In this study, we found that estVD/VTrest and BMI < 18.5 kg/m2 were risk factors for mortality, whereas a BMI of 24-26.9 kg/m2 was a protective factor. To the best of our knowledge, this is the first report to link estVD/VT with the risk of death in patients with COPD without cancer. The mechanism underlying the predictive ability of estVD/VTrest for COPD death is not clear; however, it may be due to its broad associations with lung and non-lung factors.23 Based on the hierarchical stratification concept, estVD/VTrest was the primary pulmonary factor and BMI was the non-pulmonary factor.
Lung function variables. Although FEV14,13,14 DLCO16 and IC/TLC17 are the primary pulmonary factors and have been reported to contribute to mortality in patients with COPD, they were not selected in this study. Huang et al investigated the performance of seven staging methods of FEV1, i.e. GOLD, quartiles of FEV1%, z-score of FEV1, quartiles and Miller’s cut-off points (FEV1·height− 2Miller, range: 0.3, 0.4, 0.5) of the ratio of FEV1 over height squared, quartiles of the ratio of FEV1·height− 3 and FEV1 quotient (FEV1Q i.e. FEV1 in liters/0.5 L for males; 0.4 L for females) in predicting outcomes of patients with COPD15, and found that staging based on quartile of FEV1Q was the best predictor, followed by FEV1·height− 2Miller. We tested these two variables in univariate survival analysis, and found that the fourth quartile of FEV1Q (i.e. ≥3.82) had a protective effect (0.41 [0.17–0.98], p = 0.046), whereas FEV1·height− 2Miller did not. Although other authors did not find that FEV1% was a protective factor,9 FEV1% was a risk factor for mortality in univariate analysis after excluding subjects with cancer and adjustments in this study. This type of paradoxical phenomenon occasionally happens when multiple regression and adjustments are performed.
In contrast, estVD/VTrest, estVD/VTpeak, PEF%, DLCO/VA%, and VA% were significantly contributing primary lung factors to all-cause mortality in the patients with COPD before adjustment in this study (Table 2). VD/VTrest is a sophisticated marker of resting gas exchange and is more related to cigarette smoking, carboxyhemoglobin level, pulmonary hypertension and PaCO2 than FEV1%.23 Moreover, elevated PaCO2 is related to mortality.18,24−26 VD/VTrest can be estimated by cigarette consumption, minute ventilation/CO2 output, arterial oxyhemoglobin saturation, and tidal volume ⋅ inspiratory duty cycle.23 In this study, we re-derived the estVD/VTrest prediction equation using lung function variables and demographic data alone. This simplification may allow for the more general use of the prediction equation, even though the predictive power was lower than the previous equation, calculating estVD/VTrest requires many variables, and the method has yet to be validated. After multivariable Cox regression analysis, estVD/VTrest remained a significant risk factor (Table 3), however estVD/VTpeak did not. This might be because estVD/VTpeak and estVD/VTrest were co-linear (r = 0.36, p = 0.02).
Table 3
Stepwise Cox proportional hazard model analysis for risk of all-cause mortality.
| Crude hazard ratio (95% C.I.) | Adjusted hazard ratio (95% C.I.) |
estVD/VTrest, % | 1.07 (1.01–1.13)* | 1.07 (1.01–1.13)* |
estVD/VTpeak, % | 1.05 (1.00-1.10)* | |
FVC% | 1.00 (0.98–1.01) | |
FEV1% | 1.00 (0.99–1.02) | |
MMEF% | 1.01 (0.99–1.03) | |
PEF% | 0.99 (0.97-1.00)* | |
TLC% | 1.00 (0.98–1.02) | |
RV/TLC% | 1.00 (0.99–1.02) | |
DLCO/VA% | 0.98 (0.97-1.00)✝ | |
VA% | 0.98 (0.96-1.00)* | |
Age | 1.03 (1.00-1.06)* | |
Sex | | |
Female | 1 | |
Male | 0.88 (0.21–3.64) | |
Body mass index, kg/m2 | | |
18.5–23.9 | 1 | 1 |
< 18.5 | 2.68 (1.29–5.57)✝ | 2.43 (1.15–5.14)* |
24-26.9 | 0.34 (0.14–0.83)* | 0.34 (0.14–0.84)* |
≥ 27 | 0.42 (0.17–1.02) | 0.42 (0.17–1.02) |
Charlson comorbidity index; | | |
≤ 2 | 1 | |
> 2 | 2.99 (0.72–12.30) | |
Bronchodilator use | | |
None | 1 | |
Yes | 0.85 (0.45–1.61) | |
Hospitalized AECOPD | | |
0 | 1 | 1 |
1 | 0.58 (0.18–1.91) | 0.39 (0.12–1.30) |
≥2 | 2.50 (1.31–4.77)✝ | 1.73 (0.88–3.39) |
Abbreviations: AECOPD, acute exacerbation of COPD; DLCO, diffusing capacity of lung for carbon monoxide; estVD/VTrest and estVD/VTpeak, estimated dead space and tidal volume ratios at rest and at peak exercise; FEV1, forced expired volume in one second; FRC, functional residual capacity; FVC, forced vital capacity; MMEF, maximum mid-expiratory flow; PEF, peak expiratory flow; RV, residual volume; TLC, total lung capacity; VA, alveolar volume. * p < 0.05, ✝ <0.01. |
BMI. Reduced BMI is an independent risk factor for COPD and mortality.36,37 Even in matched-FEV1, BMI has still been reported to be a marker of COPD phenotypes.38 COPD patients with a reduced BMI may have a higher rate of impaired peripheral oxygenation (anemia, circulation impairment and deconditioning), where they are taller and more malnourished, anemic and have more hyper-inflation, air-trapping, and diffusion impairment.38 According to the 10-point BODE index (B: body mass index [BMI], O: obstruction of airflow, D: dyspnea score, E: exercise capacity delineated by six-minute walking distance), BMI > 21 kg/m2 is protective for survival, whereas BMI ≤ 21 kg/m2 is detrimental.3 Moreover, FEV1% and mid-thigh muscle cross sectional area obtained by computed tomography are related to survival,39 suggesting that fat-free muscle mass is important; however, BMI is easily measured. In this study, the under-weight COPD patients (BMI < 18.5 kg/m2) had a high risk of death (HR of 2.68), the overweight patients (BMI, 24-26.9 kg/m2) had a low risk of death (HR of 0.34), and the obese patients (BMI ≥27 kg/m2) had neither effect (Fig. 2). However, obesity was a risk factor for poor COPD-related outcomes (quality of life, dyspnea, six-minute walking distance, and severe AECOPD) and was dose-dependent.40
Other factors. Some studies have not encompassed AECOPDs or co-morbidities when performing survival analysis. For example, Martinez et al included many risk variables in a Cox regression model9; other studies constructed composite indexes: BODE3; DO (D: dyspnea score, O: obstruction of airflow i.e. pre-bronchodilator FEV1% or GOLD grade and exertional dyspnea);41 and ADO (A: age, D: dyspnea score, O: obstruction of airflow).5 In addition, Soler-Cataluna et al integrated exacerbations into the BODE index, where the “E” of BODE was replaced.10 This omitted the cumbersome six-minute walking test and did not lose power of survival prediction.10 Although acute respiratory failure accounted for half of the causes of non-cancer death in the current study, acute respiratory failure death only accounted for 9 of 109 AECOPD. The co-morbidity test (COTE) index has been used to assess the risk of mortality in patients with COPD.21 Recently, individual diseases such as heart failure and ischemic heart disease rather than CCI score have been shown to have a large effect size on mortality prediction.42 However, CCI score was used in the current study and stratified as ≤2 or > 2. Thus, AECOPD and co-morbidities did not contribute to the risk of mortality in multivariable analysis. Even though there are several prediction models, none of them is perfect, and the current study may be helpful to refine future models.
Hierarchical stratification of risk factors for survival analysis. None of the aforementioned studies mentioned the concept of hierarchical stratification of risk factors for survival analysis. All of these factors can be stratified from primary to quaternary. A similar study on hospitalizations and all-cause mortality used logistic regression with stepwise risk stratification but not hierarchical stratification was published recently43, in which demographic and COPD-specific data, and multi-morbidities were used. However, these variables were not sorted clearly, i.e. hemoglobin as a COPD-specific variable and anemia as a co-morbidity variable; BMI as a COPD-specific variable. They also included cancer and lung cancer as co-morbidities; however, mortality was best predicted by disease severity (area under the curve [AUC] 0.816; 95% CI 0.805– 0.827), and the predictive ability was only marginally enhanced by adding multi-morbidity indices (AUC 0.829; 95% CI 0.818–0.839). In contrast, we found that cancer was the only risk factor associated with mortality after adjustments, with an adjusted HR of 8.46 (95% CI 3.91–18.31). After excluding patients with cancer, BMI and estVD/VTrest were significantly associated. The discrepancies across these studies may be due to the use of different variables such as CCI or individual diseases42 to score co-morbidities, and differences in the stratification of risk factors.43 Lastly, composite indexes are generally thought to be better than the primary lung variables for survival prediction. However, some reports do not support this notion, partly because survival prediction is not their initial purpose.4,6,7,13,14
Study limitations. As this was a retrospective study, some variables could not be assessed, such as emphysema9, secondary factors i.e. hypoxemia, hypercapnia,18 and other gas exchange variables, other tertiary factors i.e. dyspnea, peak oxygen uptake, frailty19,44, and health-related quality of life.21 In this context, composite indexes3, 5–7,10,13 cannot be calibrated with this dataset. In addition, COPD-asthma overlap, a phenotype with different outcomes,45 cannot be fully excluded, even though none of the subjects had a bronchodilator effect in spirometry and were former or current cigarette smokers. Being a current smoker is a strong risk factor for mortality in patients with COPD42,46; however, the status of cigarette smoking in some patients was not clear. In addition, as estVD/VTrest and estVD/VTpeak were not measured and their predictive formulae have not been externally validated and only a few patients were included in the study, further studies are needed to verify our results.