INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Patients with chronic obstructive pulmonary disease (COPD) demonstrate widely variable exercise capacities. Until recently, it was felt that the FEV₁ provides a good expression of the mechanical changes by which COPD affects exercise tolerance. This notion was based on excellent correlations (r up to 0.78) between FEV₁ and peak O₂ (e.g., 1-8). In earlier studies, however, correlations were carried out without normalization of one or both variables. This treatment enhances the correlation because variability in unnormalized data incorporates anthropometric differences between subjects. Earlier correlations, therefore, do not reflect the specific effect of disease on these variables. In a recent study (9) we plotted, for the first time, normalized FEV₁ versus normalized peak O₂ (i.e., % predicted O₂max). The correlation coefficient was quite low (0.49) indicating that disease-related changes in mechanics, as reflected in FEV₁ (% predicted), account for less than 25% of the variability in peak O₂. Evidently, how COPD affects exercise tolerance is not well represented in the FEV₁. The downgrading of the FEV₁ as a good index of the mechanical derangements limiting exercise makes it necessary to pursue other possible reasons for variability in exercise tolerance. In the present study we explore the role of three factors not hitherto examined.

(1) FEV₁: Partial expiratory flow-volume (PEFV) curves are frequently used to study normal persons as well as patients with reversible and/or chronic airflow limitation mainly to evaluate airway hyperresponsiveness (10, 11). In our experience, as with that of others (12), patients with COPD often exceed the limits determined by the maximal expiratory flow-volume (MEFV) loop during resting ventilation. The difference in flow rates at iso-volume between MEFV and PEFV maneuvers varies considerably among patients with COPD. We hypothesized that the ability to exceed the MEFV boundary during partial maneuvers, refered to here as FEV₁, offers an advantage during exercise in that expiratory flow rates in the operating volume range would be higher than expected based on the usual MEFV curves. This could decrease the degree of dynamic hyperinflation, hence allowing these patients to increase minute ventilation and achieve higher levels of exercise.

(2) There is considerable variability in the shape of MEFV curve; for the same FEV₁ a patient may have a high peak flow and marked scooping while another may show a lower peak and less scooping. We hypothesized that a higher ratio of FEF₅₀ to peak expiratory flow would theoretically decrease the degree of dynamic hyperinflation, allowing these patients to achieve higher levels of exercise.

(3) Dynamic hyperinflation: One of the main consequences of expiratory flow limitation, especially at high levels of ventilation, is the generation of dynamic hyperinflation (13, 14). Different studies have demonstrated the presence of dynamic hyperinflation during exercise in patients with obstructive lung disease. The role of FEV₁ and airflow resistance had been evaluated previously (15, 16). However, little is known about other determinants of dynamic hyperinflation (namely shape of MEFV curve, FEV₁, and inspiratory duty cycle (TI/Ttot) during increasing levels of exercise or about the true effect of this abnormality on maximum ventilation, and oxygen consumption at peak exercise in COPD patients.

The results pertaining to these newly examined variables were incorporated with the results of conventional tests and with the ventilatory response to exercise (which was shown to be of importance earlier [9]) in multiple regression analysis.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Patients

Twenty-four patients, 13 men and 11 women, with chronic obstructive pulmonary disease (COPD), were studied prospectively. All patients met the following criteria: (1) physician-diagnosed COPD; (2) pulmonary function tests compatible with COPD (i.e., FEV₁/FVC ratio of less than 70%, and bronchodilator response less than 15%); (3) exercise limitation by shortness of breath or general fatigue. Additionally, all patients entered into this study were stable at the time of admission. Patients with cardiovascular problems or primary pulmonary vascular disease were excluded. All patients had a basal set of lung volumes and single breath diffusing capacity for carbon monoxide performed within the last year at our pulmonary function laboratory. These data as well as anthropometric data are listed in Table 1. New spirometry was performed the same day of the progressive exercise test in all patients. The new spirometric data are also presented in Table 1.

Equipment

Spirometry was performed using a No. 3 Fleisch pneumotachograph coupled with a differential pressure transducer (model MP45-14; Validyne Engineering Corp., Northridge, CA) and connected to a carrier demodulator (Validyne Engineering Corp.). The flow signal was electronically integrated to obtain volume. Lung volume changes and inspiratory and expiratory flow rates were displayed on the axes of an X-Y oscilloscope (Tektronix Inc., Portland, OR). On line data acquisition was performed using Windaq^® (Dataq Instruments Inc., Akron, OH). Both signals were sampled at 100 Hz and stored on a computer disk for later analysis. The ATS standards for spirometry (17) were followed. The pneumotachograph was calibrated on a daily basis. The predicted normal values for spirometric measurements were those of Cherniack and Raber (18). Normal values for maximum inspiratory flow (MIF) were obtained from Bass (19).

Exercise testing was performed on electronically braked cycle ergometer (Lode standard, Groningen, The Netherlands). Electrocardiogram leads were placed on the chest to monitor heart rate. The subjects breathed through a mouthpiece attached to a two-way Hans-Rudolph valve (Hans-Rudolph, Kansas City, MO) with a combined dead space of 115 ml. A nose clip was used for all the tests. Inspiratory and expiratory flows were measured by separate No. 3 Fleisch pneumotachographs attached to the corresponding lines. The flow signal was electronically integrated to provide ventilation. The expiratory line was connected to a 10-liter mixing chamber with baffles. Fractional concentrations of CO₂ and O₂ in the mixing chamber were determined by a mass spectrometer (MGA-1100; Perkin-Elmer Corp., Pomona, CA). Oxygen saturation values were obtained by a pulse-oximeter (N-200; Nellcor, Hayward, CA). The signals representing flow, volume, gas concentrations, and the electrocardiogram were continuously displayed on a Gould 2400S recorder.

Procedure

Post bronchodilator spirometry was performed on the same day as the exercise test. The bronchodilator consisted of two puffs of salbutamol. Maximal flow-volume curves (minimum of three) were obtained for each patient in the sitting position with a nose clip. The best of all efforts was selected to determine the patient's FEV₁ and FVC. At the end of forced expiration the subject inhaled maximally to TLC. MIF was taken as the highest inspiratory flow obtained during any of these maximal inspiratory efforts. The subject was then asked to breathe normally for 1 to 2 min. With the mouthpiece still on, the subject inspired slowly to a lung volume below TLC (between 40 and 75% of VC approx.) under direct vision and then started a forced expiration to RV. At the end of the forced expiration the subject was asked to take a maximum forced inspiration targeting TLC. This last point was used during the analysis as the reference point to locate the MEFV as well as the partial expiratory flow volume curves over the volume axis (Figure 1). The partial expiratory maneuver was repeated at least three times. Finally another set of control MEFV were obtained after the PEFV.

View larger version (21K): [in this window] [in a new window]   Figure 1.   Maximum (MEFV) and partial (PEFV) expiratory flow-volume curves initiated at different volumes (%VC). Diagonal arrow pairs indicate the volumes reached after 1 s of the partial maneuver (right arrow of each pair) and after one second if the forced expiration had started at the same volume as in the PEFV curve but had proceeded along the lower limit described by the full MEFV curve (left arrow of each pair). FEV1 is the difference between these two 1 s values. TLC = total lung capacity; RV = residual volume.

Each subject was then seated on the cycle ergometer and connected to the circuit. After 1 to 2 min of resting measurements, the subject performed an incremental exercise test. The initial work rate was 10 watts. The subjects were asked to maintain a pedalling rate between 50 and 80 revolutions/min using speedometer feedback. The load was increased by 10 to 30 watts at the end of each minute until the subject could no longer sustain exercise. At the end of each minute, before increasing the load, subjects were instructed to inspire deeply to TLC and inspiratory capacity (IC) was noted.

Data Analysis

FEV₁ was calculated from the maximum and the partial expiratory flow-volume loops. Peak expiratory flow (FEFmax) and FEF₅₀ were measured from both maneuvers and were indexed as percent of MEF and FEF₅₀ predicted for a subject of the same age, height, and gender. We determined the volume, below TLC, at which each partial maneuver started. We next determined the volume that would be exhaled in 1 s, beginning from the same absolute volume but where flow is that observed during a maximum maneuver in the relevant volume range. To do that, we displayed the volume versus time data of the maximum maneuver, determined the time at which the starting volume of the partial maneuver was reached and measured the volume exhaled over the next 1-s period. FEV₁, an index of the patient's ability to exceed the maximum curve during partial maneuvers, was calculated as the difference between the FEV₁ determined from the partial maneuver and the FEV₁ computed from the maximum curve, beginning at the same volume, as described above (Figure 1). The FEV₁ from the maneuver nearest end-expiratory level will be reported. It must be pointed out, however, that in a given patient FEV₁ was not substantially affected by the volume from which expiration started (Figure 1). Accordingly, the precise location of the partial loop relative to end-expiratory volume was not critical.

Minute ventilation (E), tidal volume (VT), respiratory frequency (f), heart rate (HR), oxygen uptake (O₂), CO₂ production (CO₂) and respiratory exchange ratio (R) were calculated for each minute of exercise using standard formulas (20, 21). Ventilatory parameters were expressed at BTPS, and O₂ and CO₂ were expressed at STPD.

In order to obtain an index that represented the shape of the expiratory flow-volume loop (shape index), we used the value of FEF₅₀ from the maximum expiratory maneuver and divided it by peak expiratory flow (FEFmax) (shape index = FEF₅₀/FEFmax).

The ventilatory response to exercise was plotted according the method of Riddle and coworkers (22, 23). As shown in Figure 2, O₂ is expressed as % of maximal predicted for the patient taking into account age, sex, and height, and the ventilatory response of the patient is compared to the predicted average response for the normal subject of the same age, height, and sex*. The observed ventilation at the subject's peak O₂ (A, Figure 2) was divided by the E predicted at the same oxygen consumption in a normal subject with the same age, height and sex (B, Figure 2) to arrive at a ratio (E_max/E_pred). Furthermore, because of its possible relevance to DH, we calculated the ratio of inspiratory time to total breath duration (TI/Ttot) at peak exercise.

View larger version (12K): [in this window] [in a new window]   Figure 2.   Graphical analysis of the ventilatory response. The predicted line describes the average normal response for a subject with the same age, height, and gender (see footnote).

To evaluate the magnitude and pattern of dynamic hyperinflation we plotted the relation between E (L/min) and inspiratory capacity (IC, L) during the increasing levels of exercise (Figure 3). This treatment assumes a constant TLC, an assumption that is justified by the data of Stubbing and colleagues (24). Values were obtained from each one minute interval. A dynamic hyperinflation index (DHI) was computed from:

(1)

View larger version (11K): [in this window] [in a new window]   Figure 3.   Graphical analysis of the dynamic hyperinflation index (DHI) in one patient. Minute ventilation (L/min) is plotted against inspiratory capacity at different levels of exercise.

where IC_rest and IC_ex represent inspiratory capacity at rest and peak exercise, respectively, and E_ex is the difference in E between peak exercise and rest. The units are %IC_rest per unit increase in E. We used two tests to evaluate the pattern of change in DH with E. First, we determined the goodness of fit (correlation coefficient) to a linear regression model. Second, we determined the rate of change in IC over the first 10 L/min increase in E (DHI₁₀) and divided it by the average DHI over the entire E range (DHI as above). A value that is less than one denotes an accelerating rate of DH with E, and vice versa.

Statistical analysis was carried out both with absolute values as well as with normalized data where differences in age, sex and height were taken into account. For most variables normalization was carried out by dividing by the relevant predicted normal values (e.g., FEV₁, IC, MIF, O₂max, etc.). Peak exercise E, and FEV₁ were normalized by dividing by predicted FEV₁. DHI was normalized (DHI(N)) by using E/predFEV₁, instead of E, as the denominator in Equation 1. In effect, normalized DHI(N) = DHI × predFEV₁. Some indexes were dimensionless (e.g., FEV₁/FVC, shape index, TI/Ttot, and ventilatory response (E_max/E_pred) and, therefore, did not require normalization.

A correlation matrix was generated for the relation between individual variables and each of the other variables. Stepwise regression was then carried out to determine the main predictors of the three dependent variables of interest, namely normalized DHI, E_max and peak O₂. We used the backward stepwise regression method because of the existence of significant interrelations between the various independent variables. With the backward stepwise regression the first step was a multiple linear regression between the dependent variable and all the independent variables of interest without regard to the initial individual correlation between the dependent and independent variables (which may be importantly affected by correlations with other variables). The result of this first step gives the relative importance of each independent variable after taking these "side" correlations into account. The variable having the least significant correlation with the dependent variable was then deleted and multiple linear regression between the dependent variable and remaining variables were repeated (step 2). The least significant variable was then deleted. The process was repeated until all remaining variables correlated significantly (p < 0.05) with the dependent variable.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Table 1 shows the average (± SD) of anthropometric and resting spirometric data and Table 2 shows relevant results at peak exercise. Table 3 is a matrix of correlations between selected variables expressed in their absolute units. Table 4 is a similar correlation matrix but where the results were normalized. Because the correlations between unnormalized data are enhanced by anthropometric differences between patients and do not accurately reflect the effect of disease (9), only the correlations between normalized data will be discussed. Table 3 is provided only for contrast to emphasize the importance of normalization (see DISCUSSION). Please note that in Tables 3 and 4 the ratio E_max/E_pred was inverted (E_p/E_m). This form of expressing the ventilatory response was preferred because the ratio E_max/E_pred would, theoretically, produce a hyperbolic relation with peak O₂, which would complicate statistical analysis (see Reference 9 for additional details).

Partial Expiratory Maneuvers

All patients developed higher flow rates during the partial maneuvers than at the corresponding volume in the maximum expiratory maneuver. There was considerable variability between patients. Figure 4 illustrates the results in two patients with roughly the same FEV₁ and demonstrates a marked difference in FEV₁. FEV₁ for the partial maneuver nearest resting end-expiratory volume averaged (± SD) 208 ± 87 ml (Table 1) with a range of 74 to 376 ml.

View larger version (13K): [in this window] [in a new window]   Figure 4.   Maximum and partial expiratory flow volume curves (60-70% VC) in two patients with similar FEV1 (A = 1.05 L, B = 0.95 L). Note the considerable difference in FEV1. For definition of FEV1, see Figure 1.

The only spirometric variable that correlated with FEV₁(N) was FEV₁(N). The correlation was, however, weak (r = 0.47, Table 4). With respect to possible impact of FEV₁ on exercise, Table 4 shows that FEV₁(N) significantly correlated with DHI(N) (r = 0.43, p < 0.05), E_max(N) (r = 0.66, p < 0.001) and peak O₂(N) (r = 0.50, p < 0.02). Figure 5 is a scatter plot of FEV₁(N) versus E_max(N) and peak O₂(N).

View larger version (16K): [in this window] [in a new window]   Figure 5.   (A) Relation between maximum exercise ventilation, expressed as multiples of predicted FEV1, and FEV1. (B) Relation between peak exercise O2 and FEV1. For definition of FEV1, see Figure 1.

Shape of MEVF Curve

There was considerable variability in the shape of the curve. Figure 6 illustrates the contrast in shape between curves from two patients with roughly the same FEV₁. On average, the shape index was 0.15 ± 0.05 (Table 1). The shape index correlated significantly with FEV₁(N) (r = 0.52, p < 0.01, Table 4) but not with any other index of mechanics. Shape had no impact on E_max(N) or peak O₂(N) (Table 4).

View larger version (10K): [in this window] [in a new window]   Figure 6.   Maximum expiratory flow volume curve in two patients with similar FEV1 (A = 1.07 L; B = 1.04 L). Note the variability in shape index (FEF50/FEFmax); A = 7%, B = 23%.

Dynamic Hyperinflation

All patients sustained DH during exercise. At peak exercise IC averaged 21.7 ± 9.2 (% IC rest). IC at peak exercise correlated significantly with IC_rest (r = 0.46, p < 0.05) but not with peak O₂(N) (r = 0.12). There was, however, a considerable range in maximum DH (8%  40% IC_rest).

Within each patient there was a highly significant linear relation between IC and E (e.g., Figure 3). The lowest correlation coefficient was 0.85. This emphasizes the paramount dependence of DH on level of ventilation. Although significant linear correlations were observed, the relation between IC and E displayed non-linearities in many patients. In 12 patients, the ratio DHI₁₀/DHI was between 0.75 and 1.25, indicating that DH increased in an essentially linear fashion with E. In two patients, the slope tended to decrease as E increased while in the remaining 10 patients it tended to increase with E. On average, the ratio DHI₁₀/DHI was 0.86 ± 0.4.

The tendency to develop DH, as expressed by the DHI, varied considerably among patients. Thus the range was 0.18 to 3.03% IC/L/min (mean ± SD = 1.11 ± 0.72). With respect to factors that may, theoretically, increase the tendency for DH, stepwise regression produced the following model:

(2)

Table 5 gives details of the stepwise regression and shows the changes in the r value for the relation between different variables and DHI(N) at different steps of the regression. Although FEV₁(N) showed a stronger individual correlation with DHI(N) than did FEV₁(%) (0.59 versus 0.43, Table 4), the situation was reversed during the first step of the stepwise regression with FEV₁(N) emerging as the stronger correlate (Table 5). In fact the correlation between FEV₁(N) and DHI(N) became positive. There was a highly significant negative correlation between TI/Ttot and DHI(N) (r = 0.57, p < 0.01, Table 4).

Ventilatory Response During Exercise

With the exception of three patients, E at maximum exercise exceeded the value predicted at the same O₂ for normal subjects of the same age, sex, and height. The average E_max/ E_pred was 1.27 ± 0.25 (Table 2) with a range of 0.85 to 1.97. The ventilatory response did not correlate significantly with any index of respiratory mechanics (Table 4). It was also not correlated with DHI(N) (r = 0.01, Table 4), O₂ saturation at end-exercise (r = 1.7) or with diffusing capacity (r = 0.39).

The effect of ventilatory response on maximum exercise was examined after inverting the ratio (i.e., E_pred/E_max). There was a significant positive correlation between E_pred/ E_max and peak O₂ (Figure 7).

View larger version (18K): [in this window] [in a new window]   Figure 7.   Relation between peak exercise O2 and ventilatory response to exercise. Note that a lower x value denotes a higher ventilatory response.

Determinants of Maximum Exercise Ventilation

Significant correlations were observed between normalized E_max and FEV₁(N), FEV₁(N) and DHI(N) (Table 4). The correlation with FEV₁(N) was by far the most significant (r = 0.66, p < 0.001). Stepwise regression provided the following equation:

Details of the stepwise regression are shown in Table 6.

Determinants of Peak V·O₂

Peak O₂(N) correlated with FEV₁(N), E_pred/E_max, FEV₁ (N), MIF(N), IC_rest(N) and DHI(N) (Table 4). Stepwise regression (Table 7) provided the following equation:

(3)

Note that the units of the different variables are the same as those used in Tables 1 and 2. Also note that the correlation between peak O₂(N) and TI/Ttot, which was positive during the individual correlations (Table 4), became negative during stepwise regression.

Figure 8 shows the relation between actual peak O₂ and the values predicted according to Equation 3.

View larger version (14K): [in this window] [in a new window]   Figure 8.   Relation between actual peak O2 (% pred) and predictions based on the model of Equation 3: peak O2 (% pred) = 81.9 ± 75.2  Epred/ Emax + 0.55 MIF (% pred)  10.7 DHI(N)  221 TI/Ttot.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

In the present prospective study we have confirmed several findings established in an earlier retrospective study (9), notably the impact of normalization and the importance of the ventilatory response to exercise in determining exercise tolerance. In addition, we have provided a detailed description of the evolution of dynamic hyperinflation during exercise, the intersubject variability in the tendency to develop DH and the impact this has on variability of exercise performance. Finally, we have demonstrated the relevance of the difference between partial and maximum forced expiratory maneuvers on exercise tolerance in COPD.

The present study provides further support for the need to normalize dependent and independent variables if one is to dissect the impact of disease. Thus, virtually all significant correlations observed between unnormalized data (Table 3) were downgraded after normalization (Table 4) and some became insignificant, thereby indicating that the significance in unnormalized data was substantially enhanced by anthropometric differences between patients. Noteworthy are the changes in correlations between peak O₂ and IC (0.68 to 0.46), E_max and FEV₁ (0.73 to 0.45), E_max and MIF (0.64 to 0.34), E_max and IC (0.71 to 0.30) and, of particular relevance to this study, the correlation between peak O₂ and FEV₁ (0.75 to 0.43). For the above reasons all subsequent discussion will focus on relationships observed between normalized data.

Partial versus Maximal Expiratory Flow Rates

Unlike normal subjects, in patients with COPD flow rates during forced expirations beginning below TLC (PEFV) often exceed the rates developed at the same volume when forced expiration begins at TLC (MEFV). The mechanisms of this phenomenon have been subject to much study. In essence, four factors may contribute. (1) Compression of airways contributes to expiratory flow during forced expiratory efforts (25). This excess flow would occur early in the maneuver regardless of the volume from which it is initiated (11, 25). (2) Differences in intrathoracic gas volume due to different degrees of gas compression in the two maneuvers (particularly during the early part of the partial maneuvers). (3) Effect of volume history prior to forced expiration on airway resistance and lung compliance (12). (4) The presence of mechanical nonhomogeneities with the faster units emptying first, resulting in high initial flow, regardless of volume from which the maneuver is started (11).

There has been little discussion of the functional significance of the difference between partial and maximal flow rates. In theory, factors 1-3 can be viewed as technical artifacts and may not be expected to offer functional advantages. A contribution from "fast" compartments to expiratory flow, which is independent of volume at which expiration begins, may however be expected to offer advantages in that higher expiratory flow rates (than expected from the MEFV maneuver) are made available at any operating volume. Such flow rates would otherwise be unavailable unless the patient inspired to TLC each breath. The energetic advantage is obvious. Furthermore, there should be a lesser likelihood for dynamic hyperinflation since, for a given end-inspiratory volume and a given expiratory time, the volume reached at end-expiration should be lower. Tidal volume and ventilation at a given inspiratory effort would, thus, be higher (13, 14).

The present study is the first to assess the importance of this phenomenon to exercise performance. We found significant correlations between FEV₁(N) and dynamic hyperinflation index (DHI(N)) (p < 0.05, Table 4), E_max(N) (p < 0.0001, Table 4) and peak O₂(N) (p < 0.02, Table 4). That these significant correlations were not fortuitously related to correlations between FEV₁ and other more conventional variables (e.g., FEV₁) was demonstrated by the results of the stepwise regression. This analysis confirmed that FEV₁(N) exerts a significant influence on DHI(N) and E_max(N) that is independent of other indices of respiratory mechanics. FEV₁(N) did not survive the stepwise regression for peak O₂(N) and does not appear in the final model. This, however, is related to the inclusion of DHI in this analysis. To the extent that FEV₁(N) correlates strongly with DHI(N) (Table 5), and likely exerts its influence via this variable (i.e., DHI(N)), only one of these two variables can survive. In fact, if DHI(N) is not entered in the stepwise regression for peak O₂(N). FEV₁ emerges as the variable with the most significant correlation (with peak O₂) among other mechanical indices.

By demonstrating a functional utility to FEV₁, the present findings indirectly indicate that the mechanism of nonhomogeneous emptying of lung units (mechanism 4, above) contributes importantly to the differences observed between PEFV and MEFV in this patient population. Had the difference between the two maneuvers been related to the other factors (mechanisms 1-3), a functional advantage would not have been expected.

There was a significant positive correlation between FEV₁ (N) and FEV₁(N) (r = 0.47, Table 4). Thus, patients with more severe disease tended to show a smaller difference between PEFV and MEFV, as expressed by FEV₁. This positive correlation may be explained as follows (we are grateful to Dr. S. Permutt, Baltimore, for this explanation): In the normal lung, all lung units are fairly fast. There is little nonhomogeneity and little reason for a positive FEV₁ to exist. In patients with far advanced disease, there cannot be an important fast compartment. Other than for the airway compression mechanisms (Equation 1 above), there is again little reason for a positive FEV₁ to exist. Patients with moderate disease have, therefore, the greatest likelihood of having important fast compartments amid slow regions. Nonhomogeneity of mechanical properties may, therefore, be expected to be highest with moderate disease and to decrease with greater and lesser severity. Since our patients encompassed a range from moderate to very severe COPD, the correlation may be expected to be positive.

Dynamic Hyperinflation (DH)

The development of DH during exercise in COPD patients has been well documented (24, 26, 27). The present study extends these observations in several respects:

Factors contributing to DH. As may be expected, the magnitude of DH was very significantly correlated with level of E in each subject. This emphasizes the paramount dependence of DH on E. The tendency to develop DH, expressed as the average slope of the relation between DH and E(DHI), varied considerably among patients. On theoretical grounds it would be expected that at a given ventilation the amount of DH would depend on TI/Ttot and the relation between maximum possible flow and lung volume. For a given E, TI/Ttot determines the average expiratory flow while the maximum expiratory F-V relation determines where, on the volume axis, this expiratory flow can be achieved. We found no correlation between DHI(N) and the shape index (r = 0.09). In retrospect, this is not surprising since the shape index (FEF₅₀/FEFmax) is derived from efforts beginning at TLC. The relation between FEF₅₀, so determined and the maximal flow that can actually be generated in the middle of vital capacity during spontaneous breathing is highly variable as evident from the results of partial maneuvers. Interestingly, FEV₁/FVC ratio showed a significant correlation with DHI(N) (Table 4) and its influence was clearly apparent at the end of the stepwise regression (r = 0.60, p = 0.002, Table 5). By contrast, FEV₁(N), which showed a strong individual correlation with DHI(N) (Table 4), did not emerge as an important determinant of DH during the stepwise regression (Table 5). This may be related to the fact that the volume range over which FEV₁ occurs (near TLC in these patients) is usually above the resting VT volume range. FEV₁ may thus not provide a good reflection of maximum flow in the operating volume range.

The stepwise regression for DHI(N) produced two unexpected results: (1) Whereas a straight correlation between MIF(N) and DHI(N) produced a negative relation (r = 0.33, Table 4), the independent relation between MIF(N) and DHI(N), as unearthed by stepwise regression (Table 5), was strongly positive (r = 0.54, p = 0.007). This positive relation is counter intuitive. Thus, if the patient's capacity to generate inspiratory flow is high, he should be able to reduce TI/Ttot, decreasing the expiratory flow requirement at a given E and permitting him to operate at a lower lung volume.

(2) There was a strong negative correlation between TI/ Ttot and DHI(N) (r = 0.57, p < 0.01, Table 4). This relation became even stronger after the stepwise regression (r = 0.73), p = 0.00005). One might have expected that, all else being equal, a smaller TI/Ttot should provide relatively more time for expiration and hence a lower expiratory flow requirement at the same ventilation. The negative correlation, therefore, indicates that, rather than being an independent variable in this relation, TI/Ttot is the dependent variable; the more dynamic hyperinflation exists, the shorter TI/Ttot becomes. Two explanations may be entertained. First, in response to greater DH, neural TI/Ttot is reduced as a compensatory strategy to mitigate the DH. Second, with DH, mechanical TI/Ttot is artifactually reduced (i.e., relative to neural TI/Ttot) since part of neural TI becomes incorporated into mechanical TE due to inspiratory time taken to reverse flow (14). Regardless of the reason, the current findings indicate that differences in neural TI/Ttot do not actively contribute to the development of DH but, if anything, rather act to moderate it.

The relation between MIF(N) and TI/Ttot was itself curious and unexpected. There was a positive correlation between these two variables. This correlation was so high that it is quite unlikely to be fortuitous (r = 0.8, p < 0.00001, Table 4). Again, and as discussed under (1) immediately above, one might have expected a negative relation.

We offer a hypothesis that may explain the counterintuitive relations described above, between MIF(N), TI/Ttot and DHI(N). Expiratory flow limitation produces an unpleasant sensation (28). In an effort to minimize this sensation the patient has two options. First, he can actively increase lung volume into a region where maximum available flow is higher without necessarily reducing TI/Ttot. Second, he can actively reduce TI/Ttot in order to reduce the required expiratory flow at a given ventilation. The first option requires tonic activation of inspiratory muscles and is energetically costly. The second option requires deliberate reduction of TI/Ttot from its naturally selected value. Although energetically superior, this approach may have its own sensory consequences. It is possible that patients with strong inspiratory muscles, as reflected in a high MIF(N), choose the first option whereas those with weaker muscles are obligated to follow the second course. According to this scenario MIF(N) is simply a marker of inspiratory muscle strength. Patients with high MIF(N) would thus develop greater DH per unit ventilation (DHI(N)), which is in part active, and would not need to reduce TI/Ttot. The opposite would occur in patients with weak muscles (as reflected in a low MIF(N)). A positive correlation would thus develop between MIF(N) and TI/Ttot. Although development of tonic inspiratory activity has been demonstrated under other circumstances associated with DH (29, 30), it has not been documented in COPD patients during exercise. It would also be of interest to see whether tonic inspiratory activity, if any, occurs primarily in patients with strong muscles.

Evolution of DH during progressive exercise. Our findings indicate that, by and large, DH increases in a nearly linear fashion with exercise ventilation. Where a non-linearity occurs, it is in the direction of accelerating rate of DH as E increases. There was no significant correlation between severity of obstruction, as expressed by FEV₁(N), and the curvature index (DHI₁₀/DHI). The factors that lead to an accelerating rate of DH are likely complex.

The maximum amount of DH at end-exercise varied considerably and ranged between 8 and 40% of IC. That there was no correlation between maximum DH and peak O₂(N) indicates that patients do not achieve greater peak O₂ by tolerating more DH. That the range of maximum DH was so wide indicates either that tolerance for DH varies widely or that the absolute value of DH is only one of many factors that contribute to exercise limitation.

Impact of dynamic hyperinflation index on peak exercise ventilation and V·o₂. This aspect will be discussed below (see DETERMINANTS OF MAXIMUM EXERCISE VENTILATION and DETERMINANTS OF PEAK O₂).

Ventilatory Response to Exercise

The present study confirms our earlier finding that variability in ventilatory response is an important determinant of exercise performance in COPD (9). This study further confirms that ventilatory response is not correlated with any of the variables traditionally correlated with exercise performance such as respiratory mechanics, O₂ saturation or DCO. It is, therefore, a fairly independent variable. The correlation between E_pred/E_max and peak O₂ was not as impressive in the current study as in the previous study (0.45 versus 0.62). This is likely related to a narrower range of ventilatory response in the present study (87% to 197% versus 88% to 290% of predicted). Nonetheless, the ventilatory response survived the stepwise regression (Table 7) and remains an important determinant of exercise performance. The possible reasons for variability in ventilatory response to exercise have been discussed previously (9) and will not be repeated.

Determinants of Maximum Exercise Ventilation

FEV₁(N) emerged as the most significant correlate with normalized E_max, and the only variable that was ultimately included in the stepwise regression. After FEV₁(N) was entered, FEV₁(N) and DHI(N) became insignificant (Table 6). This is not surprising since there were significant correlations between FEV₁(N) on one hand, and FEV₁(N) and DHI(N) on the other (Table 4). The advantages offered by the difference between MEFV and PEFV (as expressed by FEV₁) have been discussed earlier. DHI(N) did not emerge as an independent determinant of E_max(N). We do not believe that this negates an important role for DH since one of the mechanisms by which FEV₁ affects E_max is through producing lesser tendency for DH.

The statistical analysis for correlation with E_max(N) included multiple inspiratory and expiratory mechanical indices. These are most frequently affected by COPD. The fact that r² of the final model was so low (r ² = 0.44) suggests that such mechanical abnormalities do not limit exercise primarily by imposing a ventilatory "ceiling." It is more likely that exercise is terminated prematurely by sensations related to the abnormal mechanics, to excessive work of breathing or to other respiratory and non-respiratory factors. The reduction in E observed as maximum exercise may accordingly be more a reflection of, rather than a cause of, reduced peak O₂. In fact, if the stepwise regression is repeated with peak O₂ being considered as an independent variable (for E_max), the final model becomes:

As can ben seen FEV₁(N) continues to be a significant variable, but the overall correlation improves substantially.

Determinants of Peak V·O₂

In the present study we assessed the potential impact of several variables not hitherto examined on peak O₂(N). These included FEV₁(N), DHI(N), shape and TI/Ttot. In addition, we included in the analysis the ventilatory response to exercise (E_pred/E_max) the importance of which has only very recently been described (9). The stepwise regression also included most of the conventional variables that were examined in the past (e.g., FEV₁, lung volumes, IC, MIF). It was of interest that, with the exception of MIF, when these "new" variables were introduced, all other indices of mechanics became insignificant (Table 7). Perhaps this is because the abnormal mechanics in COPD operate via the DHI(N) which may, thus, provide a more direct link between the pathology and exercise performance.

The inclusion of these "new" variables (including E_pred/ E_max) also resulted in a much more robust model (Equation 3) with a correlation coefficient of 0.82 (accounting for 67% of the variability in peak O₂) (Table 7). Without these "new" variables, the model that emerges includes only MIF(N) and has a much lower predictive value (r = 0.49, r² = 0.25).

Finally, although correlations do not necessarily reflect mechanisms, the model that emerged (Equation 3, Table 7) is particularly appealing since all the variables included in it impact directly on inspiratory muscle function, thereby providing a plausible unifying mechanisms for exercise limitation in COPD: excessive inspiratory muscle effort relative to these muscles' capacity. Thus, for a given increase in E, a greater tendency for dynamic hyperinflation, as reflected in DHI(N), imposes a greater inspiratory threshold load while decreasing the pressure generating ability of inspiratory muscles. A higher ventilatory demand at a given level of exercise (as reflected in a lower E_pred/E_max) imposes a relatively greater stress on the inspiratory muscles and, for the same DHI(N), should result in more DH. A low MIF(N) reflects weak inspiratory muscles and/or high inspiratory resistance. In either case, at a given E, inspiratory muscles are placed under greater stress than if MIF(N) were higher. The impact of TI/Ttot on inspiratory muscle function is not so clear cut. On one hand, at a given E, a shorter TI/Ttot translates into a greater inspiratory flow. The inspiratory muscles need to generate more pressure during inspiration. On the other hand, the inspiratory muscles contract for a shorter fraction of cycle time which tends to reduce the tension-time index (31). In a recent computer simulation, using realistic values of the abnormal mechanics in COPD, we found that the net effect of reducing TI/Ttot is a lower mean inspiratory pressure at the same E (13, 14).

In summary, our results indicate that much of the variability in maximum exercise performance in COPD is related to differences in ventilatory response to exercise, in maximum inspiratory flow and in susceptibility to dynamic hyperinflation. The latter is importantly influenced by the patient's ability to generate higher expiratory flow rates during expiration beginning below TLC than, at the same volume, during forced expirations beginning from TLC. Measurement of FEV₁ and of the ventilatory response to exercise may be helpful in the assessment of mechanism of disability in this disease and it may be suspected that correction of reasons for excessive ventilation, when present, may improve exercise tolerance.