Articles

Prediction of mortality in patients with chronic obstructive

pulmonary disease with the new Global Initiative for Chronic
Obstructive Lung Disease 2017 classification: a cohort study
Anne Gedebjerg, Szimonetta Komjáthiné Szépligeti, Laura-Maria Holm Wackerhausen, Erzsébet Horváth-Puhó, Ronald Dahl, Jens Georg Hansen,
Henrik Toft Sørensen, Mette Nørgaard, Peter Lange, Reimar Wernich Thomsen

Summary
Background The Global Initiative for Chronic Obstructive Lung Disease (GOLD) 2017 classification separates the Lancet Respir Med 2018
spirometric 1–4 staging from the ABCD groups defined by symptoms and exacerbations. Little is known about how Published Online
this new classification predicts mortality in patients with chronic obstructive pulmonary disease (COPD). We aimed January 10, 2018
http://dx.doi.org/10.1016/
to establish the predictive ability of the GOLD 2017 classification, compared with earlier classifications, for all-cause
S2213-2600(18)30002-X
and respiratory mortality, both when using its main ABCD groups and when further subdividing according to
See Online/Comment
spirometric 1–4 staging. http://dx.doi.org/10.1016/
S2213-2600(18)30001-8
Methods In this nationwide cohort study, we enrolled patients with COPD with data available in the Danish registry for Department of Clinical
COPD. To be included in this registry, individuals must have been outpatients in hospital-based pulmonary clinics in Epidemiology, Institute of
Denmark. Eligible patients were aged 30 years or older; had received a primary diagnosis of COPD (International Clinical Medicine
(A Gedebjerg MD,
Classification of Diseases [ICD]-10 J44.X) or acute respiratory failure (ICD-10 J96.X) in combination with COPD (ICD-10 S K Szépligeti MSc,
J44.X) as a secondary diagnosis; and had complete data on FEV1, body-mass index, modified Medical Research Council L-M H Wackerhausen MD,
dyspnoea scale score, and smoking status. We categorised eligible patients with complete data according to the 2007, E Horváth-Puhó PhD,
2011, and 2017 GOLD classifications at the first contact with an outpatient clinic. For the GOLD 2017 classification, we J G Hansen DMSc,
Prof H T Sørensen DMSc,
further subdivided the patients by spirometry into 16 subgroups (1A to 4D). We calculated adjusted hazard ratios (HRs) M Nørgaard PhD,
for all-cause and respiratory mortality and compared the predictive ability of the three GOLD classifications (2007, 2011, R W Thomsen PhD) and
and 2017) using receiver operating curves. Department of Pulmonary
Medicine (R Dahl DMSc), Aarhus
University Hospital, Aarhus,
Findings We enrolled 33 765 patients with COPD, who were outpatients in Danish hospitals between Jan 1, 2008, and Denmark; Department of
Nov 30, 2013, in the main cohort assessed for all-cause mortality. 22 621 of these patients had data available on cause- Pulmonary Medicine, Aalborg
specific mortality (respiratory) and were included in a subcohort followed from Jan 1, 2008, to Dec 31, 2011. For the University Hospital, Aalborg,
Denmark
GOLD 2017 classification, 3 year mortality increased with increasing exacerbations and dyspnoea from group A (all-
(L-M H Wackerhausen);
cause mortality 10·0%, respiratory mortality 3·0%) to group D (all-cause mortality 36·9%, respiratory mortality 18·0%). GlaxoSmithKline, Brentford,
However, 3 year mortality was higher for group B patients (all-cause mortality 23·8%, respiratory mortality 9·7%) than UK (R Dahl); Department of
for group C patients (all-cause mortality 17·4%, respiratory mortality 6·4%). Compared with group A, adjusted HRs for Clinical Medicine, Aarhus
University, Aarhus, Denmark
all-cause mortality ranged from 2·05 (95% CI 1·87–2·26) for group B, to 1·47 (1·31–1·65) for group C, and
(R Dahl); Department of Social
to 3·01 (2·75–3·30) for group D. Area under the curve for all-cause mortality was 0·61 (95% CI 0·60–0·61) for GOLD Medicine, Institute of Public
2007, 0·61 (0·60–0·62) for GOLD 2011, and 0·63 (0·53–0·73) for GOLD 2017. Area under the curve for respiratory Health, Copenhagen
mortality was 0·64 (0·62–0·65) for GOLD 2007, 0·63 (0·62–0·64) for GOLD 2011, and 0·65 (0·53–0·78) for GOLD 2017. University, Copenhagen,
Denmark (Prof P Lange DMSc);
The GOLD 2017 classification based on ABCD groups only did not predict mortality better than the earlier 2007 and
and Medical Department,
2011 GOLD classifications. However, when 16 subgroups (1A to 4D) were defined, the new classification predicted Section of Respiratory
mortality more accurately than the previous systems (p<0·0001). Medicine, Herlev and Gentofte
University Hospital, Herlev,
Denmark (Prof P Lange)
Interpretation We showed that the new GOLD 2017 ABCD classification does not predict all-cause and respiratory
Correspondence to:
mortality more accurately than the previous GOLD systems from 2007 and 2011.
Dr Anne Gedebjerg, Department
of Clinical Epidemiology, Aarhus
Funding Danish Lung Association, Program for Clinical Research Infrastructure. University Hospital,
Aarhus DK-8200, Denmark
Introduction the well established association between COPD severity aged@clin.au.dk

The Global Burden of Disease study1 estimates that chronic
obstructive pulmonary disease (COPD) will become the
third most prevalent cause of death world­wide by 2030.
Risk stratification of patients according to COPD severity
is clinically important and forms the basis of therapeutic
recommendations.2 Since 2007, COPD has been classified
according to the Global Initiative for Chronic Obstructive
Lung Disease (GOLD) classification system. Because of
and FEV1, the first GOLD class­ification in 20073 was based
solely on patients’ FEV1 thresholds compared with
predicted normal values.2 Because lung function might not
fully describe disease complexity, the GOLD 2011 revision4
presented an ABCD classification, combining respiratory
symptoms, risk of exacerbations, and airflow limitations
as indicated by FEV1. The intention was to provide
improved understanding of the disease’s effect on

www.thelancet.com/respiratory Published online January 10, 2018 http://dx.doi.org/10.1016/S2213-2600(18)30002-X 1

 Articles
Research in context
Evidence before this study
We searched PubMed without language restrictions for
epidemiological and clinical studies published between
Jan 1, 2000, and Sept 15, 2017, using the search terms “COPD”,
“follow-up studies”, “diagnosis”, “mortality”, “prognosis”,
“survival analysis”, “severity of illness index”, “GOLD
classification”, “GOLD 2007”, “GOLD 2011”, “GOLD 2017”, and
“prediction”. Searches were supplemented by review of the
reference lists of the identified articles. Earlier Global Initiative
for Chronic Obstructive Lung Disease (GOLD) classifications
were based on either FEV1 thresholds (GOLD 2007) or a
combination of spirometry, history of exacerbations, and
symptoms (GOLD 2011). The GOLD 2017 classification is based
on a composite of spirometry and symptoms or exacerbations
and separates the spirometric 1–4 staging from the ABCD
groups defined by symptoms and exacerbations. Previous
studies have shown that the GOLD 2011 classification does not
improve mortality prediction in patients with chronic
obstructive pulmonary disease (COPD) compared with the
GOLD 2007 classification. We identified only one study
comparing mortality for all three GOLD classification systems in
our search of the literature. That study included only
524 patients with COPD and did not assess predictive
accuracies. We identified no previous large-scale studies
examining how well the new GOLD 2017 classification predicts
all-cause and respiratory mortality in patients with COPD.
Added value of this study
We prospectively enrolled a cohort of 33 765 patients with
COPD with data in the Danish registry for COPD. All patients
individual patients.4 The GOLD 2007 and 2011
classifications were designed to guide treatment, but also
have been used widely for prognostic prediction.2,5 Previous
studies5–10 have shown that the GOLD 2011 classification
does not predict mortality or respiratory outcomes better
than the GOLD 2007 classification.
The new GOLD 2017 classification is based on a com­
posite of spirometry and symptoms and exacer­bations,
separating the spirometric 1–4 staging from the ABCD
groups.2 To our knowledge, no previous studies have
examined the association between the GOLD 2017
classification and mortality or established how well the
new classification predicts mortality compared with the
earlier 2007 and 2011 GOLD classifications. Therefore, we
aimed to establish the predictive ability of the GOLD 2017
classification, com­pared with earlier classifications, for all-
cause and respiratory mortality, both when using its main
ABCD groups and when further subdividing according to
spirometric 1–4 staging.
Methods
Study population
In this nationwide cohort study, we enrolled patients with
data in the Danish registry for COPD.11 This registry was
2 www.thelancet.com/respiratory Published online January 10, 2018 http://dx.doi.org/10.1016/S2213-2600(18)30002-X
had attended hospital outpatient clinics in Denmark.
We categorised each patient according to the ABCD groups in
the GOLD 2017 classification. We further subcategorised by
spirometry 1–4 staging, resulting in 16 subgroups (1A to 4D).
After adjustment for confounders, we found that 3 year
all-cause and respiratory mortality increased from group A to
group D. All-cause and respiratory mortality were higher in
group B than in group C. Cardiovascular disease-related
mortality was high in group B and was similar to
cardiovascular disease-related mortality in group D. Receiver
operating curve analyses showed that the GOLD 2017
classification based on symptoms and exacerbation risk
(ABCD groups) only did not predict all-cause and respiratory
mortality more accurately than the older classification
systems (GOLD 2007 and 2011). When spirometry results
were added to the GOLD 2017 ABCD groups and 16 subgroups
were created, we found that 3 year mortality increased
successively from subgroup 1A (all-cause mortality 5·4%) to
subgroup 4D (all-cause mortality 42·5%) and that the
16 subgroups predicted all-cause and respiratory mortality
more accurately than either of the previous GOLD
classification systems.
Implications of all the available evidence
This cohort study is the first large-scale study to show that the
new GOLD 2017 classification does not predict all-cause and
respiratory mortality more accurately than the older GOLD
systems, unless the spirometric stages 1–4 and ABCD groups
based on symptoms and exacerbations are combined into
16 subgroups.
established in 2008 after Danish health-care authorities
began to monitor evidence-based care of patients with
COPD in all hospitals in Denmark.11 The Danish registry
for COPD draws on patient-level medical data from the
Danish National Patient Registry, which covers all
hospitals in Denmark and contains computerised
discharge records for all hospital admissions since 1977
and for all hospital outpatient visits to specialist clinics or
emergency departments since 1995.12 To be enrolled in
the Danish registry for COPD, patients have to be
outpatients in hospital-based pulmonary clinics.
Our study population consisted of incident and
prevalent patients aged 30 years or older who had visited
any hospital outpatient clinic in Denmark from
Jan 1, 2008, to Nov 30, 2013, and had received a primary
diagnosis of COPD (J44.X) or acute respiratory failure
(J96.X), in combination with COPD (J44.X) as a secondary
diagnosis (diagnosis codes from the International
Classification of Diseases tenth revision [ICD-10]).
Eligible patients also had complete data on FEV1, body-
mass index (BMI), modified Medical Research Council
(MRC) dyspnoea scale score, and smoking status. All
spirometry was done by trained nurses or laboratory
technicians at outpatient pulmonary clinics. Although no

existing data describe the overall quality of spirometry in
Denmark, all data came from specialised units with a
focus on lung function measurements, and we assume
that the quality of the data was high.
This study was approved by the Danish Data Protection
Agency (record number 2012-41-0793). According to
Danish legislation, this registry-based study did not
require further ethical approval because it did not involve
any patient contact or intervention.
GOLD classifications
We categorised each patient in our cohort according to all
three GOLD classifications (2007, 2011, and 2017) at the
first contact date with an outpatient clinic with complete
data (index date). These classifications are summarised
in the panel.
To define the risk of exacerbations as either low or
high, we used information from the Danish National
Patient Registry about COPD-related inpatient admis­
sions (ICD-10 codes J41-J44, J96, J12-J18 [pneumonia],
or J20-J22 [acute lower respiratory infections] at 1 year
before index date) and information from the Danish
National Health Service Prescription Database about oral
corticosteroids (Anatomical Therapeutic Chemical
Classification System [ATC] code H02AB) and antibiotics
(ATC code J01) used to treat exacerbations.13 A low risk of
exacerbations was defined as no hospital admissions for
an acute exa­cerbation and fewer than two prescriptions
for corticosteroids, with or without antibiotics, within
1 year before the index date. A high risk of exacerbations
was defined as one or more acute hospital admissions for
an exacerbation or two or more prescriptions for
corticosteroids, with or without antibiotics, within 1 year
before the index date.
Outcomes
The primary outcomes were all-cause and respiratory
mortality. For all-cause mortality, the exact date of death
for all patients, when applicable, was obtained from the
Danish Civil Registration System up to Nov 30, 2013.
Data on underlying cause-specific (respiratory) mortality
were available from the Danish Registry of Causes of
Death only up to Dec 31, 2011. Therefore, the analysis of
cause-specific death was restricted to a subcohort
followed from Jan 1, 2008, to Dec 31, 2011 (ICD-10 codes
for this subcohort are shown in the appendix).14,15
Covariates
Covariates were extracted from the Danish registry for
COPD and included age, gender, number of hospital
admissions for acute COPD exacerbations in the previous
year, smoking status, BMI, time from first COPD
diagnosis to index date, comorbidities, marital status (as
an indicator of socioeconomic status), and prescriptions
for selected drugs. We categorised comor­bidities according
to the Charlson Comorbidity Index within a period of
10 years before the index date.16,17 The Chronic Pulmonary
www.thelancet.com/respiratory Published online January 10, 2018 http://dx.doi.org/10.1016/S2213-2600(18)30002-X 3
Articles
Panel: Global Initiative for Chronic Obstructive Lung Disease (GOLD) classifications
GOLD 2007 classification3
• Based solely on FEV1 thresholds compared with predicted normal values, with
FEV1:FVC less than 0·70
• Patients are classified as stage 1–4:
• Stage 1 (mild)—FEV1 greater than or equal to 80% of predicted value
• Stage 2 (moderate)—FEV1 greater than or equal to 50% but less than 80% of
predicted value
• Stage 3 (severe)—FEV1 greater than or equal to 30% but less than 50% of predicted value
• Stage 4 (very severe)—FEV1 less than 30% of predicted value
GOLD 2011 classification4
• ABCD classification:
• Group A—low exacerbation risk or mild or moderate airflow limitation and score of
0–1 in the modified Medical Research Council (MRC) dyspnoea scale
• Group B—low exacerbation risk or mild or moderate airflow limitation and score
of 2 or higher in the modified MRC dyspnoea scale
• Group C—high exacerbation risk or severe or very severe airflow limitation and score
of 0–1 in the modified MRC dyspnoea scale
• Group D—high exacerbation risk or severe or very severe airflow limitation and score
of 2 or higher in the modified MRC dyspnoea scale
2017 GOLD classification2
• Separates spirometric staging (1–4) in the 2007 classification from the ABCD groups
defined by symptoms and exacerbations in the 2011 classification
• We used a classification based on a composite of spirometry and symptoms or
exacerbations, resulting in 16 subgroups (1A to 4D)
Disease category was not included in the Charlson
Comorbidity Index scoring system because it constituted
the index disease of our cohort. We obtained information
about patients’ marital status (married, divorced, widowed,
or never married) from the Civil Registration System as a
marker of socioeconomic status.18 Through the Danish
National Health Service Prescription Database, we also
obtained data on all filled prescriptions for antihypertensive
drugs, antithrombotic drugs, and statins or other lipid-
lowering drugs for each patient within 1 year before the
index date. The unique central personal registry number
given to each Danish resident at birth or on immigration
allowed individual-level linkage of Danish registry for
COPD data with data in other Danish registries.19
Statistical analyses
We estimated cumulative mortality, including 1 year and See Online for appendix
3 year mortality in accordance with the GOLD 2017
classification, using the Kaplan-Meier method for all-
cause mortality and considered death from other causes
as a competing risk for respiratory mortality.20 We then
constructed cumulative incidence curves for all-cause
mortality and respiratory mortality for groups A–D.
To assess the association between the GOLD 2017
classification and mortality, we used Cox regression to
compute crude and adjusted hazard ratios (HRs) with
95% CIs for the individual outcomes associated with the
ABCD groups in the GOLD 2017 classification, with
 Articles

We did a sub-analysis using the 16 subgroup class­

53 374 eligible patients in Danish registry ification (1A to 4D) based on a composite of spirometry
for COPD
and symptoms and exacerbations, with analyses of patient
baseline characteristics, cumulative mortality, HRs, and
19 597 had missing data on one or more variables predictive accuracy. Finally, we did a non-response
(FEV1, BMI, modified MRC dyspnoea scale
score, or smoking status)
analysis to compare patients who were excluded because
of missing exposure values with patients included in the
final study cohort. All analyses were done with SAS
33 777 with complete data version 9.4M4.

12 younger than 30 years

Role of the funding source
The funding source had no role in the study design, data
collection, data analysis, data interpretation, or writing of
33 765 in study cohort the report. AG and RWT had full access to all the study
data and had final responsibility for the decision to
submit for publication.
22 621 with data available on cause-specific
mortality (respiratory) in subcohort
Results
Figure 1: Flow diagram Of the 53 374 patients with COPD who visited a hospital
COPD=chronic obstructive pulmonary disease. BMI=body-mass index. outpatient clinic between Jan 1, 2008, and Nov 30, 2013,
MRC=Medical Research Council. 33 765 (63%) were aged 30 years or older and had
complete data on FEV1, BMI, modified MRC dyspnoea
scale score, and smoking status, and could thus be
Group A Group B Group C Group D Total
classified according to GOLD 2017 (figure 1). These
Total 5974 (17·7%) 9585 (28·4%) 4178 (12·4%) 14 028 (41·5%) 33 765 (100·0%) patients, prospectively enrolled in the Danish registry for
Stage 1 865 (2·6%) 222 (0·7%) 371 (1·1%) 203 (0·6%) 1661 (4·9%) COPD, were then followed from the index date until
Stage 2 3507 (10·4%) 2963 (8·8%) 2285 (6·8%) 3017 (8·9%) 11 772 (34·9%) death, migration, or Nov 30, 2013, whichever came first.
Stage 3 1403 (4·2%) 4139 (12·3%) 1301 (3·9%) 6056 (17·9%) 12 899 (38·2%) Table 1 shows the distribution of the patients in our
Stage 4 199 (0·6%) 2261 (6·7%) 221 (0·7%) 4752 (14·1%) 7433 (22·0%)
cohort according to the GOLD 2017 classification. Most
of the 33 765 patients with COPD were classified as
Data are n (%). The denominator is 33 765. GOLD=Global Initiative for Chronic Obstructive Lung Disease. group D (14 028 [41·5%]) and group B (9585 [28·4%]),
COPD=chronic obstructive pulmonary disease.
followed by group A (5974 [17·7%]) and group C
Table 1: Distribution of 33 765 hospital outpatients with COPD according to the GOLD 2017 classification (4178 [12·4%]). In subcategorisation, the majority of
patients in the cohort were categorised into less than half
of the subgroups.
group A as the reference group. We adjusted for age, Table 2 shows the baseline characteristics and outcomes
gender, marital status, comorbidities, BMI, smoking of all patients, presented according to the ABCD groups in
status, and number of preventive cardiovascular disease the GOLD 2017 classification. Patients with multiple
medications (antilipidaemic, antihypertensive, and anti­ symptoms and many exacerbations (group D) were older,
thrombotic therapies) on the index date. To assess the more likely to be previous smokers, less likely to be
same associations in only patients with a first-ever married and more likely to be widowed, and had more
hospital-recorded COPD diagnosis, we repeated the comorbidities than patients with fewer symptoms and
analyses after excluding all patients with a previous exacerbations (group A). Subgroup analyses showed that
hospital-based (inpatient or outpatient) COPD diagnosis patients with low FEV1 (subgroups 4A to 4D) were more
between 1977 and the index date. likely to have lower BMI and to be male than patients with
We plotted receiver operating curves for 3 years of follow- higher FEV1 (subgroups 1A to 1D; appendix).
up and estimated area under the curve to access the The entire patient cohort was followed for a median of
predictive abilities of the time-to-event models based on 2·2 years (IQR 1·1–3·6) for all-cause mortality. Data from
the GOLD 2007, 2011, and 2017 classifications.21,22 Time- the Danish Registry of Causes of Death were not available
dependent receiver operating curves and area-under-the- after Dec 31, 2011, and, therefore, the subcohort was
curve functions summarise the predictive accuracy at followed for a median of 1·6 years (0·8–2·7) for
specific times. To assess the calibration of the GOLD 2007, respiratory mortality. All-cause and respiratory mortality
2011, and 2017 classifications in a time-to-event setting, increased with increasing exacerbations and dyspnoea
we used a modification of the D’Agostino-Nam test (table 3, figure 2). Thus, 1 year and 3 year all-cause and
(appendix).23 We compared the three classifications using respiratory cumulative mortality were lowest for group A
Uno’s concordance statistics; a two-sided p value of less and highest for group D (table 3). The cumulative
than 0·05 was considered significant.24 incidence of all-cause and respiratory mortality was

4 www.thelancet.com/respiratory Published online January 10, 2018 http://dx.doi.org/10.1016/S2213-2600(18)30002-X

 Articles

generally higher in groups B and D (many symptoms)

Group A Group B Group C Group D
than in groups A and C (few symptoms; table 3, figure 2). (n=5974) (n=9585) (n=4178) (n=14 028)
Generally, the subclassification into 16 subgroups
Baseline characteristics
reflected all-cause and respiratory cumulative mortality
First-time COPD diagnosis
well, with cumulative mortality increasing from sub­
No 1318 (22·1%) 4129 (43·1%) 2819 (67·5%) 11 508 (82·0%)
groups 1A to 4D (table 3; appendix).
Yes 4656 (77·9%) 5456 (56·9%) 1359 (32·5%) 2520 (18·0%)
Patients in group B had fairly high all-cause mortality
Time with COPD diagnosis 5·2 (4·4) 6·1 (4·2) 3·0 (4·3) 4·7 (4·7)
(table 3). When comparing cause-specific mortality in before index date (years)
group B across different spirometry-defined groups (1–4), Gender
patients in subgroup 2B had lower mortality from
Female 2790 (46·7%) 4833 (50·4%) 2257 (54·0%) 7845 (55·9%)
respiratory disease than from cardiovascular disease
Male 3184 (53·3%) 4752 (49·6%) 1921 (46·0%) 6183 (44·1%)
(2·4% vs 3·2%; appendix). By contrast, mortality from
Age (years) 65·4 (10·9) 69·5 (10·5) 68·7 (10·8) 71·9 (10·0)
respiratory disease was higher than from cardiovascular
Body-mass index (kg/m2) 25·7 (5·3) 26·2 (6·5) 25·3 (5·6) 25·5 (6·5)
disease in subgroups 3B (6·6% vs 2·6%) and 4B (11·6%
FEV1 (%) 62 (49–73) 43 (31–56) 57 (45–69) 38 (27–49)
vs 2·7%; appendix). Although patients in subgroup 2D
Modified MRC dyspnoea scale
had the highest cardiovascular disease-related
0 1525 (25·5%) 0 (0·0%) 829 (19·8%) 0 (0·0%)
mortality (3·7%) of all subgroups, it was similar to
1 4449 (74·5%) 0 (0·0%) 3349 (80·2%) 0 (0·0%)
cardiovascular disease-related mortality among patients
2 0 (0·0%) 4621 (48·2%) 0 (0·0%) 4909 (35·0%)
in subgroups 2B (3·2%), 1D (3·4%), and 3D (3·4%).
3 0 (0·0%) 3019 (31·5%) 0 (0·0%) 4551 (32·4%)
Patients in subgroup 4D—who had low FEV1 and high
4 0 (0·0%) 1945 (20·3%) 0 (0·0%) 4568 (32·6%)
risk of exacerbations and symptoms—had the poorest
prognosis with all-cause mortality of 44%. Smoking status

Calculation of adjusted HRs with group A as the Never 292 (4·9%) 237 (2·5%) 137 (3·3%) 345 (2·5%)

reference group showed that group D had the highest Current 2388 (40·0%) 2911 (30·4%) 1556 (37·2%) 3711 (26·5%)
mortality: the adjusted HR for group D was 3·01 (95% CI Previous 3294 (55·1%) 6437 (67·2%) 2485 (59·5%) 9972 (71·1%)
2·75–3·30) for all-cause mortality and 5·11 (3·99–6·55) for Marital status
respiratory mortality (table 4). Adjusted HRs for all-cause Currently married 3242 (54·3%) 4542 (47·4%) 2041 (48·9%) 6381 (45·5%)
and respiratory mortality were higher in group B than in Never married 650 (10·9%) 953 (9·9%) 375 (9·0%) 1061 (7·6%)
group C (table 4). In subgroup analyses, adjusted HRs Divorced 1190 (19·9%) 2055 (21·4%) 836 (20·0%) 2719 (19·4%)
increased with decreasing FEV1 (from GOLD stages 1 to 4) Widowed 886 (14·8%) 2025 (21·1%) 924 (22·1%) 3857 (27·5%)
and from groups A to D, with lowest mortality observed in Number of acute hospital admissions for COPD exacerbations during the previous year
subgroups 1A (reference) and 2A (all-cause mortality: 0 5974 (100·0%) 9585 (100·0%) 688 (16·5%) 2773 (19·8%)
adjusted HR 1·26, 95% CI 0·92–1·74; respiratory 1 0 (0·0%) 0 (0·0%) 2449 (58·6%) 6412 (45·7%)
mortality: 0·78, 0·32–1·91) and highest mortality observed ≥2 0 (0·0%) 0 (0·0%) 1041 (24·9%) 4843 (34·5%)
in subgroup 4D (all-cause mor­ tality: 5·90, 4·37–7·96; Charlson Comorbidity Index score
respiratory mortality: 8·54, 3·82–19·12; appendix). 0 3538 (59·2%) 4385 (45·7%) 2044 (48·9%) 5576 (39·7%)
Figure 3 shows receiver operating curves for all-cause 1 1095 (18·3%) 2050 (21·4%) 882 (21·1%) 3068 (21·9%)
and respiratory mortality at 3 year follow-up for the 2 774 (13·0%) 1521 (15·9%) 629 (15·1%) 2396 (17·1%)
three GOLD classifications. Changes in the area under the ≥3 567 (9·5%) 1629 (17·0%) 623 (14·9%) 2988 (21·3%)
curve over 3 year follow-up for the GOLD 2007, 2011,
Outcomes
and 2017 classifications are shown in the appendix. For all-
All-cause mortality 534 (8·9%) 2225 (23·2%) 648 (15·5%) 4958 (35·3%)
cause and respiratory mortality, all three GOLD
Respiratory mortality* 67 (1·8%) 429 (6·6%) 98 (3·7%) 1277 (13·2%)
classifications were well calibrated (appendix). For all three
Cardiovascular disease-related 39 (1·0%) 181 (2·8%) 51 (1·9%) 297 (3·1%)
GOLD classifications, the time-dependent area under the mortality*
curve for all-cause and respiratory mortality was stable over
follow-up (appendix). The area under the receiver operating Data are n (%), mean (SD), or median (IQR). COPD=chronic obstructive pulmonary disease. MRC=Medical Research
Council. GOLD=Global Initiative for Chronic Obstructive Lung Disease. *Subcohort of 22 621 patients with data
curve for all-cause mortality was 0·61 (95% CI 0·60–0·61) available for 2008–11 (n=3745 in group A, n=6529 in group B, n=2672 in group C, n=9675 in group D). †ABCD groups
for GOLD 2007, 0·61 (0·60–0·62) for GOLD 2011, and 0·63 based on symptoms and exacerbation risk only.
(0·53–0·73) for GOLD 2017. The area under the receiver
Table 2: Baseline characteristics and incidence of outcomes among 33 765 hospital clinic outpatients
operating curve for respiratory mortality was 0·64 with COPD according to the GOLD 2017 classification†
(0·62–0·65) for GOLD 2007, 0·63 (0·62–0·64) for
GOLD 2011, and 0·65 (0·53–0·78) for GOLD 2017. Thus,
predictive accuracy did not differ significantly between the the older classifications (0·65 for all-cause mortality and
three GOLD classification systems (appendix). 0·69–0·70 for respiratory mortality; p<0·0001; appendix).
When using the 16 subgroup 1A to 4D classification of Time-dependent area under the curve for all-cause
GOLD 2017, the time-dependent area under the curve for mortality stratified by gender is shown in the appendix.
all-cause and respiratory mortality was larger than with We found no difference in the predictive ability of the

www.thelancet.com/respiratory Published online January 10, 2018 http://dx.doi.org/10.1016/S2213-2600(18)30002-X 5

 Articles

2009 (21%); exclusions were then stable at around

All-cause Respiratory
cumulative mortality cumulative mortality* 13% per year from 2010 to 2013, suggesting that the main
reason for missing exposure variables was late imple­
1 year 3 year 1 year 3 year
mentation of data registration in some hospitals during
Group A 2·3% 10·0% 0·7% 3·0% the early years of operation of the Danish registry for
Subgroup 1A 0·8% 5·4% 0·5% 1·9% COPD.
Subgroup 2A 2·1% 8·8% 0·6% 2·0%
Subgroup 3A 3·1% 14·0% 1·1% 4·9% Discussion
Subgroup 4A 5·0% 17·8% 1·7% 8·7% To the best of our knowledge, this cohort study is the first
Group B 6·6% 23·8% 2·3% 9·7% large-scale, nationwide study to examine the association
Subgroup 1B 5·0% 17·1% 1·9% 5·2% between the GOLD 2017 classification and all-cause
Subgroup 2B 5·0% 17·2% 0·9% 3·7% and respiratory mortality in patients with COPD, and
Subgroup 3B 7·4% 24·9% 2·5% 9·3% the ability of this new classification to predict these
Subgroup 4B 7·6% 29·8% 3·6% 16·1% outcomes. We found that the GOLD 2017 ABCD
Group C 5·2% 17·4% 1·7% 6·4% classification (based on symptoms or exacerbations alone)
Subgroup 1C 3·8% 11·4% 1·4% 3·0% was not able to better predict all-cause or respiratory
Subgroup 2C 4·3% 14·8% 0·9% 4·7% mortality than either of the GOLD 2007 and GOLD 2011
Subgroup 3C 6·3% 21·5% 2·2% 8·1% classifications. Additionally, we found that when using
Subgroup 4C 10·7% 29·5% 7·3% 15·4% a composite of spirometry and symptoms or exacerba­
Group D 11·6% 36·9% 5·7% 18·0% tions (ie, a 16 subgroup 1A to 4D classification), the
Subgroup 1D 11·6% 35·4% 4·1% 10·1% GOLD 2017 classification was a better predictor of all-
Subgroup 2D 10·0% 30·9% 3·8% 10·2% cause and respiratory mortality than the older GOLD
Subgroup 3D 11·4% 35·0% 5·4% 15·9% classifications. Given the high prevalence of COPD, even
Subgroup 4D 12·8% 42·5% 7·1% 23·6% minor improvements in prediction might be of public
health importance. Although the GOLD 2017 classification
COPD=chronic obstructive pulmonary disease. GOLD=Global Initiative for Chronic
also predicted cardiovascular disease-related mortality, it
Obstructive Lung Disease. *Subcohort of 22 621 patients with COPD, considering
cause-specific death as a competing risk, 2008–11. had poor discriminatory power for this outcome, with an
area under the curve of around 0·57, and was not better
Table 3: Cumulative mortality risk among 33 765 patients according to than the GOLD 2007 and 2011 classifications. Our
severity of COPD based on the GOLD 2017 classification
findings for the GOLD 2007 and 2011 classifications were
consistent with previous studies5,9 of their performance in
16 subgroup classification between men and women predicting all-cause, respiratory, and cardiovascular
based on overlapping 95% CIs. The results for respiratory disease-related mortality. However, our observation that
mortality by gender are not shown because the model was the GOLD 2017 classification was not a better predictor of
not well calibrated. all-cause and respiratory mortality in men than in women
Data on cardiovascular disease-related mortality is differed from the findings of a large observational study5
available in the appendix. Cardiovascular disease-related of pooled data for 15 632 patients from 22 COPD cohorts.
mortality was higher in patients in groups B and D than in Our finding of increasing all-cause and respiratory
patients in groups A and C, even after adjustment for mortality between GOLD stage 1 and GOLD stage 4, and
known cardiovascular risk factors. Cardiovascular disease- from group A to D, substantiates the importance of FEV1
related mortality in group B was equal to cardio­vascular as a predictor of poor survival. The population-based
disease-related mortality in group D. The time-dependent PLATINO study25 of 524 patients with mainly mild-to-
area under the curve for cardiovascular disease-related moderate COPD showed no clear pattern with regard to
mortality was low and almost equal in the three GOLD mortality according to the GOLD 2017 classification.
classifi­cations at around 0·57. Subgroup analyses showed However, because only patients with clinical COPD
that cumulative cardiovascular disease-related mortality attending hospital-based pulmonary clinics can be enrolled
leveled off or de­creased between GOLD stages 3 and 4, in the Danish registry for COPD, patients in our cohort
within groups A to D. generally had more severe COPD than patients in the
The results of sensitivity analyses restricted to patients PLATINO cohort. Therefore, most exacerbations in
with a first-ever COPD diagnosis with recorded hospital patients in group D were related to previous inpatient
contacts were consistent with those of the main analysis admissions for COPD (84% of all patients) and not to
(appendix). We detected no major difference between treatment in primary care with corticosteroids or
included and excluded patients in our non-response antibiotics (45% of all patients). This fact also explains the
analyses (appendix). The excluded patients had a slightly low proportion of patients in the GOLD stage 1 category,
lower exacerbation risk and were more recently diag­ because most patients who might be categorised as stage 1
nosed with COPD than the included patients. The largest in Denmark are either undiagnosed or treated solely by
proportion of exclusions occurred in 2008 (29%) and general practitioners (GPs) and thus not enrolled in the

6 www.thelancet.com/respiratory Published online January 10, 2018 http://dx.doi.org/10.1016/S2213-2600(18)30002-X

 Articles

Danish registry for COPD.26,27 We observed that, according A

to the ABCD groups of the GOLD 2017 classification, all- 40 Group A
cause, respiratory, and cardiovascular disease-related Group B
Group C
mortality were higher for patients in group B than for Group D
patients in group C, and that cardiovascular disease-related 30

All-cause mortality (%)

mortality was equal in groups B and D. This finding is in
line with previous reports that patients in group B are at 20
particularly high risk and is probably associated with the
fact that dyspnoea can be caused by diseases other than
COPD, such as heart disease.6 By contrast, in our 10
subanalysis, cardiovascular disease-related cumulative
mortality leveled off or decreased between GOLD stages 3
0
and 4, within groups A to D. This trend is in line with the 0 1 2 3
Number at risk
hypothesis that the main cause of death in patients with Group A 5974 4739 3430 2267
moderate-to-severe COPD is cardiovascular disease, Group B 9585 7559 5500 3680
Group C 4178 3251 2314 1469
whereas in patients with very severe airflow limitation, Group D 14 028 10 502 7179 4505
respiratory failure caused by COPD is more common.
The main strengths of this study of prospectively B
40
collected data are the comprehensive and detailed assess­
ments of lifestyle, demographic factors, and clinical
factors available in the Danish registry for COPD and 30
Respiratory mortality (%)

their linkage with population-based health registries.

These resources provided nearly 100% complete data for
demographic and clinical characteristics and prescription 20

data for the patients in our study. The likelihood of

misdiagnosis was low because all patients were followed 10
in hospital-based pulmonary outpatient clinics and
because the presence of chronic airflow limitation was
mandatory for enrolment in this cohort. 0
0 1 2 3
Our study also has several limitations. First, although we Follow-up (years)
were able to include more than 30 000 patients with COPD Number at risk
in our final cohort, a considerable number of patients in Group A 3745 2540 1469 513
Group B 6529 4601 2882 961
the Danish registry for COPD were excluded because of Group C 2672 1766 904 301
missing records of exposure variables. However, missing Group D 9675 6457 3585 1229
exposure data primarily resulted from late initiation of
Figure 2: All-cause (A) and respiratory (B) mortality over 3 years according to GOLD 2017 ABCD group
data registration with the Danish registry for COPD in
GOLD=Global Initiative for Chronic Obstructive Lung Disease.
some hospitals (ie, they were missing at random) and was
not related to intrinsic patient char­ acteristics.28 This
assumption is consistent with the findings of our non-
All-cause mortality Respiratory mortality*
response analysis, and missing data are therefore unlikely
to affect the validity of our findings. Second, cause-of-death Crude HRs Adjusted HRs† Crude HRs Adjusted HRs†
(95% CI) (95% CI) (95% CI) (95% CI)
registries might have fairly low validity, and COPD might
be under-reported as a cause of death in mild-to-moderate Group A 1 (ref) 1 (ref) 1 (ref) 1 (ref)

cases compared with severe cases.29 Third, some Group B 2·58 (2·35–2·83) 2·05 (1·87–2·26) 3·44 (2·66–4·45) 2·74 (2·11–3·54)
exacerbations might have been treated with antibiotics Group C 1·82 (1·62–2·04) 1·47 (1·31–1·65) 2·19 (1·61–2·99) 1·78 (1·31–2·43)
alone without corticosteroids, leading to underestimation Group D 4·40 (4·03–4·81) 3·01 (2·75–3·30) 7·63 (5·97–9·75) 5·11 (3·99–6·55)
of the incidence of exacer­bations.30 Fourth, symptoms in HRs=hazard ratios. GOLD=Global Initiative for Chronic Obstructive Lung Disease. COPD=chronic obstructive pulmonary
the registry are only assessed with the modified MRC disease. *Subcohort of 22 621 patients with COPD who had cause-specific mortality data available for 2008–11.
dyspnoea scale and not with the COPD assessment test. †Adjusted for age, gender, marital status, comorbidity, body-mass index, smoking status, and number of
antihypertensive drugs, antithrombotic drugs, statins, and other lipid-lowering drugs.
Fifth, we relied on real-life clinical data, with routine
spirometry results obtained from different hospital clinics; Table 4: Risk of all-cause mortality and respiratory mortality among 33 765 patients with COPD,
therefore, our measurements might have been of lower according to the GOLD 2017 A–D groups
quality than could have been obtained in a prospective
analysis. Although we adjusted for different potential
confounder variables, our results might have been affected enrolment to the Danish registry for COPD was done in
by residual confounding variables, such as socioeconomic specialty outpatient clinics and patients constituted a
factors other than marital status. Finally, because mixture of incident and prevalent COPD cases, particularly

www.thelancet.com/respiratory Published online January 10, 2018 http://dx.doi.org/10.1016/S2213-2600(18)30002-X 7

 Articles

in the first few years after initiation of the registry, our

A
1·00 GOLD 2007 AUC 0·61 (95% CI 0·60–0·61)
patients probably had more severe COPD than newly
GOLD 2011 AUC 0·61 (95% CI 0·60–0·62) diagnosed patients or patients followed by GPs.
GOLD 2017 AUC 0·63 (95% CI 0·53–0·73) In conclusion, this cohort study of patients with COPD
followed in Danish hospital clinics showed that the new
GOLD 2017 ABCD classification did not predict all-cause
and respiratory mortality better than the GOLD 2007 and
0·75 GOLD 2011 classifications. However, we found that com­
bining the spirometric 1–4 staging with the ABCD
grouping (based on symptoms or exacerbations) im­proved
the predictive accuracy of GOLD 2017. However, none of
the GOLD classifications appeared to have sufficient
discriminatory power to be used as a stand alone tool for
Sensitivity

0·50 risk classification of mortality in patients with COPD.

Contributors
AG, L-MHW, RD, JGH, HTS, MN, PL, and RWT contributed to the study
concept and design. AG, L-MHW, RD, JGH, HTS, MN, PL, and RWT
collected or interpreted the data. AG wrote the first draft of the manuscript,
with help from MN, PL, HTS, and RWT. All authors critically reviewed the
manuscript and approved the final version for submission. SKS and EH-P
0·25 did the statistical analyses. RWT obtained funding. MN, PL, and RWT
supervised the study.
Declaration of interests
RD works at GlaxoSmithKline. All other authors declare no competing
interests.
Acknowledgments
0 This study was funded by the Danish Lung Association and the Program
for Clinical Research Infrastructure.
B References
1·00 GOLD 2007 AUC 0·64 (95% CI 0·62–0·65) 1 GBD 2015 Chronic Respiratory Disease Collaborators. Global,
GOLD 2011 AUC 0·63 (95% CI 0·62–0·64) regional, and national deaths, prevalence, disability-adjusted life
GOLD 2017 AUC 0·65 (95% CI 0·53–0·78) years, and years lived with disability for chronic obstructive
pulmonary disease and asthma, 1990–2015: a systematic analysis for
the Global Burden of Disease Study 2015. Lancet Respir Med 2017;
5: 691–706.
2 Vogelmeier CF, Criner GJ, Martinez FJ, et al. Global strategy for the
diagnosis, management, and prevention of chronic obstructive lung
0·75 disease 2017 report. GOLD executive summary.
Am J Respir Crit Care Med 2017; 195: 557–82.
3 Rabe KF, Hurd S, Anzueto A, et al. Global strategy for the
diagnosis, management, and prevention of chronic obstructive
pulmonary disease: GOLD executive summary.
Am J Respir Crit Care Med 2007; 176: 532–55.
4 Vestbo J, Hurd SS, Augsti AG, et al. Global strategy for the
Sensitivity

diagnosis, management, and prevention of chronic obstructive

0·50
pulmonary disease: GOLD executive summary. Am J Respir Crit
Care Med 2013; 187: 347–65.
5 Soriano JB, Lamprecht B, Ramirez AS, et al. Mortality prediction in
chronic obstructive pulmonary disease comparing the GOLD 2007
and 2011 staging systems: a pooled analysis of individual patient
data. Lancet Respir Med 2015; 3: 443–50.
6 Lange P, Marott JL, Vestbo J, et al. Prediction of the clinical course
0·25 of chronic obstructive pulmonary disease, using the new GOLD
classification: a study of the general population.
Am J Respir Crit Care Med 2012; 186: 975–81.
7 Leivseth L, Brumpton BM, Nilsen TI, Mai XM, Johnsen R,
Langhammer A. GOLD classifications and mortality in chronic
obstructive pulmonary disease: the HUNT study, Norway. Thorax
2013; 68: 914–21.
8 Agusti A, Edwards LD, Celli B, et al. Characteristics, stability and
0 outcomes of the 2011 GOLD COPD groups in the ECLIPSE cohort.
0 0·25 0·50 0·75 1·00 Eur Respir J 2013; 42: 636–46.
1-specificity 9 Johannessen A, Nilsen RM, Storebo M, Gulsvik A, Eagan T,
Bakke P. Comparison of 2011 and 2007 global initiative for chronic
Figure 3: Receiver operating curves for all-cause (A) and respiratory (B) mortality at 3 year follow-up obstructive lung disease guidelines for predicting mortality and
GOLD=Global Initiative for Chronic Obstructive Lung Disease. AUC=area under the curve. hospitalization. Am J Respir Crit Care Med 2013; 188: 51–59.

8 www.thelancet.com/respiratory Published online January 10, 2018 http://dx.doi.org/10.1016/S2213-2600(18)30002-X


10 Goossens LM, Leimer I, Metzdorf N, Becker K, Rutten-van Molken
MP. Does the 2013 GOLD classification improve the ability to
predict lung function decline, exacerbations and mortality:
a post-hoc analysis of the 4-year UPLIFT trial. BMC Pulm Med 2014;
14: 163.
11 Lange P, Tøttenborg SS, Sorknæs AD, et al. Danish Register of
Chronic Obstructive Pulmonary Disease. Clin Epidemiol 2016;
8: 673–78.
12 Lynge E, Sandegaard JL, Rebolj M. The Danish National Patient
Register. Scand J Public Health 2011; 39: 30–33.
13 Johannesdottir SA, Horvath-Puho E, Ehrenstein V, Schmidt M,
Pedersen L, Sorensen HT. Existing data sources for clinical
epidemiology: the Danish National Database of Reimbursed
Prescriptions. Clin Epidemiol 2012; 4: 303–13.
14 Pedersen CB. The Danish Civil Registration System.
Scand J Public Health 2011; 39: 22–25.
15 Juel K, Helweg-Larsen K. The Danish registers of causes of death.
Dan Med Bull 1999; 46: 354–57.
16 Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of
classifying prognostic comorbidity in longitudinal studies:
development and validation. J Chronic Dis 1987; 40: 373–83.
17 Thygesen SK, Christiansen CF, Christensen S, Lash TL,
Sorensen HT. The predictive value of ICD-10 diagnostic coding
used to assess Charlson Comorbidity Index conditions in the
population-based Danish National Registry of Patients.
BMC Med Res Methodol 2011; 11: 83.
18 Lange P, Marott JL, Vestbo J, Ingebrigtsen TS, Nordestgaard BG.
Socioeconomic status and prognosis of COPD in Denmark. COPD
2014; 11: 431–37.
19 Schmidt M, Pedersen L, Sorensen HT. The Danish Civil
Registration System as a tool in epidemiology. Eur J Epidemiol 2014;
29: 541–49.
20 Gooley TA, Leisenring W, Crowley J, Storer BE. Estimation of
failure probabilities in the presence of competing risks: new
representations of old estimators. Stat Med 1999; 18: 695–706.
www.thelancet.com/respiratory Published online January 10, 2018 http://dx.doi.org/10.1016/S2213-2600(18)30002-X 9
Articles
21 Kamarudin AN, Cox T, Kolamunnage-Dona R. Time-dependent
ROC curve analysis in medical research: current methods and
applications. BMC Med Res Methodol 2017; 17: 53.
22 Uno H, Cai T, Tian L, Wei LJ. Evaluating prediction rules for t-year
survivors with censored regression models. J Am Stat Assoc 2007;
102: 527–37.
23 Demler OV, Paynter NP, Cook NR. Tests of calibration and
goodness-of-fit in the survival setting. Stat Med 2015; 34: 1659–80.
24 Uno H, Cai T, Pencina MJ, D’Agostino RB, Wei LJ. On the
C-statistics for evaluating overall adequacy of risk prediction
procedures with censored survival data. Stat Med 2011; 30: 1105–17.
25 Menezes AM, Wehrmeister FC, Perez-Padilla R, et al. The
PLATINO study: description of the distribution, stability, and
mortality according to the Global Initiative for Chronic Obstructive
Lung Disease classification from 2007 to 2017.
Int J Chron Obstruct Pulmon Dis 2017; 12: 1491–1501.
26 Çolak Y, Afzal S, Nordestgaard BG, Vestbo J, Lange P. Prognosis of
asymptomatic and symptomatic, undiagnosed COPD in the general
population in Denmark: a prospective cohort study.
Lancet Respir Med 2017; 5: 426–34.
27 Lange P, Marott JL, Dahl M, Ingebrigtsen TS, Vestbo J,
Nordestgaard BG. Substantial need for early diagnosis,
rehabilitation and treatment of chronic obstructive pulmonary
disease. Dan Med J 2012; 59: A4396.
28 Tottenborg SS, Thomsen RW, Nielsen H, Johnsen SP,
Frausing Hansen E, Lange P. Improving quality of care among
COPD outpatients in Denmark 2008–2011. Clin Respir J 2013;
7: 319–27.
29 Jensen HH, Godtfredsen NS, Lange P, Vestbo J. Potential
misclassification of causes of death from COPD. Eur Respir J 2006;
28: 781–85.
30 Salwan AA, Spigt M, Laue J, Melbye H. Predictors of treatment with
antibiotics and systemic corticosteroids for acute exacerbations of
asthma and chronic obstructive pulmonary disease in primary care.
BMC Fam Pract 2015; 16: 40.