. Moreover, although PEEP_i,st remains a reliable parameter, there is an important reason why PEEP_i,mod should be obtained: PEEP_i,st is a static parameter that provides the lung recoil pressure at a given volume when all the stress adaptation transients have dissipated. This does not represent the situation regarding the pressures inside the lung at the end of a normal dynamic breath.

In summary, in this study we have adopted a recursive least square algorithm combined with the classical first-order model of respiratory mechanics and continuous measurement of airflow and airway pressure to quantify PEEP_i in real time. The method constructs, from the recursive estimation of P₀ during inspiration (see Eq. 1), a weighted mean and standard deviation of dynamic intrinsic PEEP. This value is updated on a cycle-by-cycle basis to give real-time monitoring of this important clinical index in ventilator-dependent patients with ARF of different etiologies, including COPD. The latter is the condition in which the highest values of PEEP_i are commonly found. Our method is limited to patients without any respiratory activity during controlled mechanical ventilation. Furthermore, it must be stressed that our on-line estimate of PEEP_i is really an estimate of PEEP_t, i.e., the total positive pressure at the end of the expiration. Without PEEP set by the ventilator, PEEP_t essentially reflects PEEP_i, i.e., the end-expiratory elastic recoil. When PEEP was set by the ventilator, PEEP_i could be computed by subtracting the set value of PEEP from the estimated PEEP_t, and this can also be done automatically in the algorithm. The value of PEEP_i estimated by our on-line model more closely reflects a true PEEP_i,dyn than the conventional one-point measurement of PEEP_i,dyn at zero flow, because it is influenced by almost all the values of PEEP_i occurring during mechanical lung inflation.

Although limited to controlled mechanical ventilation, our method for monitoring PEEP_i (PEEP_t) on-line may provide the possibility for useful clinical applications. This can be determined by additional studies.