Exercise Ventilatory Inefficiency Adds to Lung Function in Predicting Mortality in COPD

ABSTRACT

Severity of resting functional impairment only partially predicts the increased risk of death in chronic obstructive pulmonary disease (COPD). Increased ventilation during exercise is associated with markers of disease progression and poor prognosis, including emphysema extension and pulmonary vascular impairment. Whether excess exercise ventilation would add to resting lung function in predicting mortality in COPD, however, is currently unknown. After an incremental cardiopulmonary exercise test, 288 patients (forced expiratory volume in one second ranging from 18% to 148% predicted) were followed for a median (interquartile range) of 57 (47) months. Increases in the lowest (nadir) ventilation to CO₂ output (VCO₂) ratio determined excess exercise ventilation. Seventy-seven patients (26.7%) died during follow-up: 30/77 (38.9%) deaths were due to respiratory causes. Deceased patients were older, leaner, had a greater co-morbidity burden (Charlson Index) and reported more daily life dyspnea. Moreover, they had poorer lung function and exercise tolerance (p < 0.05). A logistic regression analysis revealed that ventilation/VCO₂ nadir was the only exercise variable that added to age, body mass index, Charlson Index and resting inspiratory capacity (IC)/total lung capacity (TLC) ratio to predict all-cause and respiratory mortality (p < 0.001). Kaplan–Meier analyses showed that survival time was particularly reduced when ventilation/VCO₂ nadir > 34 was associated with IC/TLC ≤ 0.34 or IC/TLC ≤ 0.31 for all-cause and respiratory mortality, respectively (p < 0.001). Excess exercise ventilation is an independent prognostic marker across the spectrum of COPD severity. Physiological abnormalities beyond traditional airway dysfunction and lung mechanics are relevant in determining the course of the disease.

KEYWORDS:

• Survival; COPD; ventilation; cardiopulmonary exercise test; physiology

Introduction

Chronic obstructive pulmonary disease (COPD) is a leading cause of mortality worldwide (1). The severity of airflow limitation [as estimated by forced expiratory volume in one second (FEV₁)] only partially explains increased risk of death in these patients (2, 3). More recently, lung hyperinflation (4) and transfer factor (2) have been found to be useful prognostic indicators. Unfortunately, a large fraction of deaths remains unpredicted by resting functional evaluation (5) and a recent document co-sponsored by the American Thoracic Society and European Respiratory Society called for additional research on predictors of mortality in COPD (3).

Exercise tolerance decreases in tandem with progressing COPD severity (6). In fact, even simple estimates of functional capacity, such as the 6-minute walking distance, provide useful prognostic information, particularly when considered in association with severity of airway obstruction and exertional dyspnea on daily life [as reviewed in ref. (5)]. Measurement of cardiocirculatory and ventilatory variables by means of cardiopulmonary exercise testing (CPET) might provide important complementary information to “field” tests as those measurements are more directly linked to the pathophysiology of COPD and may elucidate the mechanisms underlying exercise intolerance in individual patients (7). Determination of CPET variables, which are more closely related to mortality are therefore key to expand our knowledge on the mechanistic bases of the reported inter-relationships between poor functional capacity, greater activity-related dyspnea and poor prognosis in COPD.(5, 8)

In this context, there is a growing recognition that CPET-based markers of ventilatory inefficiency [increased ventilation (VE) relative to CO₂ production (VCO₂)] are related to clinically relevant outcomes in COPD (9–11). These abnormalities are thought to reflect an enlarged dead space (VD) as a fraction of a lower tidal volume (VT) and/or increased neural ventilatory drive due to a low PaCO₂ “set-point” and heightened afferent ventilatory stimulation (12, 13). Of note, several pathophysiological derangements underlying increased VD/VT, including emphysema progression (14), pulmonary microvascular abnormalities (15) and mechanical VT constraints (16) are ultimately relevant to prognosis in COPD (5, 17). High neural ventilatory drive may be a consequence of increased pulmonary arterial pressures (18), sympathetic over-excitation and increased stimulation of peripheral ergoreceptors (19), which are all well-known negative prognostic markers in chronic cardiopulmonary diseases (20, 21). Thus, similar to heart failure (22) and pulmonary hypertension (21), exercise ventilatory inefficiency may constitute a relevant physiological biomarker with negative prognostic implications in COPD.

This study aimed to determine whether incremental CPET would add to standard pulmonary function tests in estimating the risk of death in patients with mild to very severe COPD. We specifically hypothesized that ventilatory inefficiency [i.e., increased VE/VCO₂ ratio at its lowest point (“nadir”)] (23) would compound with established resting functional indices in predicting all-cause and respiratory mortality in these patients.

Methods

Subjects

The study population comprised 288 outpatients with established COPD presenting with a wide range of airflow obstruction who underwent resting functional and CPET evaluations from October 1999 to June 2012 at the Respiratory Investigation Unit, Kingston General Hospital. Those evaluations were performed as screening tests for potential participation in subsequent physiologic studies. Thus, patients’ diagnosis was carefully confirmed and treatment (inhaled β₂-agonists, anticholinergics, and inhaled steroids) optimized according to the guidelines at the time of testing. Data from this cohort have been used in a communication dealing with the effects of COPD progression on exercise ventilation (11). Causes of death were obtained from the patients’ centralized electronic medical records from Kingston General Hospital and coded as “respiratory” or “non-respiratory” by two independent reviewers (AA and JAN). Discrepancies in coding (N = 4) were resolved by consensus. Patients had no evidence of asthma or another lung disease or a known contraindication for a CPET (7). The institutional review board approved the use of these retrospective, anonymous data sets and waived the need for patient informed consent (DMED-1659-13).

Measurements

Clinical variables

Age, gender, and body mass index (BMI, kg/m²) were recorded. Dyspnea was evaluated by the Baseline Dyspnea Index (BDI) scale (24). Co-morbidity burden was determined by the combined 19-disease Charlson Index (25).

Lung function tests

Spirometry [including inspiratory capacity (IC) measurements], body plethysmography [residual volume (RV) and total lung capacity (TLC)], and measurements of transfer factor for carbon monoxide (TL_CO) were performed using automated testing equipment (2130 spirometer with 6200 Autobox DL or V6200 Autobox; SensorMedics; Yorba Linda, CA, USA).

Cardiopulmonary exercise test

Symptom-limited incremental CPET was performed on an electronically braked cycle ergometer using the Vmax229d System (SensorMedics). Tests were performed without supplemental oxygen. The rate of work rate increment was individually selected according to reported exercise tolerance and resting functional impairment (typically 5–10 W/min in the Global Initiative for Chronic Obstructive Lung Disease (GOLD) stages 3 and 4 and 10–15 W/min in stages 1 and 2). VE (L/min), VCO₂ (L/min), oxygen uptake (VO₂, L/min), end-tidal partial pressure for carbon dioxide (PETCO₂, kPa), and VT (L) were averaged at 30-second intervals. Peak VE was also expressed relative to the estimated maximal voluntary ventilation [MVV (L/min) = FEV₁ × 35]. VE/VCO₂ (ventilatory equivalent for CO₂) was the lowest test data point (23). IC maneuvers tracked operating lung volumes and, assuming a constant TLC, were used to calculate inspiratory reserve volume (IRV, L) (16). Arterial oxygen saturation was measured noninvasively by pulse oximetry (SpO₂, %). Breathlessness was rated according to the modified 10-point Borg category-ratio scale.

Statistical analysis

Values are reported as means ± standard deviation (SD) unless otherwise specified (IBM^ۚ SPPS^ۚ Statistics version 22.0.0.0). The time frame from CPET to the last attendance or death was recorded. Survivors and nonsurvivors were contrasted by nonpaired t or Mann–Whitney's test or a χ² test for differences in proportions. The survival time was calculated with the life table method. Receiver operating characteristics (ROC) curve analyses selected the optimal threshold values (highest Youden's J index = Sensitivity + Specificity − 1) for the lung function and CPET variables. Areas under the curves were compared according to DeLong et al. (26) (MedCalc 14.12.0, Ostend, Belgium). Univariate and multivariate binary logistic regression analyses evaluated the predictive value of the different variables, using all-cause and respiratory mortality as the outcome. Clinical variables were used as continuous variables, except the categorical variables of gender and smoking status. Results were expressed in terms of the estimated relative risks with corresponding 95% confidence intervals. Multicollinearity was checked by calculating variable's level of tolerance and the respective variance inflation factor. Kaplan–Meier analysis assessed the survival characteristics of lung function and CPET variables retained in the final logistic regression models with statistical significance being determined by the log-rank test.

Results

General characteristics

Main clinical, functional, and CPET variables describing the study population are shown in Table 1. Patients presented with a large range of resting functional impairment (e.g., FEV₁ ranged from 18% to 148% pred); however, 178/288 (61.8%) patients had mild-to-moderate airflow obstruction. Exercise impairment, as indicated by peak VO₂ below the lower limit of normality (83% predicted) (7), was found in 209/288 (72.6%) patients. As expected, patients showed increased peak VE/MVV ratios, lower peak HR, poor ventilatory efficiency, and varying reduction in IC and SpO₂. Moreover, dyspnea was the main limiting symptom in the majority of patients (Table 1).

Table 1. Clinical, resting functional, and CPET responses for the whole sample and for patients separated by mortality status.

	All patients (N = 288)	Survivors (N = 211)	Nonsurvivors (N = 77)
Age, years	65.5 ± 9.1	64.8 ± 8.9	67.4 ± 9.1^*
Males, N (%)	160 (55.6)	110 (52.1)	50 (64.9)^*
BMI, kg/m^2	27.7 ± 6.1	28.7 ± 6.1	25.5 ± 5.2^*
Current smoking, N (%)	84 (29.2)	67 (31.8)	17 (22.1)
Cumulative smoking, pack-years	50.0 ± 21.1	50.3 ± 20.1	48.1 ± 23.1
BDI, total score	7.1 ± 2.1	7.4 ± 2.3	6.1 ± 1.8^*
Charlson Index (total score)	4 [3.5]	3 [3.5]	5.5 [4.0]^*
Coronary artery disease, N	51	32	19
Chronic heart failure, N	34	13	21^*
Cerebrovascular disease, N	22	12	10
Diabetes mellitus, N	39	28	11
COPD-related medications
Short-acting bronchodilators, N	288 (100)	211 (100)	77 (100)
Long-acting bronchodilators, N	243 (84.4)	171 (81.0)	72 (93.5)^*
Inhaled corticosteroids, N	216 (75.0)	149 (70.6)	67 (87.0)^*
Roflumilast, N	24 (8.3)	10 (4.7)	14 (18.2)^*
Lung function
FEV1/FVC	0.49 ± 0.11	0.51 ± 0.11	0.44 ± 0.10^*
FEV1, % pred	61.2 ± 25.0	65.6 ± 25.2	49.1 ± 21.1^*
GOLD stages
I	75 (26.0)	67 (31.8)	8 (10.4)^*
II	103 (35.8)	82 (38.9)	21(27.3)
III/IV	110 (38.2)	62 (29.4)	48 (62.3)^*
TLC, % pred	113.9 ± 16.9	111.1 ± 15.7	121.7 ± 17.8^*
RV, % pred	158.5 ± 61.1	145.5 ± 57.2	193.4 ± 58.0^*
TLCO, % pred	68.6 ± 22.1	70.8 ± 21.1	62.5 ± 23.6^*
TLCO/VA, % pred	85.1 ± 19.5	85.1 ± 19.5	72.9 ± 23.5^*
IC, % pred	85.8 ± 24.1	91.5 ± 23.9	70.2 ± 16.9^*
IC/TLC	0.35 ± 0.11	0.38 ± 0.10	0.27 ± 0.07^*
Cardiopulmonary exercise test^a
Metabolic and cardiovascular responses
VO2, mL/min/kg	16.2 ± 6.0	17.1 ± 6.4	13.9 ± 4.0^*
VO2, % pred	73.5 ± 27.4	77.8 ± 28.3	61.6 ± 20.8^*
HR, bpm	123 ± 20	126 ± 22	117 ± 18^*
HR, % pred	74.5 ± 12.6	75.7 ± 13.0	71.1 ± 10.8^*
O2 pulse, mL/beat	9.9 ± 3.	10.8 ± 4.0	8.6 ± 3.4^*
Ventilatory and gas exchange responses
VE/VCO2 rest	40.2 ± 8.5	38.1 ± 7.3	43.3 ± 10.1^*
VE/VCO2 nadir	34.5 ± 7.1	32.6 ± 6.7	38.7 ± 9.2^*
VE/VCO2 peak	35.0 ± 6.8	33.4 ± 6.1	39.0 ± 9.4^*
PETCO2 nadir, kPa	5.13 ± 0.87	5.37 ± 0.78	5.30 ± 0.94^*
SpO2, %	92.9 ± 5.1	93.3 ± 4.1	91.8 ± 2.7^*
Δ Peak-rest IC, L	−0.42 ± 0.44	−0.40 ± 0.49	−0.46 ± 0.38
IC, L	1.98 ± 0.70	2.08 ± 0.79	1.59 ± 0.52^*
IRV, L	0.58 ± 0.38	0.63 ± 0.41	0.42 ± 0.28^*
VT, L	1.40 ± 0.30	1.46 ± 0.32	1.18 ± 0.20^*
VT/IC	0.71 ± 0.13	0.70 ± 0.13	0.74 ± 0.11^*
Sensorial responses
Dyspnea score	5 [3]	5 [3.5]	5 [2.5]
Leg effort score	5 [3]	5 [3.3]	5 [1.8]^*
Dyspnea ≥ leg effort, N (%)	161 (55.9)	112 (48.3)	59 (76.6)^*
Dyspnea–leg effort	0 [2.5]	0 [3]	0 [1.5]^*

Values are mean ± SD or median [interquartile range]. ^*p < 0.05.

^aPeak exercise responses (unless stated otherwise).

Abbreviations: BMI = body mass index; BDI = baseline dyspnea index; COPD = chronic obstructive pulmonary disease; FEV₁ = forced expiratory volume in one second; FVC = forced vital capacity; GOLD= Global Initiative for Chronic Obstructive Lung Disease; TLC = total lung capacity; RV = residual volume; TL_CO = transfer factor; VA= alveolar volume; IC = inspiratory capacity; IRV = inspiratory reserve volume; VO₂ = oxygen uptake; HR = heart rate; VE = minute ventilation; VCO₂ = carbon dioxide output; PETCO2= end-tidal partial pressure for CO2; SpO₂ = oxy-hemoglobin saturation by pulse oximetry; VT= tidal volume.

Mortality

The mean ± SD follow-up time for the total population was 65 ± 39 months, while the median (interquartile range) was 57 (47) months. Seventy-seven patients (26.7%) died during the follow-up period. Seventy-two patients died after hospital admission and five patients arrived dead in the Hospital Emergency Department. Autopsies were allowed to be performed in 42 patients (54.5%) and the main cause of death was based on the conclusions in these patients. Thirty patients (38.9%) died from respiratory causes: COPD was listed as the main contributing factor in all but two patients. Most common nonrespiratory causes included cardiovascular disease (N = 18), cerebrovascular disease (N = 8), and cancer (N = 7). Median (95% CI) survival for the total population was 123 (105–158) months. No patient underwent lung transplant or lung volume reduction surgery during the follow-up time.

Clinical, functional, and CPET variables in survivors and nonsurvivors

Table 1 shows the differences in patient characteristics separated by mortality status. Nonsurvivors were more frequently males, older, leaner, more dyspneic on daily life, and had a higher co-morbidity burden than the survivors (p < 0.05). From the most prevalent co-morbidities included in the Charlson Index (25), only chronic heart failure was more frequent in nonsurvivors than survivors (Table 1). Nonsurvivors also had more severe lung function impairment as indicated by lower FEV₁, worse lung hyperinflation and air trapping, and lower DL_CO. Metabolic and cardiovascular variables from the CPET indicated than nonsurvivors had lower peak VO₂, larger chronotropic reserves, and lower O₂ pulse. They also showed greater ventilatory response at rest and during exercise, more O₂ desaturation, and greater mechanical constraints. Moreover, dyspnea was more frequently cited as the limiting symptom in this group compared to survivors (p < 0.05; Table 1).

Predictors of mortality

ROC curve analyses contrasting the resting functional variables revealed that IC/TLC has the highest area under the curve for all-cause (Figure 1A) and respiratory (Figure 1C) mortality prediction followed by RV and FEV₁. These variables, however, were closely inter-related (r values ranging from 0.70 to 0.89; p < 0.01). A similar analysis involving the CPET responses showed better prognostic performance for VO₂ peak and VE/VCO₂ nadir (Figure 1B and D) with a lower between-variable correlation (r = −0.57; p < 0.05). Tables 2 and 3 show the optimal thresholds for all-cause and respiratory mortality prediction according to key resting and exercise variables. This analysis revealed that VE/VCO₂ nadir > 34 was the best cut-off for all-cause and respiratory mortality prediction. In contrast, the IC/TLC cut-offs differed: ≤0.34 for all-cause and ≤0.31 for respiratory mortality (Tables 2 and 3). Univariate logistic regression analysis showed that age, BMI, and Charlson Index were also significantly associated with all-cause (Table 4) and respiratory (Table 5) mortality (p < 0.05).

Figure 1. Receiver operating characteristics curve analyses contrasting lung function and exercise variables performance in predicting all-cause and respiratory mortality in patients with COPD. Best cut-offs for individual variables are marked by circles and shown in Tables 2 and 3, respectively. p < 0.05: ^* vs. FEV₁, TLC, and TL_CO/VA (lung function) and vs. O₂ pulse, HR, and SpO₂ (exercise). Abbreviations: COPD = chronic obstructive pulmonary disease; AUC = area under the curve; CI = confidence interval; IC = inspiratory capacity; TLC = total lung capacity; RV = residual volume; FEV₁ = forced expiratory volume in one second; TL_CO = transfer factor; VO₂ = oxygen uptake; VE = minute ventilation; VCO₂ = carbon dioxide output; HR = heart rate; SpO₂ = oxy-hemoglobin saturation by pulse oximetry.

Table 2. All-cause mortality prediction: performance of optimal thresholds for individual lung function and exercise variables.

	Criterion	Youden Index	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)
Lung function
FEV1, % pred	≤55	0.37	0.75 (0.64–0.84)	0.62 (0.55–0.69)	2.01 (1.62–2.54)	0.39 (0.30–0.60)
TLC, % pred	>115	0.32	0.67 (0.55–0.77)	0.64 (0.58–0.71)	1.93 (1.54–2.43)	0.50 (0.42–0.71)
RV, % pred	>150	0.43	0.79 (0.68–0.87)	0.54 (0.57–0.70)	2.23 (1.81–2.84)	0.32 (0.21–0.51)
TLCO, % pred	≤63	0.26	0.64 (0.53–0.75)	0.61 (0.54–0.68)	1.69 (1.33–2.16)	0.57 (0.42–0.84)
IC/TLC	≤0.34	0.48	0.83 (0.72–0.90)	0.65 (0.58–0.72)	2.40 (1.91–3.08)	0.26 (0.24–0.42)
Cardiopulmonary exercise test^a
VO2, % pred	≤75	0.32	0.81 (0.71–0.89)	0.50 (0.43–0.57)	1.64 (1.43–2.09)	0.36 (0.24–0.51)
HR, % pred	≤72	0.24	0.64 (0.53–0.75)	0.59 (0.52–0.65)	1.59 (1.32–2.07)	0.59 (0.43–0.83)
O2 pulse, mL/beat	≤8.5	0.23	0.59 (0.47–0.70)	0.63 (0.56–0.70)	1.64 (1.34–2.16)	0.63 (0.52–0.84)
VE/VCO2nadir	>34	0.31	0.67 (0.54–0.76)	0.65 (0.58–0.72)	1.91 (1.51–2.47)	0.52 (0.43–0.72)
SpO2, %	≤91	0.17	0.37 (0.26–0.49)	0.79 (0.73–0.84)	1.85 (1.24–2.76)	0.78 (0.62–0.92)

^a Peak exercise responses (unless stated otherwise).

CI = confidence interval; PPV = positive predictive value; NPV = negative predictive value; FEV₁ = forced expiratory volume in one second; TLC = total lung capacity; RV = residual volume; TL_CO = transfer factor; IC = inspiratory capacity; VO₂ = oxygen uptake; HR = heart rate; VE = minute ventilation; VCO₂ = carbon dioxide output; SpO₂ = oxy-hemoglobin saturation by pulse oximetry.

Table 3. Respiratory mortality prediction: performance of optimal thresholds for individual lung function and exercise variables.

	Criterion	Youden Index	Sensitivity (95%CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)
Lung function
FEV1, % pred	≤50	0.39	0.79 (0.66–0.88)	0.60 (0.53–0.66)	2.00 (1.64–2.51)	0.34 (0.24–0.66)
TLC, % pred	>120	0.35	0.70 (0.57–0.82)	0.64 (0.57–0.70)	1.98 (1.60–2.50)	0.46 (0.31–0.72)
RV, % pred	>160	0.42	0.79 (0.66–0.88)	0.63 (0.56–0.69)	2.17 (1.81–2.73)	0.33 (0.24–0.55)
TLCO, % pred	≤63	0.29	0,68 (0.55–0.80)	0.43 (0.53–0.66)	1.74 (1.46–2.28)	0.51 (0.32–0.82)
IC/TLC	≤0.31	0.48	0.75 (0.62–0.86)	0.70 (0.63–0.75)	2.53 (2.08–3.22)	0.34 (0.22–0.51)
Cardiopulmonary exercise test^a
VO2, % pred	≤66	0.37	0.74 (0.61–0.84)	0.63 (0.56–0.69)	2.03 (1.62–2.60)	0.41 (0.34–0.62)
HR, % pred	<70	0.18	0.62 (0.48–0.74)	0.56 (0.49–0.63)	1.43 (1.14–1.82)	0.67 (0.51–1.00)
O2 pulse, mL/beat	≤10	0.30	0.93 (0.83–0.98)	0.37 (0.31–0.44)	1.49 (1.32–1.72)	0.18 (0.07–0.54)
VE/VCO2 nadir	≥34	0.45	0.79 (0.66–0.88)	0.66 (0.59–0.72)	2.34 (1.92–2.92)	0.31 (0.20–0.54)
SpO2, %	≤92	0.24	0.53 (0.39–0.66)	0.70 (0.64–0.76)	1.83 (1.31–2.53)	0.66 (0.54–0.96)

^a Peak exercise responses (unless stated otherwise).

CI = confidence interval; PPV = positive predictive value; NPV = negative predictive value; FEV₁ = forced expiratory volume in one second; TLC = total lung capacity; RV = residual volume; TL_CO = transfer factor; IC = inspiratory capacity; VO₂ = oxygen uptake; HR = heart rate; VE = minute ventilation; VCO₂ = carbon dioxide output; SpO₂ = oxy-hemoglobin saturation by pulse oximetry.

Table 4. All-cause mortality prediction: univariate and multivariate logistic regression modeling.

	Univariate analysis		Multivariate analysis^a
	Relative risk (95% CI)	p	Relative risk (95% CI)	p
Clinical
Age, per 1 year increase	1.032 (1.002–1.064)	0.042	1.047 (1.007–1.087)	0.019
Males	0.998 (0.990–1.096)	0.321
BMI, per 1 kg/m^2 less	0.881 (0.832–0.934)	<0.001	0.905 (0.844–0.970)	0.005
BDI, per 1 unit less	0.923 (0.813–1.023)	0.282
Charlson Index, per 1 unit increase	1.101 (1.065–1.142)	<0.001	1.095 (1.058–1.137)	0.001
Lung function
FEV1 ≤ 55% pred	5.100 (2.832–9.186)	<0.001
TLC > 115% pred	3.544 (2.031–6.182)	<0.001
RV > 150% pred	6.772 (3.649–12.565)	<0.001
TLCO ≤ 63% pred	2.913 (1.691–5.018)	0.001	-	-
IC/TLC ≤ 0.34	7.326 (4.423–15.673)	<0.001	2.659 (1.208–5.852)	0.015
Cardiopulmonary exercise test^b
VO2 ≤ 75% pred	3.922 (2.124–7.241)	<0.001	−	−
HR ≤ 72% pred	2.603 (1.520–4.458)	0.004	−	−
O2 pulse ≤ 8.7 mL/beat	2.447 (1.436–4.169)	0.001	−	−
VE/VCO2 nadir > 34	3.708 (2.137–6.433)	<0.001	2.081 (1.170–3.766)	0.010
SpO2 ≤ 91%	2.360 (1.154–4.198)	0.003	−	−

^a Backward stepwise method with Wald's criteria adjusted at nine iterations. Variables presenting multicollinearity with IC/TLC were not considered in the initial model (FEV₁, RV, and TLC). Only variables with statistically significant relative risks are shown.

^b Peak exercise responses (unless stated otherwise).

CI = confidence interval; BMI = body mass index; BDI = baseline dyspnea index; FEV₁ = forced expiratory volume in one second; TLC = total lung capacity; RV = residual volume; TL_CO = transfer factor; IC = inspiratory capacity; VO₂ = oxygen uptake; HR = heart rate; VE = minute ventilation; VCO₂ = carbon dioxide output; SpO₂ = oxy-hemoglobin saturation by pulse oximetry.

Table 5. Respiratory mortality prediction: univariate and multivariate logistic regression modeling.

	Univariate analysis		Multivariate analysis^a
	Relative risk (95% CI)	p	Relative risk (95% CI)	p
Clinical
Age, per 1 year increase	1.036 (1.006–1.070)	0.021	1.046 (1.007–1.087)	0.02
Males	0.998 (0.990–1.096)	0.321
BMI, per 1 kg/m^2 less	0.893 (0.840–0.950)	0.010	0.910 (0.849–0.974)	0.04
BDI, per 1 unit less	0.923 (0.813–1.023)	0.282
Charlson Index, per 1 unit increase	1.081 (1.045–1.122)	0.014	1.035 (1.010–1.121)	0.04
Lung function
FEV1 ≤ 50% pred	6.214 (2.715–10.124)	<0.001
TLC > 120% pred	2.724 (1.984–7.321)	<0.001
RV > 160% pred	6.487 (3.631–12.624)	<0.001
TLCO ≤ 63% pred	2.874 (1.641–5.128)	0.001	−	−
IC/TLC ≤ 0.31	8.154 (4.125–16.081)	<0.001	4.486 (2.156–9.337)	<0.001
Cardiopulmonary exercise test^b
VO2 ≤ 66% pred	3.628 (2.091–7.187)	<0.001	−	−
HR ≤ 70% pred	2.124 (1.314–4.474)	0.002	−	−
O2 pulse ≤ 10.0 mL/beat	2.084 (1.321–4.312)	0.001	−	−
VE/VCO2 nadir > 34	3.999 (2.148–7.021)	<0.001	1.998 (1.432–2.803)	0.010
SpO2 ≤ 92%	2.184 (1.335–4.198)	0.003	−	−

^a Backward stepwise method with Wald's criteria adjusted at nine iterations. Variables presenting multicollinearity with IC/TLC were not considered (FEV₁, RV, and TLC). Only variables with statistically significant relative risks are shown.

^b Peak exercise responses (unless stated otherwise).

CI = confidence interval; BMI = body mass index; BDI = baseline dyspnea index; FEV₁ = forced expiratory volume in one second; TLC = total lung capacity; RV = residual volume; TL_CO = transfer factor; IC = inspiratory capacity; VO₂ = oxygen uptake; HR = heart rate; VE = minute ventilation; VCO₂ = carbon dioxide output; SpO₂ = oxy-hemoglobin saturation by pulse oximetry.

Final results for adjusting the logistic regression analysis over all potential predictors of death are shown in Table 4 for all-cause and in Table 5 for respiratory mortality. FEV₁ and, as expected, RV and TLC showed high multicollinearity with IC/TLC (p < 0.01): these variables, therefore, did not remain in the final models in addition to IC/TLC. Age, BMI, Charlson Index, IC/TLC, and VE/VCO₂ nadir were therefore the independent predictors of mortality. Figure 2 presents the Kaplan–Meier survival curves: compared to patients presenting with VE/VCO₂ nadir ≤ 34 plus IC/TLC > 0.34, survival to all-cause (Figure 2A) and respiratory (Figure 2B) mortality progressively decreased in those with higher VE/VCO₂ nadir or lower IC/TLC. Patients presenting VE/VCO₂ nadir > 34 plus IC/TLC ≤ 0.34 had the poorest survival. For instance, when VE/VCO₂ nadir was added to IC/TLC 10-year survival decreased from 54.7% to 36.8% and 68.7% to 39.0% for all-cause and respiratory mortality, respectively (p < 0.0001; Figure 2).

Figure 2. Kaplan–Meyer survival curves for resting IC/TLC and VE/VCO₂ nadir optimal thresholds as defined by receiver operating characteristics curve analyses. Abbreviations: IC = inspiratory capacity; TLC = total lung capacity; VE = exercise ventilation; VCO₂ = CO₂ output.

We also investigated the determinants of increased VE/VCO₂ nadir (>34). Compared to their counterparts with lower nadirs, patients presenting with VE/VCO₂ nadir > 34 had significantly greater resting VE/VCO₂ (36.8 ± 6.3 vs. 41.9 ± 9.7; p < 0.01), higher VE–VCO₂ intercepts (3.6 ± 3.1 vs. 7.1 ± 4.1; p < 0.05), and a trend to lower VE–VCO₂ slopes (32.1 ± 5.4 vs. 30.5 ± 7.3; p = 0.15). Moreover, the nadir was reached at a significantly lower VO₂ in this group (78.5 ± 18.7% pred vs. 59.8 ± 17.1% pred; p < 0.001).

Discussion

This study investigated whether cardiocirculatory, ventilatory, and pulmonary gas exchange variables measured during incremental CPET would add to standard lung function variables in estimating the risk of mortality in patients with COPD. We found that a widely used marker of ventilatory inefficiency (increased VE/VCO₂ nadir) (22, 23, 27) was the only CPET variable to remain a predictor of mortality when well-established clinical (age, BMI and co-morbidity burden) (5) and resting functional (IC/TLC) (4) variables were accounted for in a multivariate logistic regression analysis. Of note, there was a marked increase in the risk of death when a high VE/VCO₂ nadir (>34) compounded with a low IC/TLC (≤0.34 for all-cause mortality and for ≤0.31 respiratory mortality). These results indicate that ventilatory inefficiency is a key physiological abnormality in COPD (10, 11, 28), which is relevant to mortality prediction across the spectrum of disease severity.

There is a large body of evidence supporting CPET for the prognostic assessment of patients with chronic cardiopulmonary diseases (7, 20, 21). However, CPET remains largely underutilized for risk stratification in COPD (5). In fact, the few previous studies that examined this issue in COPD confined their analysis to traditional end-exercise variables such as peak VO₂ (29). Moreover, those studies typically evaluated patients with severe or very severe COPD in whom variables reflecting advanced lung function impairment (e.g., low TL_CO or hypoxemia) (2, 17) likely overshadowed the prognostic role of CPET.

A key finding of the present study was the independent role of a submaximal, effort-independent CPET variable (VE/VCO₂ nadir) (22, 23, 27) in predicting all-cause and respiratory mortality in a sample with a sizeable number of patients with mild-to-moderate COPD. Of note, VE/VCO₂ nadir overruled more traditional CPET markers of poor prognosis, such as peak VO₂, heart rate, and SpO₂ (7, 29). VE/VCO₂ nadir was particularly relevant in enhancing the prognostic performance of resting lung hyperinflation (IC/TLC) (4). In this context, it has long been established that exercise ventilation is inversely related to PaCO₂ set-point and proportional to the fraction of VT “wasted” in the VD.(12) Although mechanical constraints as COPD progresses may reduce tidal volume (VT) and impair the rate of increase in VE (leading to a low VE–VCO₂slope) (9, 11, 14), the linear VE–VCO₂ relationship shifts upward (i.e., high VE–VCO₂intercept) in tandem with increasing VD/VT (9, 11, 30, 31). Consequently, discrete VE values at a given VCO₂ (VE/VCO₂ ratio or ventilatory equivalent for CO₂) (23) should increase in proportion to V_D/VT in order to maintain PaCO₂ close to the chosen set-point (12). We focus our analysis on VE/VCO₂ nadir as it is not influenced by between-subject differences in lactacidotic drive (12), it is easily obtained from CPET reports, and, as outlined above, conflates the physiological information provided by both VE–VCO₂ slope and intercept (11, 13).

In this context, we previously demonstrated that increased VD/VT was the most consistent pulmonary gas exchange abnormality associated with greater ventilatory inefficiency in patients with largely preserved FEV₁ (28). VD/VT ratio may increase secondary to emphysema as alveolar destruction increases VD and, by promoting lung hyperinflation (reduced IC/TLC), decreases exercise VT (16). Impaired pulmonary microvascular blood flow leading to a dead space effect (high ventilation/perfusion ratios) might be a consequence of vessel loss preceding anatomical emphysema (32) and/or compression by regional hyperinflation (33). Hueper et al. recently found reduced pulmonary microvascular blood flow in nonemphysematous lung regions in mild COPD, suggesting pulmonary vascular destruction/dysfunction before the establishment of macroscopic emphysema (32). Moreover, Holverda et al. (18) and Vonbank et al. (34) showed that COPD patients presenting with pulmonary hypertension at rest had greater exercise ventilation than their counterparts with normal pulmonary vascular pressures. At the other end of the disease spectrum, Armstrong et al. (35) and Kim et al. (36) showed that excision of regional lung units destroyed by emphysema via lung volume reduction surgery significantly lowered VE/VCO₂ nadir. Taken together, these findings suggest that higher VE/VCO₂ nadir in nonsurvivors resulted primarily from higher VD/VT and, accordingly, greater pulmonary microvascular abnormalities and/or larger emphysema extent (5, 6, 15, 17).

It is noteworthy that approximately two-thirds of our patients had a nonrespiratory cause of death. In fact, cardiovascular and cerebrovascular causes of death jointly accounted for a similar number of deaths compared to respiratory causes. In this context, worse lung hyperinflation (low IC/TLC) may have compounded with underlying pulmonary vascular disease to further increase pulmonary vascular resistance and decrease left ventricular filling in nonsurvivors (37). There is recent evidence that pulmonary endothelial dysfunction leading to impaired microvascular blood flow (38) (and high VE/VCO₂ nadir) might be etiologically related to low-grade systemic inflammation and oxidative stress, two well-known negative prognostic markers in cardiocirculatory diseases (37). Moreover, ventilatory inefficiency has been associated with increased neural ventilatory drive secondary to sympathetic over-excitation and increased risk of cardiac death (20, 22). Not unexpectedly, therefore, left ventricular dysfunction (either systolic or diastolic) has been related to higher exercise VE/VCO₂ in COPD.(30) It is thus conceivable that ventilatory inefficiency was associated with greater central hemodynamic impairment and/or higher sympathetic activation during exercise with ominous prognostic implications.

We confirmed the seminal results of Casanova et al. (4) showing that IC/TLC is a key resting functional predictor of respiratory mortality in COPD. Patients with low IC/TLC are likely to be particularly ill-prepared to face greater expiratory flow limitation and acute-on-chronic lung hyperinflation during COPD exacerbations (39). Increased ventilatory requirements secondary to greater wasted ventilation during exacerbations is likely to further increase the burden upon the already overloaded respiratory muscles in patients with increased VE/VCO₂ (39). This might help explain why patients showing the combination of low IC/TLC and high VE/VCO₂ nadir were at a greater risk of exacerbation-related respiratory failure and death. Moreover, assuming that a high VE/VCO₂ primarily reflects poorer ventilation–perfusion matching in COPD (15, 28), it is possible that those with greater ventilatory inefficiency would present with more pronounced changes in arterial blood gases during exacerbations thereby increasing the risk of death (39).

The current investigation has, naturally, some limitations. Our sample was relatively small compared to large epidemiological studies dealing with prognosis in COPD (3), i.e., the number of patients at risk in each prognostic category was not large and further decreased after ∼ 10 years of follow-up (Figure 2). In fact, treatment paradigms for COPD changed over the follow-up period of time and this might have potentially influenced patient's outcome depending on the date of CPET. Notwithstanding these limitations, it constitutes one of the largest series to date to investigate the prognostic relevance of CPET in COPD and the tests were carefully performed in a specialized referral center. The optimal threshold for VE/VCO₂ nadir in COPD (i.e., >34) coincided with that proposed for heart failure (22). The relevance of high VE/VCO₂ nadir to predict all-cause mortality, however, cannot be fully ascribed to increased mortality in co-existent COPD–heart failure as only a quarter of deceased patients had this co-morbidity (Table 1). Although we did not assess test–retest VE/VCO₂ nadir reliability, VE/VCO₂ nadir has been found to be highly reproducible (23, 40). Finally, only a minority of our patients had very severe COPD and it could be argued that pronounced decreases in mechanical–ventilatory reserves in GOLD stage 4 patients would preclude an adequate ventilatory response to metabolic demand. Thus, it remains to be demonstrated whether ventilatory inefficiency would remain a useful prognostic marker in hypercapnic patients with end-stage COPD.

In conclusion, this study is the first to demonstrate that ventilatory inefficiency during incremental CPET (i.e., VE/VCO₂ nadir > 34) adds to established clinical (age, BMI, and co-morbidities) and resting lung function (IC/TLC) variables in estimating the risk of all-cause and respiratory mortality in patients with COPD. Thus, CPET can add important prognostic information to traditional measures of resting lung hyperinflation across the spectrum of COPD severity. These results were obtained in a sample showing predominantly mild-to-moderate airflow obstruction, a large sub-population in which the identification of negative prognostic markers may have a larger impact on disease management compared to more severe COPD. Ventilatory inefficiency constitutes a physiological biomarker linked to key clinical outcomes in COPD: exercise tolerance (9, 11), dyspnea (10, 28), and, as newly established in this study, risk of death. Ventilatory inefficiency in COPD likely conflates the overlapping influences of disease-related functional abnormalities and pathophysiological derangements associated with multiple co-morbidities. Our work also provides novel evidence that physiological abnormalities leading to poorly efficient pulmonary gas exchange and increased wasted ventilation during exercise are relevant in predicting the course of the disease. Therapeutic approaches aiming at improving those abnormalities and decreasing exercise ventilatory inefficiency might have beneficial effects on survival in COPD.

Additional information

Funding

Financial support to Maria C. Alencar and Flavio Arbex was provided by the Brazilian Higher Education National Council. Daniel M. Hirai received funds from the Senate Advisory Research Committee, Queen's University, Ontario, Canada. J. Alberto Neder has been funded by a New Clinician Scientist Program from the Southeastern Ontario Academic Medical Association, Canada. The funders had no role in the study design, data collection and analysis, or preparation of the manuscript.