Comparing the 2007 and 2011 GOLD Classifications as Predictors of all-Cause Mortality and Morbidity in COPD

ABSTRACT

To better classify patients with chronic obstructive pulmonary disease (COPD) for prognostic purposes and to tailor treatment, the Global Initiative for Chronic Obstructive Lung Disease (GOLD) 2007 classification was revised in 2011. The primary aim of the current data analyses was to evaluate the accuracy of the GOLD 2007 and 2011 GOLD classifications to predict all-cause mortality and morbidity in a well-described COPD cohort. The prognostic values of both GOLD classifications, expressed as the C-statistic, were assessed in the Cohort of Mortality and Inflammation in COPD (COMIC) study of 795 COPD patients, with a follow-up of 3 years. Outcomes were all-cause mortality and morbidity. Morbidity was defined as time until first COPD-related hospitalisation and time until first community-acquired pneumonia (CAP). The prognostic value of the GOLD 2011 classification was compared between symptom classification based on the modified Medical Research Council (mMRC) score and the Clinical COPD Questionnaire (CCQ) scores with two different thresholds. Although the GOLD 2011 CCQ classification had the highest accuracy to predict mortality and morbidity in our study, the C-statistics differed only numerically. Furthermore, our study showed that the instrument used to determine the level of symptoms in the GOLD 2011 classification has not only important consequences on the mortality prognosis, but also affects the morbidity prognosis in COPD. Therefore, patients' estimated prognosis could alter when different types of tools are used to evaluate the prognosis.

KEYWORDS:

• Acute exacerbation COPD
• C-statistic
• pneumonia
• prognosis survival

Abbreviations List

AECOPD	=	Acute exacerbation of COPD
BMI	=	Body mass index
BODE	=	Body mass index, airflow obstruction, dyspnea, and exercise capacity
CAP	=	Community-acquired pneumonia
CAT	=	COPD Assessment Test
CCQ	=	Clinical COPD Questionnaire
CI	=	Confidence interval
COMIC	=	Cohort of Mortality and Inflammation in COPD
COPD	=	Chronic obstructive pulmonary disease
COTE	=	COPD-specific co-morbidity test
FEV1	=	Forced expiratory volume in 1 second
GOLD	=	Global Initiative for Chronic Obstructive Lung Disease
HR	=	Hazard ratio
IQR	=	Interquartile range
mMRC	=	modified Medical Research Council Dyspnoea Grade
SD	=	Standard deviation

Introduction

COPD is an important cause of morbidity and mortality. Until 2011, severity classification in COPD was solely based on FEV¹ (% predicted). To improve classification of patients for prognostic purposes and to tailor treatment, the 2011 revision of the GOLD strategy (1) recommends the assessment of patients with COPD using three different domains: severity of airflow obstruction, its impact on dyspnoea or patient's health status, and risk of future events (estimated by the severity of airflow limitation and the history of previous exacerbations) (Figure 1) (1).

Figure 1. Risk classification based on GOLD 2011.

Although the scientific committee of the GOLD strategy acknowledges that the disease classification in ABCD categories is mainly for disease management and not for prognostic purposes, several studies have already compared its survival or exacerbation risk predictive capacity with the previous FEV¹-based classification (GOLD 2007) (2–8).

Agusti et al.^{Figure 1} (5), Soriano et al. (2, 8) and Johannessen et al. (6) already showed that the prognostic validity of the GOLD 2011 classification to predict time to death is no different than the GOLD 2007 staging based on spirometry only, while Leivseth et al. (3) even showed that the GOLD 2007 classification predicted mortality better than the GOLD 2011 classification. However, Agusti et al. did show that compared to the 2007 GOLD classification, the 2011 GOLD classification had a higher concordance probability for predicting exacerbations and hospitalisations, which was also observed by Lange et al. (4).

In all the studies mentioned above, the 2011 GOLD classification was based on the modified Medical Research Council Dyspnoea Grade (mMRC) instead of the COPD Assessment Test (CAT) for assessing symptoms other than dyspnoea, since this questionnaire was not available at the onset of all these cohorts. Also the Clinical COPD Questionnaire (CCQ), which was proposed in a later stage for assessing symptoms, was not used in these studies.

Very recently, Casanova et al. (9) published results on the prognostic value of the CAT and CCQ scores in terms of mortality and compared it with the mMRC score. They observed that the CAT and the CCQ had similar ability for predicting all-cause mortality in patients with COPD, but that they were inferior to the mMRC score. They suggested new thresholds for CAT and CCQ scores based on mortality risk that could be useful for the new GOLD grading classification.

Although the study of Casanova et al. compared mMRC, CAT and CCQ, it only studied the prognostic value in terms of mortality. The primary aim of the current analyses was therefore to evaluate the accuracy of the GOLD 2007 and GOLD 2011 classifications, based on the mMRC score and on the CCQ scores with different thresholds, to predict not only all-cause mortality but also morbidity in a well-defined COPD cohort.

Methods

Setting and study population

The COMIC study is a single-centre cohort study from Enschede, the Netherlands (10). From December 2005 till April 2010, 795 patients were included with a follow-up period of at least three years. Patients had to meet the following criteria: a) a clinical diagnosis of COPD according to the GOLD guidelines, b) current or former smoker, c) age ≥40 years, d) no medical condition compromising survival within the follow-up period or serious psychiatric morbidity, e) absence of any other active lung disease (e.g. sarcoidosis), f) no maintenance therapy with antibiotics and g) ability to speak Dutch. Patients were enrolled when visiting the outpatient clinic in stable state (stable-state group) or when hospitalised for an acute exacerbation in COPD (acute exacerbation of COPD [AECOPD] group).

Patients from the AECOPD group who never became stable again before dying were excluded from the present analyses. For all other patients mMRC and CCQ were assessed in stable state and follow-up was calculated from this time point.

GOLD classifications

The number of exacerbations in the year prior to COMIC inclusion was extracted from patients’ pharmacy records. Exacerbations were defined as a prescription of oral corticosteroids for at least five days to a maximum of 15 days, with a minimal dose of 20 mg. The time between two prescriptions of oral corticosteroids had to be at least 28 days. If not, the exacerbation was regarded as a relapse of the prior exacerbation (11).

All hospital admissions in the year prior to inclusion and within the three-year follow-up period were screened for COPD exacerbations and were deemed severe AECOPD.

The mMRC (12) and CCQ (13) questionnaires were obtained at stable state and the cut-off points, as defined by the GOLD guidelines, of >1 for mMRC, and >1.0 or >1.5 for the CCQ score were used.

Outcomes

The primary outcome parameter was survival, based on all-cause mortality. Date of death was verified from public registries.

Morbidity was defined as time until first hospitalisation for an AECOPD (severe AECOPD) and time until first community-acquired pneumonia (CAP).

AECOPD was defined as an acute negative change from baseline, reported by the patient, in dyspnoea and/or sputum volume and/or colour of sputum (yellowish or greenish sputum) and/or cough, which may warrant additional treatment of prednisolone with or without antibiotics by a physician, in a patient^{Figure 2} with underlying COPD. Pneumonia was defined as an acute respiratory tract illness associated with radiographic shadowing on a chest radiograph consistent with infection which was neither pre-existing nor of any other known cause (14). All X-rays were double read by a radiologist and a pulmonary physician. In case of doubtful shadows in the report, the X-ray was presented to another pulmonary physician for final judgment.

Demographic data were collected from medical records. Spirometry was performed by trained respiratory technicians according to the American Thoracic Society guidelines (15). Smoking status was determined by the Vlagtwedde questionnaire (16). Data on common co-morbidities like myocardial infarction, congestive heart failure and diabetes mellitus were collected.

Statistics

Continuous variables are expressed as mean with standard deviation (SD) or median with interquartile range (IQR); categorical variables as counts with the corresponding percentages.

We analysed time to death, first severe AECOPD and first pneumonia by Kaplan–Meier survival curves.

We used univariate and multivariate Cox proportional hazard regression models to establish the relationship of the different GOLD classifications, with all-cause mortality, first severe AECOPD and first pneumonia. All multivariate models were corrected for the confounders age at inclusion, BMI, gender, and presence of the co-morbidities myocardial infarction, congestive heart failure and diabetes mellitus. Furthermore, the C-statistic (Harrell's C) (17) for the univariate and multivariate Cox-regression models were calculated. All tests were two-sided and p ≤ 0.05 was considered statistically significant. Data were analysed using SPSS, version 20 (SPSS Inc. Chicago IL).

Results

In total 665 of the 795 patients from the COMIC study had complete data for the GOLD 2007 and GOLD 2011 classification with both mMRC and CCQ as symptom scores. Table 1 shows the baseline characteristics.

Table 1. Baseline characteristics.

	N  =  665
Age (SD)	67 (10)
Males, N (%)	402 (61)
Lung function
Mean FEV^1 litres (SD)	1.4 (0.6)
Mean FEV^1% pred (SD)	53 (19)
Mean FEV^1%/VC (SD)	44 (13)
Current smokers, N (%)	173 (26)
Median pack-years (IQR)	36 (23–50)
Mean BMI (SD)	27 (5)
Mean mMRC score (SD)	1.8 (1.3)
Mean CCQ score (SD)	1.8 (1.0)

Figure 2 shows^{Figure 3} how patients were reclassified based on the GOLD 2007 into the GOLD 2011 classifications with mMRC and CCQ (with cut-off 1.0 and 1.5) as symptom scores. The reclassification proportions into A, B, C or D of the GOLD 2011 classification differed according to the symptom score and cut-off value used. Reclassification based on the mMRC score produced a similar pattern as reclassification based on the CCQ score with cut-off 1.5. When the lower cut-off of 1.0 was used for CCQ, a higher percentage of patient from GOLD 2007 stages I and II were reclassified as stage B instead of A, and a higher percentage of patient from stages III and IV were reclassified as stage D instead of C. Although the pattern of the reclassification based on mMRC and CCQ with cut-off 1.5 were similar, they do not reclassify patients in the same stages^{Figure 4} as is shown in Figure 3 (see also Table 3).

Figure 2. Reclassification of patients based on GOLD 2007 into GOLD 2011 mMRC and CCQ scores with cut-off 1.0 and 1.5.

Figure 3. Differences in reclassification between GOLD 2011 mMRC and CCQ with cut-off 1.5.

Mortality risk prediction

Univariate

In total 142 (21.4%) patients died within the 3-year follow-up. Figure 4 shows the Kaplan–Meier curves for the GOLD 2007 and 2011 classifications. The survival curve with the GOLD 2011 mMRC score seems to differentiate most between all four stages, while the GOLD 2011 CCQ scores show overlapping risks in stages A, B and C. With the GOLD 2007 classifications there seems overlap between stages I and II and between stages III and IV. For the complete univariate risk prediction, see the online supplement.

Figure 4. Kaplan–Meier survival curves. a) survival curve GOLD 2007, b) survival curve GOLD 2011 mMRC, c) survival curve GOLD 2011 CCQ 1.0 and d) survival curve GOLD 2011 CCQ 1.5.

Multivariate

GOLD 2007

Patients in stage IV had a significantly higher risk of dying compared to patients in stages I, II and III, with hazard ratios (HRs) of 4.6 (95%confidence interval [CI] 1.9–11.1), 4.5 (95%CI 2.6–7.6) and 1.8 (95% CI 1.2–2.9), respectively. Furthermore, patients in stage III had a 2.5 fold (95%CI 1.1–5.6) and a 2.5 fold (95%CI 1.6–3.7) higher risk compared to stages I and II, respectively.^{Figure 5}

GOLD 2011 mMRC

Patients in category D had a significantly higher risk of dying than patients in category A and B, with HRs of 4.9 (95%CI 2.6–9.3) and 2.8 (95%CI 1.5–5.3), respectively. Patients in category C had a significantly higher risk of dying than patients in category A and B, with HRs of 3.6 fold (95%CI 1.8–7.3) and 2.0 (95%CI 1.0–4.2), respectively.

GOLD 2011 CCQ 1.0

Patients in category D had a significantly higher risk of dying compared to patients in category A, B and C, with HRs of 4.5 (95%CI 2.0–10.3), 3.9 (95%CI 2.3–6.6) and 3.1 (95%CI 1.5–6.4), respectively.

GOLD 2011 CCQ 1.5

Patients in category D had a significantly higher risk of dying compared to patients in category A, B and C, with HRs of 5.8 (95%CI 2.9–11.6), 3.2 (95%CI 1.8–5.9) and 2.0 (95%CI 1.3–3.2), respectively. Furthermore, patients in category C had a 2.9 fold (95%CI 1.3–6.2) higher risk compared to category A. The C-statistics for both the univariate and multivariate mortality risk prediction models are displayed in Table 2^{Table 3}, in which the GOLD 2011 CCQ models showed numerically the highest C-statistic.

Table 2. Distribution of GOLD stages/classification.

	N  =  665
GOLD 2007, N (%)
Stage I	58 (8.7)
Stage II	283 (42.6)
Stage III	263 (39.5)
Stage IV	61 (9.2)
GOLD 2011 mMRC
A	148 (22.3)
B	88 (13.2)
C	149 (22.4)
D	280 (42.1)
GOLD 2011 CCQ 1.0
A	85 (12.8)
B	151 (22.7)
C	82 (12.3)
D	347 (52.2)
GOLD 2011 CCQ 1.5
A	140 (21.1)
B	96 (14.4)
C	154 (23.2)
D	275 (41.4)

Table 3. C-statistic of the univariate and multivariate prediction models.

	GOLD 2007	GOLD 2011 mMRC	GOLD 2011 CCQ 1.0	GOLD 2011 CCQ 1.5
Univariate prediction models
Survival	0.63	0.64	0.65	0.66
Hospitalisation AECOPD	0.62	0.65	0.67	0.67
Pneumonia	0.59	0.59	0.63	0.63
Multivariate prediction models
Survival	0.75	0.76	0.77	0.78
Hospitalisation AECOPD	0.69	0.69	0.72	0.71
Pneumonia	0.70	0.72	0.74	0.75

Multivariate prediction models were corrected for age at inclusion, BMI, gender and presence of the co-morbidities myocardial infarction, congestive heart failure and diabetes melitus.

Severe AECOPD risk prediction

Univariate

In total 279 (42.0%) patients had a severe AECOPD. Figure 5^{Figure 6} shows the Kaplan–Meier curves for the GOLD 2007 classification and the GOLD 2011 classifications. The severe AECOPD curves of the GOLD 2007 classification seem to differentiate most between all four stages, followed by the GOLD 2011 CCQ 1.0 score, who showed a remarkable higher risk for an severe AECOPD in group B than in C. The curves of GOLD mMRC and CCQ 1.5 scores showed overlapping risks in stages B and C. For the complete univariate risk prediction, see the online supplement.

Figure 5. Kaplan–Meier curves for time until first severe AECOPD. a) GOLD 2007, b) GOLD 2011 mMRC, c) GOLD 2011 CCQ 1.0 and d) GOLD 2011 CCQ 1.5.

Multivariate

GOLD 2007

Patients in stage IV had a significantly higher risk of a severe AECOPD compared to patients in stages I, II and III with HRs of 6.4 (95%CI 3.2–12.8), 3.1 (95%CI 2.1–4.7) and 1.6 (95% CI 1.1–2.3), respectively. Furthermore, patients in stage III had a 4.1 fold (95%CI 2.2–7.5) and 2.0 fold (95%CI 1.5–2.6) higher risk compared to patients in stage I and II, respectively, while patient in stage II had a 2.0 fold (95%CI 1.1–3.8) higher risk compared to patients in stage I.

GOLD 2011 mMRC

Patients in category D had a significantly higher risk of a severe AECOPD compared to patients in category A, B and C with HRs of 4.2 (95%CI 2.8–6.1), 2.5 (95%CI 1.7–3.8) and 2.1 (95%CI 1.5–2.8), respectively. Furthermore, patients in category C had a 2.0 fold (95%CI 1.3–3.2) higher risk compared to patients in category A.

GOLD 2011 CCQ 1.0

Patients in category D had a significantly higher risk of a severe AECOPD compared to patients in category A, B and C with HRs of 6.8 (95%CI 3.7–12.6), 2.6 (95%CI 1.9–3.5) and 3.7 (95%CI 2.3–6.2), respectively. Furthermore, patients in category B had a 2.7 fold (95%CI 1.4–5.1) higher risk compared to patients in category A.

GOLD 2011 CCQ 1.5

Patients in category D had a significantly higher risk of a severe AECOPD compared to patients in categories A, B and C with HRs of 6.3 (95%CI 4.0–9.9), 2.1 (95%CI 1.5–3.2) and 2.5 (95%CI 1.8–3.4), respectively. Furthermore, patients in categories B and C had a 2.9 fold (95%CI 1.7–4.9) and 2.5 fold (95%CI 1.5–4.2) higher risk compared to patients in category A. The C-statistics for both the univariate and multivariate severe AECOPD risk prediction models are displayed in Table 2, in which the GOLD 2011 CCQ models showed numerically the highest C-statistic.

Pneumonia risk prediction

Univariate

In total 179 (26.9%) patients had a community acquired pneumonia. Figure 6 shows the Kaplan–Meier curves for the GOLD 2007 classification and the GOLD 2011 classifications. The pneumonia curves showed overlap in all classifications. For GOLD 2007 there was an overlap between stages I and II and between stages III and IV, similar as in the survival curves. The GOLD CCQ 1.0 score showed overlap between stages A, B and C; the GOLD mMRC score showed somewhat overlap between stages A and B and between stages C and D; and the GOLD CCQ1.5 score showed mainly overlap between stages B and C. For the complete univariate risk prediction, see the online supplement.

Figure 6. Kaplan–Meier curves for time until first pneumonia. a) GOLD 2007, b) GOLD 2011 mMRC, c) GOLD 2011 CCQ 1.0, d) GOLD 2011 CCQ 1.5.

Multivariate

GOLD 2007

Patients in stage IV had a significantly higher risk of a pneumonia compared to patients in stages I and II with HRs of 2.8 (95%CI 1.3–6.2) and 2.1 (95%CI 1.3–3.5), respectively. Patients in stage III had a 2.5 fold (95%CI 1.3–4.9) and 1.9 fold (95%CI 1.3–2.6) higher risk compared to stages I and II.

GOLD 2011 mMRC

Patients in category D had a significantly higher risk of a pneumonia compared to patients in categories A and B, with HRs of 2.3 (95%CI 1.5–3.6) and 1.8 (95%CI 1.1–3.1), respectively. Furthermore, patients in category C had a 1.8 fold (95%CI 1.1–2.9) higher risk than patients in category A.

GOLD 2011 CCQ 1.0

Patients in category D had a significantly higher risk of pneumonia compared to patients in categories A, B and C, with HRs of 3.1 (95%CI 1.7–5.7), 2.1 (95%CI 1.4–3.0) and 3.6 (95%CI 1.9–7.0), respectively.

GOLD 2011 CCQ 1.5

Patients in category D had a significantly higher risk of a pneumonia compared to patients in categories A, B and C with HRs of 3.5 (95%CI 2.1–5.7), 1.8 (95%CI 1.1–2.8) and 2.3 (95%CI 1.5–3.4), respectively. Furthermore, patients in category B had a 2.0 fold (95%CI 1.1–3.6) higher risk compared to category A. The C-statistics for both the univariate and multivariate pneumonia risk prediction models are displayed in Table 2, in which the GOLD 2011 CCQ models showed numerically the highest C-statistic.

Discussion

The current study evaluated the accuracy of the GOLD 2007 and GOLD 2011 classifications, based on the mMRC score or the CCQ scores with different thresholds, to predict all-cause mortality and morbidity in a well-defined COPD population.

The Kaplan–Meier curves seemed to show the largest differentiation between groups in survival risk with the GOLD 2011 mMRC classification and in severe AECOPD risk with the GOLD 2007 classification. However, the C-statistic, as indication of diagnostic or prognostic accuracy, for the multivariate mortality and morbidity risk prediction models was in general numerically the lowest for GOLD 2007, followed by the GOLD 2011 mMRC classification. There was only a small difference in the accuracy of the prediction of severe AECOPD between the GOLD 2011 CCQ classifications with cut-off 1.0 and 1.5, with the first having a slightly higher accuracy to predict a severe AECOPD and the latter having a slightly higher accuracy to predict mortality and pneumonia.^{Table 2Table 3}

This observation is not in line with the study of Casanova et al. (9), which showed that the CCQ was inferior to the mMRC for predicting mortality in COPD patients. Furthermore, Casanova et al. suggested a new cut-off value of >2.5 for the CCQ, since in their study it was shown to predict mortality more accurately. When we analysed (data not shown) this cut-off value we indeed observed a higher C-statistic for mortality prediction (C  =  0.78) and also the C-statistics for the risk prediction of a severe AECOPD, which was not studied by Casanova et al., was higher (C  =  0.75). The C-statistic for prediction of pneumonia was somewhat lower (C = 0.70). It therefore seems that the cut-off of 2.5 is indeed a better cut-off to predict overall prognosis, although the prediction for pneumonia was somewhat less accurate than the 1.0 and1.5 cut-offs. Although the CCQ had the highest C-statistic in our study, they are still only numerical differences. The prognostic value of our GOLD 2011 models was comparable to the prognostic value reported by de Torres et al. (18). They however showed that the BODE index, especially when it includes the COTE index, is a better tool to evaluate the severity of COPD and ultimate risk of death at short and long term than the proposed multidimensional ABCD GOLD category classification.

The new GOLD strategy states that it is unnecessary to use more than one tool, but that a more comprehensive symptom assessment should be performed rather than generating a simple measure of breathlessness (1). The choice of this tool however affects the risk classification of patients, which is shown in Figures 2 and 3 and also observed in previous studies (7, 19, 20). After this, it is shown that the risk prediction per tool not only differs for mortality, as was observed by Casanova et al., but also differs for morbidity.

Although the ascending stages or categories in all classifications were often associated with an increased mortality or morbidity risk, multiple stages and categories showed, sometimes completely, overlapping risks. While the GOLD 2007 classification showed much overlap between stages I and II, the GOLD 2011 classifications showed often overlap between stages B and C. This overlap was also observed by Han et al. (7). While Lange et al. (4) observed that survival was better in group C than in B based on the GOLD 2011 mMRC classification, we observed that the risk of a severe AECOPD and a pneumonia was higher in group B than in C based on the GOLD 2011 CCQ classification with cut-off of 1.0 and 1.5, respectively.

Our study has several limitations. First, we did not include the CAT, since it was not available at the start of the COMIC cohort. Therefore, we cannot compare all symptom tools in our study.

Second, not all patients from the COMIC study had complete data for both the GOLD 2007 and the GOLD 2011 classifications. This was mainly due to the inclusion of both patients at stable state and during an severe AECOPD. Some of the patients who were included during a severe AECOPD died before they reached stable state in which the CCQ and mMRC was administered, therefore the GOLD 2011 classification could not be defined in these patients. Third, the symptom tools were only recorded once at stable state, so the longitudinal stability of the different GOLD classifications could not be assessed.

Fourth, it could be possible that not all types of pneumonia have been identified. In case COPD patients present themselves at the emergency department of the hospital with respiratory complaints, they will receive an X-thorax as standard care on which pneumonia can be defined. However, it is possible that some outpatients have not mentioned any symptoms and signs of pneumonia and have only received a treatment for a mild or moderate AECOPD.

In conclusion, our study confirms that the instrument used to determine the level of symptoms in the GOLD 2011 classification has important consequences on the mortality prognosis. Furthermore, it showed that the choice of symptom tool also affects the morbidity prognosis in COPD. Therefore, patients' estimated prognosis could alter when different types of tools are used to evaluate the prognosis.

Funding

The COMIC study was supported by the Department of Pulmonology, Medisch Spectrum Twente and by an unrestricted grant from GlaxoSmithKline.