All test lung studies are limited by the difficulties in modeling human breathing, particularly in disease. The potential effects on the results of this study of a number of important limitations need to be considered.

Resistance. Airways resistance (Raw) in COPD has a complex relationship with airflow and lung volume, which makes modeling difficult. Selecting appropriate resistances for test lung studies is also complicated by the wide range of total Raw encountered in patients with COPD. This test lung study used narrow tubes to create resistance to airflow, an approach that has been widely used. The resistances chosen were 3.6 cm H₂O/L/s and 20 cm H₂O/L/s (measured at 60 L/min flow), as the low and high resistances, respectively. Studies of ventilated patients with COPD have often recorded inspiratory resistance (at a flow rate of 60 L/min, when stated) in the range of 10 to 25 cm H₂O/L/s (1, 4, 5). Typical figures for patients with stable COPD have been measured in the range of 5 to 8 cm H₂O/L/s (7), and approximately 3 cm H₂O/L/s for asymptomatic smokers (8). Therefore, the high resistance reflected the fairly severe end of the range, whereas the low resistance was at the mild extreme for COPD.

Although there was a large difference between these two resistances, this had only a relatively small effect on triggering time. This can be inferred by comparing the results for conditions A and B, as a major part of the difference between these two conditions was the difference in the resistances. This suggests that even with all the limitations of resistance modeling, a more precise simulation of Raw would probably not have led to substantial differences in the results.

Flow and Vt . The flow profiles for conditions A, B, and C were intended to be broadly comparable with those observed in patients with COPD, in terms of the rates of rise and fall of inspiratory flow and the maximal inspiratory and expiratory flows. The Vt values were intended to cover a broad range from a large Vt in condition A, to a small Vt in condition C, comparable to published figures for patients with severe exacerbations of COPD. Vt in condition C (444 ml) is comparable to the mean Vt of 476 ml measured in 12 patients with hypercapnic COPD (9), and of 389 ml in 14 patients with COPD at the time of successful weaning from mechanical ventilation (10). Conversely, the Vt in condition A is at the extreme upper limit of the Vt range for COPD (approximately 2 SD higher than the mean for 28 patients with COPD) (9).

The expiratory time for our test lung was shorter than would usually be experienced in clinical practice, which could raise concerns about gas trapping and intrinsic positive end-expiratory pressure (PEEPi). However, we ensured that gas trapping did not affect our results, as explained in the next section. Apart from the concern about gas trapping, the duration of expiration would not be expected to affect the performance of the ventilators studied, as they respond to events during the inspiratory rather than the expiratory phase.

Leaks. Few data are available on leak flow rates or volumes from clinical use of NIV on which to base a model. Large leaks through the mouth of approximately 1 L per breath during sleep have been recorded from patients using nasal masks (11), and at the opposite extreme, leaks of less than 100 ml per breath from around carefully adjusted, well-fitting nasal masks (12). The volume of leakage in this model averaged approximately 200 ml per breath for the 2-mm orifice and 400 ml per breath for the larger orifice, although the total volume depended on the mask pressure profile. Further details are given in the online data supplement.

The limited effect of these different leak sizes on triggering delays is perhaps surprising at first glance. The explanation is likely to be that the ventilators' cycling criteria at the beginning and end of inspiration were still met rapidly despite these added leaks. The 2-mm leak would have increased the total flow generated by the ventilators at end-inspiration, assuming a mask pressure of 20 cm H₂O, by approximately 12 L/min, and the larger leak by approximately 25 L/min. If we note the rapid decrease in inspiratory flow at end-inspiration shown in Figure 6, and consider that the flow would generally have to fall to below 25% of the peak to initiate cycling to expiration in most ventilators, we can see that the change in flow at the end of modeled inspiration is so abrupt that the addition of the leak flows to this curve would have had relatively little effect on the time taken to reach the 25% of peak flow threshold, except perhaps under condition C. At the beginning of inspiration, the smaller leak would have added approximately 5 L/min to the flow with a positive expiratory pressure of 4 cm H₂O, and the larger leak approximately 8 L/min at this mask pressure. However, these flows would not be expected to affect the trigger delays of the flow-triggered ventilators, which should respond to a change (increase) in inspiratory flow, rather than the absolute flow level. However, for the two pressure-cycled ventilators these leaks might cause a minor increase in the inspiratory trigger delays by slowing the decline in mask pressure. The leak flows did not induce auto-cycling in this model.

Fig. 6. Examples of the relationship between test lung airway flow, mask pressure, and respiratory cycle under each of conditions A, B, and C. (A) Condition A with Vision ventilator. (B) Condition B with Nippy 1 ventilator. (C) Condition C with Achieva ventilator. Ascent of the piston corresponds with neural inspiratory time. These illustrate the rate of change of flow at the beginning and end of neural inspiratory time, the determinant of cycling for most of the ventilators.

Download Figure (Part 1) | Download Figure (Part 2) | Download Figure (Part 3) | Download Powerpoint

[More] [Minimize]

An important limitation to the validity of this model of leakage is the fixed sizes of the leaks. In patients, the dimensions of leaks between the mask and the face would increase with increasing mask pressure because of the flexibility of masks and of facial soft tissues. Therefore, this model may have overestimated the leak at the beginning of inspiration when mask pressure is low, but almost certainly underestimated the leak that can occur at the end of inspiration when mask pressure is high. The results for expiratory triggering are therefore likely to underestimate the trigger delays that can occur with large mask leaks. In particular, very large leaks may prevent the ventilator flow at end-inspiration from falling below the percentage of peak flow that is necessary to allow cycling into expiration, potentially causing major problems with coordination (13).

PEEPi. PEEPi occurs in many patients with COPD receiving ventilatory support and has an important impact on ventilator triggering into inspiration. However, in this study we aimed to compare the ventilators in a simulation in which PEEPi did not occur, as this would have complicated the interpretation of the results. We designed the test lung to avoid the occurrence of gas trapping at end-expiration. To achieve this, the expiratory phase was not passive, but involved active movement of the piston; this, and the absence of flow limitation, helped ensure that the test lung cylinder emptied well during expiration. We assessed whether PEEPi was present by recording the pressure within the test lung cylinder (with the pressure transducers proximal to the resistance), to ensure that it had returned to below the desired minimum, that is, the applied PEEP (PEEPe) level by the end of expiration. The model avoided PEEPi entirely in conditions A and B; however, in condition C a limited amount of gas trapping did occur. Therefore, to ensure that this did not affect the validity of the results of the study, inspiratory triggering delays were calculated from the onset of inspiratory flow. The onset of inspiratory flow was delayed by mean periods of between 4 and 76 ms, depending on the ventilator, from the time of reversal of the piston direction, in condition C. Further details are given in the online data supplement. Where gas trapping was observed in condition C, the level of PEEPi remained constant from breath to breath, so breath stacking did not occur.

Although this model aimed to compare the ventilators in the absence of PEEPi, the potential additional effect of PEEPi on triggering delays needs to be considered. When present, PEEPi delays the onset of inspiratory flow from the onset of inspiratory muscular activity, which further prolongs inspiratory trigger delay times for some patients. The magnitude of PEEPi is partly dependent on the extent of any expiratory trigger delay, as any shortening of the effective expiratory phase increases gas trapping (14). Therefore, those ventilators with longer expiratory triggering delays would tend to cause an increase in PEEPi, further adding to the inspiratory trigger delays.

Target inspiratory pressure. The target inspiratory pressure of 20 cm H₂O was chosen as this has frequently been used in published studies of NIV in patients with acute exacerbations of COPD, particularly since the publication of two important early studies (15, 16). A range of target pressure levels has been included in other test lung studies (17– 19). In these studies, lower target pressures have tended to lead to either longer inspiratory trigger delays or little change. However, the relative performance of different ventilators has been very similar in the range of target pressures from 15 to 20 cm H₂O, which are widely used in studies of NIV in COPD.

Several comparisons of NIV ventilators using test lungs have been published previously (17-20). However, several important features relevant to the acute use of NIV in COPD have not previously been modeled to the extent as in this simulation. These features include high Raw, small Vt, high FRC, mask leaks, and a realistic interface between the test lung and ventilator. In the study by Bunburaphong and colleagues (17), the modeled neural inspiratory time was directly determined, so that inspiratory and expiratory delay times were measurable. For several of the ventilators studied, the trigger delay times quoted have large variances, particularly for expiratory trigger delay, suggesting that the triggering was inconsistent, and making comparisons with our work difficult. However, where reproducible triggering times were obtained, they were broadly comparable with those in this study. The five ventilators included in each of the studies (BP 20, BP 30, VPAP2, Onyx, PB335) performed similarly with respect to each other with both test lungs.

Previous test lung studies have explored a number of other features of NIV ventilators in addition to inspiratory and expiratory trigger delays. Some of these features, such as the “patient” work of triggering to initiate inspiration, and the imposed work during expiration are heavily dependent on the triggering times. Features that have been studied that are independent of triggering times are the rate at which the inspiratory pressure rises, a function of the power of the ventilator, and the extent of carbon dioxide rebreathing, a function of the system employed to clear exhaled gases. Although there is some evidence that rapid pressure rise times reduce patient work (21), too rapid rates may be uncomfortable. Furthermore, fast pressure increases may lead to particularly high peak inspiratory flows, which may lead to premature termination of inspiration when the fixed percentage criterion for expiratory cycling is reached (22). For those ventilators where the rise time was adjustable, the fastest rates were used in this study on the assumption that these were likely to be associated with the shortest trigger delays.

The extent of carbon dioxide rebreathing with different ventilators has been extensively studied (23, 24), and separate expiratory lines have been shown to improve carbon dioxide clearance.

The different trigger delays identified in this study could arise either from differences in the algorithms that control cycling, or from differences in the electrical and mechanical processes in the ventilators. Some information is available on the cycling criteria for NIV ventilators. In general, cycling to inspiration is triggered by increases in flow of the range 1 to 5 L/min, or decreases in mask pressure of 0.5 to 1 cm H₂O. Cycling to expiration is generally triggered by a fall in inspiratory flow to below a given percentage of maximum, often 25%, or to a percentage set by the operator (or after a given time interval for two of the ventilators, Nippy 1 and Nippy 2). If the ventilators' trigger delays arose solely from the time taken to meet these cycling criteria, then all the ventilators would have been expected to have quite similar trigger delays, all less than approximately 50 ms under the conditions in this study. As the trigger delays for any given ventilator remained broadly similar under the different conditions modeled, it is likely that most of this time is determined by the ventilator's physical ability to respond to changing conditions, rather than the stated algorithms per se.

The objective of NIV in acute COPD is to normalize the load on the respiratory muscles, as far as is possible (22), by sharing the work of the breath between the patient and the ventilator. To achieve this, the ventilator needs to contribute to the work of the breath throughout inspiration. A long inspiratory trigger delay will leave the inspiratory muscles unassisted at the beginning of inspiration, and may even impose significant isometric-like pressure loads during this period (22). An important implication of prolonged expiratory trigger times is that as the time available for expiration is shortened, lung emptying may be incomplete, leading to increased dynamic hyperinflation, PEEPi, and adding further to the work required to trigger the next ventilator breath (4, 25). Expiratory muscle recruitment may also occur if ventilator inspiration continues into expiration. The use of timed inspiration has been suggested as a way of avoiding these problems (13). In addition to the physiologic problems that result from poor synchronization, it is likely that the sensations that result from dys-synchrony are uncomfortable (4). This may be an important factor in NIV treatment failure in anxious, ill patients.

Future studies need to assess the extent to which better synchronization improves the effectiveness and tolerability of NIV for COPD in a clinical setting. Future ventilators should aim to have better electrical and mechanical response characteristics, and cycling algorithms need to be more complex with multiple cycling criteria to ensure close synchronization. Including inspiratory duration criteria in addition to flow criteria would be a relatively easy way of ensuring effective triggering to expiration when large mask leaks are present. In addition to improving synchronization, the design of future ventilators also needs to address the issue of matching the support throughout each inspiration more closely to what the patient requires (22).

Many of the ventilators available for NIV in acute exacerbations of COPD have long trigger delay times. The extent of these delays is to some extent related to the physiology of the modeled patient. However, most of the delay time appears to be due to the physical properties of the ventilators.