UvA-DARE is a service provided by the library of the University of Amsterdam (http://dare.uva.nl)
UvA-DARE (Digital Academic Repository)
Breathomics in pulmonary disease
Fens, N.
Link to publication
Citation for published version (APA):Fens, N. (2011). Breathomics in pulmonary disease.
General rightsIt is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s),other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).
Disclaimer/Complaints regulationsIf you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, statingyour reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Askthe Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam,The Netherlands. You will be contacted as soon as possible.
Download date: 17 May 2020
Subphenotyping COPD using breath profiles
COPD subphenotypes by combining pulmonary function, CT imaging and breathomics in cluster analysis. A population-based survey
Niki Fens, Annelot G.J. van Rossum, Pieter Zanen, Bram van Ginneken,
Rob J. van Klaveren, Aeilko H. Zwinderman, Peter J. Sterk
Submitted for publication
7
100 Chapter 7
ABSTRACTRationale. Classification of COPD is currently based on the presence and severity of airways
obstruction. However, this may not fully reflect the phenotypic heterogeneity of COPD in the
(ex-) smoking community. We hypothesized that cluster analysis of functional, clinical, radio-
logical and exhaled breath metabolomic features identifies clinical subphenotypes of COPD in
a community-based population of heavy (ex-) smokers.
Methods. Adults between 50-75 years with a smoking history of at least 15 packyears derived
from a random population-based survey as part of the NELSON study underwent detailed
assessment of pulmonary function, chest CT scanning, questionnaires and exhaled breath
molecular profiling using an electronic nose. Cluster analysis was performed on the subgroup
of subjects fulfilling the GOLD criteria for COPD (post-BD FEV1/FVC<0.70).
Results. 300 subjects were recruited, of which 157 fulfilled the criteria for COPD and were
included in the cluster analysis. Three clusters were identified: cluster 1 (n=19; 12%): severe
airways obstruction, low quality of life, distinct breath molecular profile; cluster 2 (n=101;
64%): mild COPD with high prevalence of cardiovascular comorbidity, distinct breath molecu-
lar profile; cluster 3 (n=37; 24%): chronic bronchitis, high percentage of current smokers. By
using lung density by CT, cluster 2 could be subdivided into emphysematous (n=51) and non-
emphysematous COPD (n=50). Cross-validation analysis showed an accuracy of 94% for the
3 main clusters and 85% for all 4 clusters (both p<0.001).
Conclusions. This unbiased taxonomy for COPD reinforces clusters found in previous studies
and allows better phenotyping of COPD.
Subphenotyping COPD using breath profiles 101
Chap
ter 7
INTRODUCTIONChronic obstructive pulmonary disease (COPD) is a major cause of mortality and disability, and
is defined by the Global Initiative for Chronic Obstructive Lung Disease (GOLD) as the presence
of not fully reversible airflow obstruction, confirmed by postbronchodilator spirometry (ratio of
forced expiratory volume in 1 second to forced vital capacity: FEV1/FVC <0.70) [1]. However, it is
increasingly recognized that COPD is a complex and heterogeneous disease that cannot be de-
scribed adequately by the severity of airflow limitation alone [2]. Patients vary widely in clinical
presentation, response to therapy, decline in lung function, CT imaging and exacerbation fre-
quency. By identification and prospective evaluation of subphenotypes of COPD, the response
to therapies and prognosis may be better predicted. Besides this, differential pathophysiologi-
cal mechanisms may thus be identified thereby allowing better targeted COPD studies [2].
It has been proposed that a new taxonomy for defining COPD is required to recognize such
subphenotypes [3-5]. Using unbiased statistical methods such as unsupervised cluster analy-
sis, individuals are grouped based on their similarities and differences in multi-scale data [5,6].
This is a way to describe patterns of the disease based on clinical, functional, and pathogenetic
features.
Several studies have adopted this approach in order to identify meaningful subphenotypes
of obstructive lung diseases [3,7-11]. However, validation and generalisibility of these clusters
has been limited by the lack of postbronchodilator spirometry and limited availability of CT
scans to assess the presence of emphysema. Therefore, a comprehensive phenotypic character-
ization of COPD patients should be based on patient-related outcomes such as health-related
quality of life as well as functional, clinical, radiological and inflammatory features.
In a recent study we showed that metabolomic profiles of volatile organic compounds
(VOCs) in exhaled air reflect different patterns of inflammatory metabolism and oxidative stress,
being associated with eosinophilic and neutrophilic airway inflammation in mild and moderate
COPD [12]. Combining exhaled air metabolomics with functional, clinical and radiological mea-
surements may therefore identify relevant molecular subphenotypes of COPD.
The identification and clinical usefulness of clusters is highly dependent of the population
under investigation [6]. In the community or primary care centers, many individuals with COPD
are underdiagnosed [13] and severe cases are scarce, in contrast to the homogeneous patient
populations in referral centers. A population-based survey may overcome the selection of a cer-
tain subset of patients and may well reflect the primary care population of COPD patients [6].
In this study we hypothesized that unsupervised cluster analysis in a community-based
population of COPD patients reveals subphenotypes of COPD using lung function parameters,
CT scanning, symptoms and other clinical parameters, combined with exhaled breath metabo-
lomics. To that end a survey on a population of heavy (ex-) smokers was undertaken. Cluster
analysis was performed on parameters from those subjects in this survey fulfilling the current
diagnostic criteria for COPD [1]. The validity of the clusters was then tested in a cross-validation
analysis.
102 Chapter 7
METHODSSubjectsSubjects in this study were a random selection at a single location (Utrecht, The Netherlands) of
the NELSON project and were recruited between February and November 2008. The NELSON
project is a population-based Dutch-Belgian multicenter lung cancer screening trial of heavy
smokers and ex-smokers [14]. Participants for the NELSON study were recruted by sending a
questionnaire about smoking history and health-related items to all citizens between the ages
of 50 and 75 years living in the area of the participating centers. Of the respondents, subjects
meeting the inclusion criterion of a smoking history of at least 15 packyears were invited for
participation in the study. COPD was defined according to the GOLD criteria (postbronchodila-
tor FEV1/FVC ratio < 0.70) [1]. Chronic bronchitis was defined as cough and sputum production
on most days for at least 3 months per year in at least 2 consecutive years [15].
The trial was approved by the Dutch Ministry of Health and by the local ethics committees
of participating centers. All subjects gave their written informed consent. This COPD substudy
of the NELSON study was registered at the Netherlands Trial Register, under NTR 1285.
DesignThe study had a cross-sectional design and participants paid a single visit to the hospital. Chest
CT scanning, exhaled breath sampling, pre- and postbronchodilator spirometry and diffusion
capacity were performed and questionnaires were taken.
MeasurementsQuestionnairesSubjects completed a validated questionnaire of the European Community Respiratory Health
Survey (ECRHS) concerning sociodemographic data, respiratory symptoms, comorbidity,
treatments and previous diagnoses, use of medications and smoking history [16]. Subjects
also completed the clinical COPD questionnaire (CCQ) which measures the disease-related
health status [17].
Lung function testingSpirometry was performed before and 10 minutes after inhalation of 400 μg of salbutamol via
a spacer according to the latest ERS recommendations using daily calibrated equipment [18].
Diffusion capacity for carbon monoxide corrected for alveolar volume (DL,CO/VA) was measured
according to the recommendations using the single breath method and an inhalation mixture
of 0.3% CO and 10% He with air [19].
Chest CT scanningCT scanning was performed using a multidetector-row CT scanner (Mx8000 IDT or Brilliance
16P, Philips Medical Systems, Cleveland, OH, USA). Scans were performed in end-inspiration
Subphenotyping COPD using breath profiles 103
Chap
ter 7
without contrast. Scanning time was within 12 s, in spiral mode with 16x0.75 mm collimation,
1.0 mm reconstruction thickness, 0.7 mm increment, voltage of 120 kVp for subjects ≤ 80 kg
or 140 kVp for subjects > 80 kg (30 mAs). Scans were analyzed using ImageXplorer (iX) (Image
Sciences Institute, Utrecht, The Netherlands) [20]. The extent of emphysema was estimated by
quantifying the percentage of the total lung voxels below 950 and 910 Hounsfield units (HU)
[20]. In addition, the 15th percentile point (perc15) was measured [21]. The perc15 is defined as
the threshold value in Hounsfield units (HU) for which 15% of all lung voxels has a lower value.
Exhaled breath collectionCollection of exhaled breath was done as previously described [12,22,23]. In short, patients
breathed normally for 5 min with the nose clipped through an inspiratory VOC-filter (A2, North
Safety, Middelburg, The Netherlands), connected to a three-way non re-breathing valve and ex-
haled through a silica reservoir. Then, patients took a single deep maximal inspiration and ex-
haled one vital capacity volume into a 10 L Tedlar bag (SKC Inc., Eighty Four, PA, USA) connected
to the expiratory port and the silica reservoir. In parallel, a Tedlar bag was filled with VOC-filtered
room air for comparison.
Breath analysis by electronic noseWithin 5 minutes after breath was collected the Cyranose 320 electronic nose (Smiths Detec-
tion, Pasadena, Ca, USA) was connected to the Tedlar bag, followed by 1 minute (100 ml) sam-
pling of the exhaled air. This was done in parallel with sampling a Tedlar bag filled with VOC-
filtered room air for comparison. Raw eNose data consists of changes in electrical resistance of
each of the 32 polymer sensors [24,25] and was used for further analysis with offline pattern-
recognition software.
Statistical analysisSPSS (version 18.0) was used for data analysis. eNose raw data (change in resistance of sensors)
were restructured by principal component analysis from the original 32 sensors to 4 principal
components (PC) capturing 95.2% of the variance within the dataset.
Missing values made up a small proportion (< 4%) and were considered to be at random
[26,27]. Multiple imputation by the method of Rubin [28] was used for the generation of miss-
ing data replacements based on values drawn from the distribution posited by the prediction
model, taking into account the relationships between the incomplete variables and all other
variables [26,27].
Cluster analysisCluster analysis is an unbiased way to assign subjects to groups with similar features without
a priori assumptions for classification. Clusters are constructed in such a way that individuals
within in a cluster are highly associated and weakly to individuals in other clusters [6].
104 Chapter 7
Subjects were selected for inclusion in the cluster analysis in case they met the spirometric
criteria for COPD: post-bronchodilator FEV1/FVC ratio < 0.70 [1]. Partitioning cluster analysis
was performed using the hierarchical and k-means methods by determination of the Ward’s
distance. All variables were standardized by using z-scores ((actual score – mean score)/stan-
dard deviation). Cut points for the clustering process were chosen to avoid clusters of less than
10 persons.
A selection of 32 variables used for the cluster analysis was made to provide measures
of airflow limitation, response to bronchodilation, gas transfer, metabolomics breath profile,
health-related quality of life, sputum production, smoking habits and radiological features of
emphysema. All variables are listed in Tables 1 and 2, with the exception of the number of sub-
jects, GOLD stage and absolute lung volumes which were not included in the cluster analy-
sis. Differences of variables between the clusters were assessed by means of Chi2, ANOVA or
Kruskal-Wallis tests, as appropriate. For validation of the clusters, a linear canonical cross-vali-
dation discriminant analysis was performed based on the leave-one-out method resulting in
cross-validated accuracy values.
RESULTSBaseline characteristicsThree hundred subjects completed the tests, of which 157 (52%) fulfilled the criteria for COPD
[1] and entered the cluster analysis. For the flow chart of the NELSON study in The Netherlands
and the COPD substudy, see Figure 1. The characteristics of the subjects that participated in
the cluster analysis are shown in Table 1.
Table 1 Subject demographics and clinical characteristics of the total cohort and the COPD clusters identi-fied by cluster analysis.
Total COPD cohort
Cluster 1 Cluster 2a Cluster 2b Cluster 3 p-value*
Number of subjects 157 19 51 50 37GOLD stage I/II/III/IV (%) 70/27/3 10/74/16 78/18/4 76/24/0 81/19/0 <0.001Sex (% female) 24 32 20 26 22 0.675Age (yr) 63.2 (5.7) 62.6 (5.8) 64.3 (6.1) 63.0 (5.3) 62.5 (5.8) 0.493BMI (kg/m2) 26.8 (3.9) 24.8 (3.7) 27.4 (3.8) 26.8 (3.4) 26.8 (4.4) 0.016Pre-BD FEV1 (L) 2.77 (0.75) 2.09 (0.63) 2.86 (0.82) 2.87 (0.67) 2.85 (0.64) <0.001Pre-BD FEV1 (%pred) 88.2 (17.6) 66.2 (15.6) 90.5 (17.2) 92.8 (15.9) 89.9 (12.9) <0.001Post-BD FEV1 (L) 2.88 (0.76) 2.22 (0.68) 3.00 (0.84) 2.97 (0.69) 2.92 (0.64) <0.001Post-BD FEV1 (%pred) 91.6 (17.3) 70.1 (16.9) 94.8 (16.1) 96.1 (16.1) 92.2 (12.5) <0.001Reversibility in FEV1 (%) 3.4 (6.4) 3.9 (4.3) 4.3 (7.0) 3.3 (7.0) 2.3 (5.7) 0.455Post-BD FEV1/FVC 0.61 (0.08) 0.50 (0.09) 0.61 (0.07) 0.64 (0.05) 0.62 (0.05) <0.001KCO (%) 78.6 (26.0) 54.0 (29.4) 83.3 (23.6) 85.4 (18.5) 75.6 (28.7) <0.001Packyears 44.8 (21.9) 48.6 (24.4) 48.1 (26.0) 41.8 (19.8) 42.3 (16.5) 0.593Current smokers, % 55 53 37 62 70 0.013
*p-value from analysis of variance or Chi-square test between the clusters.Numeric data are expressed as mean (SD). Post-BD: post-bronchodilator values after 400 µg salbutamol.All variables were included in the cluster analysis except for number of subjects, GOLD stage and absolute lung volumes.
Subphenotyping COPD using breath profiles 105
Chap
ter 7
Figure 1 Flowchart of subjects in the study.IC: informed consent *due to logistic reasons
Questionnaires sent out in The NetherlandsN=548,498
Complete IC and eligible for participationN=15,822
No responseNot fulfilling inclusion criteriaN=532,676
Randomized to CT armN=7915
Baseline screening at location UtrechtN=2997
Randomized to lung function*N=1817
Randomized for participation COPD study*N=300
Fulfilling COPD criteriaN=157
Into Cluster Analysis
Randomized to control armN=7907
Baseline screening at other locationsN=4918
Randomized to no lung functionN=1180
Randomized to no participationCOPD studyN=1517
Not fulfilling COPD criteriaN=143
Cluster analysisVariables that were included in the cluster analysis and their values per cluster are listed in
Tables 1 and 2. The cluster analysis identified 3 distinct clusters based on the 32 variables in-
cluded in the model (Table 3). Clusters showed contrasts with respect to airflow limitation,
gas transfer, airway inflammation, health-related quality of life, sputum production, current
smoking habits, comorbidity and radiologic lung density. Demographic variables, response
to bronchodilation and cumulative smoking history showed no differences between clusters.
One out of 4 electronic nose principal components (PC2) showed large differences between
clusters, whereas the other 3 PC did not (Table 2).
106 Chapter 7
Cluster 1 grouped 19 subjects (12% of total) that were highly symptomatic with respect to
dyspnea, both at exercise and at rest and had a reduced quality of life. This group showed
an impaired diffusion capacity, and markedly reduced lung function. CT emphysema scores
showed little emphysema, and electronic nose principal component 2 (PC2) showed low val-
ues indicating a distinct breath molecular profile. Additional features in this group were a high
use of inhaled corticosteroids and long-acting bronchodilators and little comorbidity (Tables
1-3).
Cluster 2 grouped 101 subjects (64% of total) with relatively mild disease. Subjects exhibited
mild airflow obstruction, good quality of life, no use of inhaled corticosteroids and the ab-
sence of symptoms of cough or chronic sputum production. Subjects reported symptoms of
dyspnea only during exercise. Electronic nose PC2 showed high values representing a char-
acteristic molecular profile. A relatively high percentage of subjects showed cardiovascular
comorbidity and diabetes (Tables 1-3). Cluster 2 could be divided into 2 subclusters based on
a single parameter: CT 15th percentile point (perc15). Subcluster 2a (51 subjects; 50% of cluster
2; 32% of total) showed a mean perc15 of -960.6 (SD 12.1) HU, indicating a lower overall lung
density. In contrast, subcluster 2b (50 subjects; 50% of cluster 2; 32% of total) showed a higher
lung density with a mean perc15 of -929.3 (SD 13.2) HU. Subclusters 2a and 2b did not differ in
any of the variables listed in tables 1 and 2, except for the percentage of current smokers: 37%
for cluster 2a vs 62% for cluster 2b.
Cluster 3 grouped 37 subjects (24% of total) with characteristics of chronic bronchitis. Airflow
limitation was mild and diffusion capacity somewhat impaired. Current smoking was reported
in 70% of cases. Perc15 and electronic nose PC2 were in the intermediate range (Tables 1-3).
validation of clustersIn order to validate the clusters that were identified, a cross-validation discriminant analysis
was performed (Table 4), showing a cross-validated accuracy of 94% (p<0.001) for clusters 1, 2
and 3 and an accuracy of 83% (p<0.001) for subclusters 2a and 2b. In an overall analysis with
classification of subjects into 4 clusters (1, 2a, 2b and 3), an accuracy of 85% (p<0.001) was
reached.
Subphenotyping COPD using breath profiles 107
Chap
ter 7
Table 2 Description of the total cohort and the COPD clusters identified by cluster analysis.
Total COPD cohort
Cluster 1 Cluster 2a Cluster 2b Cluster 3 p-value*
Number of subjects 157 19 51 50 37
Sputum production % 28 26 1 1 100 <0.001
Chronic bronchitis % 22 21 4 0 76 <0.001
Dyspnea at exercise % 53 84 45 48 54 0.011
Dyspnea at rest % 5 26 0 0 8 <0.001
Chronic unproductive cough % 11 26 0 2 30 <0.001
2nd hand smoke % 24 0 33 20 31 0.005
Cancer, any form, % 5 26 6 0 0 <0.001
Inhaled corticosteroid use % 7 37 0 0 11 <0.001
Long-acting bronchodilator use % 9 53 0 2 8 <0.001
Hypercholesterolaemia % 32 16 39 32 32 0.240
Asthma % 10 53 4 2 5 <0.001
Diabetes % 13 11 22 8 8 0.056
Cardiovascular disease % 20 5 27 22 16 0.053
Hypertension % 24 21 31 14 27 0.841
CCQ, median (IQR) 6 (4-11) 14 (7-21) 6 (3-9) 4 (2-6) 9 (6-13) <0.001
Perc15 (HU), mean (SD) -942.8 (21.5) -933.9 (33.7) -960.6 (12.1) -929.7 (13.2) -940.7 (16.0) <0.001
VI-HU910 (%), mean (SD) 41.6 (15.1) 38.7 (21.4) 54.4 (7.7) 32.1 (10.4) 38.3 (12.5) <0.001
VI-HU950 (%), mean (SD) 13.9 (8.5) 12.7 (10.4) 21.8 (6.0) 7.6 (4.5) 12.2 (5.9) <0.001
eNose PC1, mean (SD) 0.00 (1.00) -0.03 (1.21) 1.72 (0.90) -0.10 (1.13) -0.09 (0.83) 0.811
eNose PC2, mean (SD) 0.00 (1.00) -0.61 (1.01) 1.30 (0.88) 0.12 (1.00) -0.03 (1.07) 0.013
eNose PC3, mean (SD) 0.00 (1.00) 0.05 (1.25) 0.14 (1.09) -0.06 (0.85) -0.13 (0.93) 0.621
eNose PC4, mean (SD) 0.00 (1.00) 0.09 (1.00) 0.02 (1.02) -0.12 (0.98) -0.08 (1.02) 0.786
CCQ: clinical COPD questionnaire. Perc15: 15th percentile point. VI-HU910: Voxel index at a threshold of -910 Hounsfield Units. VI-HU950: Voxel index at a threshold of -950 Hounsfield Units. PC: principal component of exhaled breath electronic nose. Chronic bronchitis is defined as the presence of cough and sputum production for at least 3 months in each of 2 consecutive years [1]. All variables were included in the cluster analysis except for number of subjects.
Table 3 Description of the clusters.
Cluster 1 Cluster 2 Cluster 3
No of subjects (%) 19 (12%) 101 (64%) 37 (24%)
Symptoms ++ 0 +Inhalation medication ++ 0 +Quality of life – – 0 –Comorbidity 0 ++ +Lung function – – – –Diffusion capacity – – – –Radiologic emphysema 0 + +eNose PC2 – + 0
PC2: principal component 2 of exhaled breath electronic nose. + increased, – decreased, 0 intermediate.
108 Chapter 7
Table 4 Results of the cross-validation discriminant analysis of clusters.
validation Accuracy Cross-validated accuracy p-value
Clusters 1, 2, 3 99 % 94 % <0.001
Clusters 2a, 2b based on all variables 98 % 83 % <0.001
Clusters 2a, 2b based on perc15 only 98 % 95 % <0.001
Clusters 1, 2a, 2b, 3 96 % 85 % <0.001
DISCUSSIONThe present cluster analysis derived from a community-based population of heavy (ex-) smok-
ers identified 3 distinct clusters of COPD patients. It appeared that symptoms, spirometry, CT
lung density and exhaled molecular profiling all contributed significantly to distinguish these
COPD subphenotypes. Cluster 1 showed the largest impairment in lung function and many
subjects reported symptoms of dyspnea. Cluster 2 consisted mainly of mild COPD patients.
This cluster was subdivided into subjects with and without low lung density on chest CT.
Cluster 3 was characterized by chronic bronchitis. This study confirms phenotypes of COPD of
other studies, and could therefore provide a new taxonomy for COPD [5].
Multidimensional assessment of COPD is increasingly recognized as an effective approach
to discover clinically relevant phenotypes that may differ in natural course of disease and
response to therapies [4]. The method of cluster analysis that was used to identify COPD
phenotypes is a method for classification of subjects into homogeneous groups based on a
heterogeneous set of variables [6]. This method is considered to be hypothesis-generating
rather than potentially biased by a priori assumptions.
Previously, several groups adopted similar approaches to identify subphenotypes in pa-
tients with COPD [7-11]. Subphenotypes that were identified were severe COPD [7-9,11], mild
COPD [8,9,11], chronic bronchitis [7-10] and emphysematous COPD [10,11]. These studies
were focused on clinical COPD subphenotypes, and often did not include biomarkers or CT
parameters. In contrast, subjects in the present study were randomly recruited from a commu-
nity-based survey among heavy (ex-) smokers. This makes our results better generalisable in
the primary care population. In addition to the regular diagnostic workup for COPD, we also
included spiral CT scanning of the chest in order to assess the presence and extent of emphy-
sema, and exhaled breath metabolomics using an electronic nose. The latter is a non-invasive
integrative analysis that appears to be associated with the inflammatory profile in COPD [12].
The clusters identified in this study confirm and extend the COPD phenotypes found by others
using cluster analysis, including the severely impaired group (cluster 1) [7-9,11], the chronic
bronchitis group (cluster 3) [7-10], the emphysematous group (cluster 2a) [10,11] and the mild
COPD group (cluster 2b) [8,9,11]. This not only serves as an external validation [29], but indi-
cates that clinical subphenotypes of COPD can also be found in a general population of heavy
Subphenotyping COPD using breath profiles 109
Chap
ter 7
smokers. Besides this, molecular subphenotyping can be obtained by a fast and non-invasive
breath test.
We made large efforts to ensure the validity of our findings by careful methodological
choices. First, COPD was established based on the GOLD criteria [1], being the current gold-
standard. This makes the results easily comparable to other studies. Second, the recruitment
of subjects from the general population, rather than selected clinical groups of patients, en-
sured that the data can be generalized to COPD in the community, predominantly featuring
mild to moderate airways obstruction. Third, the variables in the cluster analysis were chosen
to reflect different aspects of COPD, including functional, clinical, inflammatory, radiological
and patient-centered outcomes. Fourth, most studies using cluster analysis excluded subjects
with missing data. This markedly decreases the available sample size, as it often occurs that a
variable is missing due to logistic reasons or technical issues. Rather than excluding subjects,
we used multiple imputation to generate replacements of the small proportion of randomly
missing data [26-28]. This procedure has been validated and is considered a reliable method
for unbiased analysis of missing data when taking into account all observed data [26].
The choice for a community-derived sample of subjects may also be considered as a limi-
tation in this study. We recognize that the different phenotypes that were found are directly
influenced by the population in which the study was carried out. However, the fact that all
COPD phenotypes in this study are consistent with those in other studies as well, strength-
ens the validity of our findings. Notably, these phenotypes appear to be already apparent at
mild or early stages of the disease [4]. A second potential limitation is that the use of disease-
modifying treatments such as inhaled corticosteroids and the presence of comorbidities may
have affected the results. Comorbidities were assessed by questionnaires and by medication
usage rather than by systematic diagnostic check-up. Therefore, we cannot completely ex-
clude missing a proportion of yet undiagnosed comorbidities. On the other hand, medication
use and comorbidity were not used as selection criteria, thereby reflecting a real life commu-
nity-derived COPD population and daily practice in primary care.
Each of the clusters showed a distinct profile: a combination of features covering the
different disease domains of COPD. This cannot be translated directly into (partly) distinct,
driving pathogenetic mechanisms. However, one may speculate that inflammatory subtype
and -activity plays an important role. Cluster 1 represents subjects with high ICS use and
asthma prevalence (53%), cluster 2a includes subjects with a high prevalence of metabolic
syndrome comorbidity and cluster 3 represents subjects with chronic bronchitis. Especially
cluster 2a may refer to the chronic systemic inflammatory syndrome with a high percentage of
subjects presenting with hypertension, diabetes, hypercholesterolaemia and cardiovascular
diseases [30]. This is likely to explain the observation that exhaled breath metabolomic profiles
significantly differ between clusters, as specific exhaled profiles can also be found in diseases
associated with metabolic or inflammatory processes such as diabetes [31-33], heart failure
[34] and asthma [23,35].
110 Chapter 7
This study is likely to have clinical implications. First, a better understanding of the different
COPD phenotypes may facilitate the development of more specific diagnostic procedures and
targeted therapies. COPD subphenotyping will benefit the design of randomized controlled
trials, especially in mild or early stages of the disease in the primary care population [5,13]. It
has been estimated that only 1 in 20 COPD patients identified from a community survey would
meet the current criteria for inclusion in an RCT for COPD [36]. Representative samples of the
most prevalent phenotypes of COPD will strengthen the implications of clinical trials. Second,
exhaled metabolomic profiling contributed to the subphenotyping of COPD. This indicates
not only that inflammatory status [12] and thereby the potential response to therapy differs
between clusters, but also that such molecular subphenotyping can be obtained by adding a
rapid, non-invasive breath test to the diagnostic work-up of COPD. Finally, in order to examine
the stability and clinical course of the COPD subphenotypes, longitudinal follow-up studies
and treatment response studies should be carried out [4].
In conclusion, in a community-derived population of predominantly mild to moderate COPD
patients, 3 distinct subphenotypes could be identified using cluster analysis of clinical, func-
tional, CT lung density and metabolomics data: markedly impaired COPD, chronic bronchitis,
and mild COPD. The latter cluster could be subdivided into emphysematous and non-emphy-
sematous COPD. Using such new taxonomy could result in better phenotyping of COPD and
thereby to potentially better management in daily practice and more focussed clinical trials.
Subphenotyping COPD using breath profiles 111
Chap
ter 7
REFERENCES
1. Rabe KF, Hurd S, Anzueto A, Barnes PJ, Buist SA, Calverley P, Fukuchi Y, Jenkins C, Rodriguez-Roisin R, van Weel C, Zielinski J. Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease: GOLD executive summary. Am J Respir Crit Care Med 2007 Sep 15;176:532-55.
2. Beasley R, Weatherall M, Travers J, Shirtcliffe P. Time to define the disorders of the syndrome of COPD. Lancet 2009 Aug 29;374:670-2.
3. Wardlaw AJ, Silverman M, Siva R, Pavord ID, Green R. Multi-dimensional phenotyping: towards a new taxonomy for airway disease. Clin Exp Allergy 2005 Oct;35:1254-62.
4. Han MK, Agusti A, Calverley PM, Celli BR, Criner G, Curtis JL, Fabbri LM, Goldin JG, Jones PW, MacNee W, Make BJ, Rabe KF, Rennard SI, Sciurba FC, Silverman EK, Vestbo J, Washko GR, Wouters EF, Martinez FJ. Chronic obstructive pulmonary disease phenotypes: the future of COPD. Am J Respir Crit Care Med 2010 Sep 1;182:598-604.
5. Shirtcliffe P, Weatherall M, Travers J, Beasley R. The multiple dimensions of airways disease: targeting treatment to clinical phenotypes. Curr Opin Pulm Med 2010 Dec 9.
6. Weatherall M, Shirtcliffe P, Travers J, Beasley R. Use of cluster analysis to define COPD phenotypes. Eur Respir J 2010 Sep;36:472-4.
7. Weatherall M, Travers J, Shirtcliffe PM, Marsh SE, Williams MV, Nowitz MR, Aldington S, Beasley R. Distinct clinical phenotypes of airways disease defined by cluster analysis. Eur Respir J 2009 Oct;34:812-8.
8. Cho MH, Washko GR, Hoffmann TJ, Criner GJ, Hoffman EA, Martinez FJ, Laird N, Reilly JJ, Silverman EK. Cluster analysis in severe emphysema subjects using phenotype and genotype data: an exploratory in-vestigation. Respir Res 2010;11:30.
9. Burgel PR, Paillasseur JL, Caillaud D, Tillie-Leblond I, Chanez P, Escamilla R, Court-Fortune, Perez T, Carre P, Roche N. Clinical COPD phenotypes: a novel approach using principal component and cluster analyses. Eur Respir J 2010 Sep;36:531-9.
10. Pistolesi M, Camiciottoli G, Paoletti M, Marmai C, Lavorini F, Meoni E, Marchesi C, Giuntini C. Identification of a predominant COPD phenotype in clinical practice. Respir Med 2008 Mar;102:367-76.
11. Garcia-Aymerich J, Gomez FP, Benet M, Farrero E, Basagana X, Gayete A, Pare C, Freixa X, Ferrer J, Ferrer A, Roca J, Galdiz JB, Sauleda J, Monso E, Gea J, Barbera JA, Agusti A, Anto JM. Identification and prospective validation of clinically relevant chronic obstructive pulmonary disease (COPD) subtypes. Thorax 2011 May;66:430-7.
12. Marsh SE, Travers J, Weatherall M, Williams MV, Aldington S, Shirtcliffe PM, Hansell AL, Nowitz MR, McNaughton AA, Soriano JB, Beasley RW. Proportional classifications of COPD phenotypes. Thorax 2008 Sep;63:761-7.
13. van Iersel CA, de Koning HJ, Draisma G, Mali WP, Scholten ET, Nackaerts K, Prokop M, Habbema JD, Oudkerk M, van Klaveren RJ. Risk-based selection from the general population in a screening trial: selec-tion criteria, recruitment and power for the Dutch-Belgian randomised lung cancer multi-slice CT screen-ing trial (NELSON). Int J Cancer 2007 Feb 15;120:868-74.
14. Fletcher CM, Pride NB. Definitions of emphysema, chronic bronchitis, asthma, and airflow obstruction: 25 years on from the Ciba symposium. Thorax 1984 Feb;39:81-5.
15. Burney PG, Luczynska C, Chinn S, Jarvis D. The European Community Respiratory Health Survey. Eur Respir J 1994 May;7:954-60.
16. van der Molen T, Willemse BW, Schokker S, ten Hacken NH, Postma DS, Juniper EF. Development, validity and responsiveness of the Clinical COPD Questionnaire. Health Qual Life Outcomes 2003;1:13.
17. Miller MR, Hankinson J, Brusasco V, Burgos F, Casaburi R, Coates A, Crapo R, Enright P, van der Grinten CP, Gustafsson P, Jensen R, Johnson DC, Macintyre N, McKay R, Navajas D, Pedersen OF, Pellegrino R, Viegi G, Wanger J. Standardisation of spirometry. Eur Respir J 2005 Aug;26:319-38.
18. Macintyre N, Crapo RO, Viegi G, Johnson DC, van der Grinten CP, Brusasco V, Burgos F, Casaburi R, Coates A, Enright P, Gustafsson P, Hankinson J, Jensen R, McKay R, Miller MR, Navajas D, Pedersen OF, Pellegrino R, Wanger J. Standardisation of the single-breath determination of carbon monoxide uptake in the lung. Eur Respir J 2005 Oct;26:720-35.
112 Chapter 7
19. Gietema HA, Schilham AM, van Ginneken B, van Klaveren RJ, Lammers JW, Prokop M. Monitoring of smoking-induced emphysema with CT in a lung cancer screening setting: detection of real increase in extent of emphysema. Radiology 2007 Sep;244:890-7.
20. Stoel BC, Stolk J. Optimization and standardization of lung densitometry in the assessment of pulmonary emphysema. Invest Radiol 2004 Nov;39:681-8.
21. Dragonieri S, Schot R, Mertens BJ, Le Cessie S, Gauw SA, Spanevello A, Resta O, Willard NP, Vink TJ, Rabe KF, Bel EH, Sterk PJ. An electronic nose in the discrimination of patients with asthma and controls. J Allergy Clin Immunol 2007 Oct;120:856-62.
22. Fens N, Zwinderman AH, van der Schee MP, de Nijs SB, Dijkers E, Roldaan AC, Cheung D, Bel EH, Sterk PJ. Exhaled breath profiling enables discrimination of chronic obstructive pulmonary disease and asthma. Am J Respir Crit Care Med 2009 Dec 1;180:1076-82.
23. Lewis NS. Comparisons between mammalian and artificial olfaction based on arrays of carbon black-polymer composite vapor detectors. Acc Chem Res 2004 Sep;37:663-72.
24. Röck F, Barsan N, Weimar U. Electronic nose: current status and future trends. Chem Rev 2008 Feb;108:705-25.
25. Marti H, Chavance M. Multiple imputation analysis of case-cohort studies. Stat Med 2011 Feb 24.26. de Groot JA, Janssen KJ, Zwinderman AH, Bossuyt PM, Reitsma JB, Moons KG. Correcting for partial veri-
fication bias: a comparison of methods. Ann Epidemiol 2011 Feb;21:139-48.27. Rubin DB, Schenker N. Multiple imputation in health-care databases: an overview and some applications.
Stat Med 1991 Apr;10:585-98.28. Fabbri LM, Rabe KF. From COPD to chronic systemic inflammatory syndrome? Lancet 2007 Sep 1;370:797-
9.29. Phillips M, Cataneo RN, Cheema T, Greenberg J. Increased breath biomarkers of oxidative stress in diabe-
tes mellitus. Clin Chim Acta 2004 Jun;344:189-94.30. Minh TD, Oliver SR, Ngo J, Flores RL, Midyett J, Meinardi S, Carlson MK, Rowland FS, Blake DR, Galassetti
PR. Non-invasive Measurement of Plasma Glucose from Exhaled Breath in Healthy and Type 1 Diabetic Mellitus Subjects. Am J Physiol Endocrinol Metab 2011;300:E1166-75.
31. Greiter MB, Keck L, Siegmund T, Hoeschen C, Oeh U, Paretzke HG. Differences in exhaled gas profiles between patients with type 2 diabetes and healthy controls. Diabetes Technol Ther 2010 Jun;12:455-63.
32. Phillips M, Cataneo RN, Greenberg J, Grodman R, Salazar M. Breath markers of oxidative stress in patients with unstable angina. Heart Dis 2003 Mar;5:95-9.
33. Fens N, Roldaan AC, van der Schee MP, Boksem RJ, Zwinderman AH, Bel EH, Sterk PJ. External validation of exhaled breath profiling using an electronic nose in the discrimination of asthma with fixed airways obstruction and chronic obstructive pulmonary disease. Clin Exp Allergy 2011 Apr 18;inpress.
34. Travers J, Marsh S, Williams M, Weatherall M, Caldwell B, Shirtcliffe P, Aldington S, Beasley R. External validity of randomised controlled trials in asthma: to whom do the results of the trials apply? Thorax 2007 Mar;62:219-23.