Abstract
Sarcoidosis is a highly variable, systemic granulomatous disease of hitherto unknown aetiology. The GenPhenReSa (Genotype–Phenotype Relationship in Sarcoidosis) project represents a European multicentre study to investigate the influence of genotype on disease phenotypes in sarcoidosis.
The baseline phenotype module of GenPhenReSa comprised 2163 Caucasian patients with sarcoidosis who were phenotyped at 31 study centres according to a standardised protocol.
From this module, we found that patients with acute onset were mainly female, young and of Scadding type I or II. Female patients showed a significantly higher frequency of eye and skin involvement, and complained more of fatigue. Based on multidimensional correspondence analysis and subsequent cluster analysis, patients could be clearly stratified into five distinct, yet undescribed, subgroups according to predominant organ involvement: 1) abdominal organ involvement, 2) ocular–cardiac–cutaneous–central nervous system disease involvement, 3) musculoskeletal–cutaneous involvement, 4) pulmonary and intrathoracic lymph node involvement, and 5) extrapulmonary involvement.
These five new clinical phenotypes will be useful to recruit homogenous cohorts in future biomedical studies.
Abstract
Five new clinical phenotypes of sarcoidosis have been identified by analysing organ manifestations of 1932 patients http://ow.ly/UYLC30jpUkq
Introduction
Sarcoidosis is a highly variable, systemic granulomatous disease of hitherto unknown aetiology. Even though the disease affects people worldwide, sarcoidosis cohorts are not homogeneous, and differ in terms of age, sex, ethnicity, type of onset and organ involvement [1–5]. The clinical course of sarcoidosis ranges from spontaneous resolution to disabling chronic disease, with lung transplantation as the last resort. More serious complications are pulmonary fibrosis, renal failure, cardiac involvement, neurosarcoidosis, defacing lupus pernio and loss of sight due to posterior uveitis. This phenotypic variability may reflect the difference in genetic background, environmental exposure profile and socioeconomic status of sarcoidosis patients [6–10]. Onset of the disease is either acute or subacute. Acute onset is characterised by fever, fatigue, weight loss and/or with the triad of erythema nodosum, bihilar lymphadenopathy and arthritis called Löfgren syndrome [11, 12], whereas subacute onset is typically nonspecific, with cough, dyspnoea and/or chest pain being the most common symptoms [13, 14].
Previous epidemiological surveys, such as the American ACCESS [2], the German–Swiss WATL [3] and the retrospective single-centre MUSC [15] studies, describe this clinical heterogeneity of sarcoidosis, and highlight an urgent need for further detailed clinical phenotypes to better understand the pathomechanisms of the disease and to plan future clinical therapeutic studies [16, 17]. The GenPhenReSa (Genotype–Phenotype Relationship in Sarcoidosis) project, a European multicentre study designed to investigate the influence of genotypes on disease phenotypes in sarcoidosis, addresses this issue by the detailed characterisation of over 2000 European sarcoidosis patients. Here, we present the results of the phenotype module of GenPhenReSa.
Methods
Subjects
31 study centres providing tertiary pulmonary care either as an academic or national centre (supplementary table S1) consecutively screened their prevalent sarcoidosis patients for the following inclusion criteria: 1) diagnosis of sarcoidosis according to the consensus statement of the American Thoracic Society, European Respiratory Society (ERS) and World Association of Sarcoidosis and Other Granulomatous Disorders [1], 2) a documented course of disease over at least 2 years prior to recruitment start, thereby providing all parameters of the phenotyping protocol, and 3) Caucasian parentage with a local ancestry over at least two generations to minimise confounding the planned genetic studies by population substructure. Individuals with granulomatous disease that could not be unequivocally diagnosed as sarcoidosis were excluded. Patients who met these criteria were asked for informed consent to participate in this study. In total, 2163 patients with sarcoidosis were included in the study over a period of 3 years.
As the definition of organ involvement in sarcoidosis varies widely, we chose to use the criteria of the ACCESS study [18]. Patients were phenotyped at the 31 study sites according to a standardised protocol that was both validated and field tested [13]. A custom-built software programme was used for data management. The phenotyping protocol included more than 200 clinical parameters covering all organ manifestations. To harmonise pulmonary function data, the ERS-modified European Community for Steel and Coal reference values were applied to the entire cohort [19, 20]. Chest radiographs were grouped according to Scadding [21] at the corresponding study centre. The observational study did not influence the clinical processes or any therapeutic decisions made by the attending physicians, who were solely responsible for the choice and initiation of treatment.
The GenPhenReSa study was aimed at generating the requested data via regular processes of patient care. Staff at all recruiting centres were trained in phenotyping their own patients either during a visit to Freiburg (Germany) or during a visit by the Freiburg team to their centres. Data management was undertaken by the Popgen biobank in Kiel (Germany), which is audited by the independent centre for data protection of the state of Schleswig-Holstein [22]. All phenotype data were checked for plausibility by the Freiburg team before logging the data in the Popgen database. 203 patients had to be excluded because of incomplete or contradictory data. A positive vote by the Ethics Committee of the University of Freiburg was obtained prior to the initiation of the study, which was also registered in the German Clinical Trials Register (identifier DRKS00000045).
Statistical analysis
Patient characteristics were summarised in the form of frequencies (categorical data) or means and standard deviations (continuous data). In case of missing values, frequencies are given for available data only. Intergroup differences were assessed for statistical significance using a Chi-squared test or Wilcoxon rank sum test, as appropriate. In an explorative manner, an unsupervised data mining method was used to identify new possible phenotype groups of patients according to organ involvement as follows. Multiple correspondence analysis (MCA) was applied to extract principal components, which were then subjected to an agglomerative hierarchical clustering (hierarchical clustering on principal components (HCPC)) analysis. MCA was chosen as it allows an unbiased analysis of patterns of relationships between several variables in a categorical data set. Ward's minimum variance method with Euclidean distance was used and the total inertia (the sum of between- and within-group variance) was calculated at each aggregating step [23]. The inertia gain is the increase in within-group variance when moving from one partition to the next. The optimal number of clusters Q is given when the increase of between-inertia between Q–1 and Q clusters is greater than that between Q and Q+1 clusters. The stability of the final cluster solution was assessed by bootstrapping (k=10 000) using Jaccard coefficients (JCs) [24]. Statistical analyses were performed using R [25] (packages FactoMineR [26] and fpc [27]). A p-value <0.05 was considered statistically significant without adjustment for multiple testing.
Results
Patient characteristics
2163 Caucasian patients with sarcoidosis were included in the study. Their mean±sd age was 47.0±12.1 years and 1290 (59.6%) were female. Noncaseating granulomas could be histologically confirmed in 1987 patients (94.6%) (table 1). A total of 327 sarcoidosis patients (15.1%) had at least one organ involvement other than lung or lymph nodes histologically confirmed. Lung involvement (n=1664 (92.6%)) was by far the most common type of organ involvement, followed by mediastinal and/or hilar lymphadenopathy (n=1355 (77.0%)), and skin (n=342 (16.1%)), eye (n=163 (7.8%)) and joint (n=158 (7.5%)) involvement. The most common presenting symptoms were fatigue (n=1112 (61.8%)), cough (n=873 (47.9%)) and dyspnoea (n=818 (44.4%)). The most frequent radiological type was Scadding I (n=764 (36.1%)); 347 of the sarcoidosis patients (16.4%) did not have sarcoidosis-associated chest radiography findings (Scadding type 0) and 83 (3.9%) had signs of lung fibrosis (Scadding type IV). Pulmonary function was only marginally impaired, on average, with forced expiratory volume in 1 s (FEV1) being the most reduced measure (92.8% predicted) (table 1).
Characteristics of the sarcoidosis patients
Sex and age differences
The characteristics of male and female sarcoidosis patients were found to differ considerably in several aspects. Thus, the age at inclusion in the study peaked much earlier in male patients (between the third and fourth decades) than in females (fifth decade) (figure 1 and supplementary table S2). Male patients had slightly worse lung function. Female patients predominantly presented with Scadding type I (37.6%); male patients predominantly presented with Scadding type II (39.7%). Female patients suffered significantly more frequently from eye (9.2% versus 5.6%; p<0.01), salivary gland (4.7% versus 2.6%; p<0.05) and skin (18.9% versus 11.9%; p<10–4) involvement. Symptoms also showed a sex difference, in that females suffered significantly more often from fatigue (66.4% versus 55.0%; p<10–5), arthralgia (54.7% versus 49.1%; p<10–6) and chest pain (23.4% versus 18.6%; p<0.05), but less from fever or subfebrile temperatures (18.5% versus 23.0%; p<0.05).
Frequency distributions of a) chest radiograph type according to Scadding [21] and b) age at baseline. The percentages are given separately for both sexes.
Another important factor influencing the clinical phenotype of patients with sarcoidosis is age (figure 1). A female/male ratio of 1:1 was observed in patients aged ≤40 years, whereas a female/male ratio of 2:1 was observed in patients aged >40 years. Younger patients presented predominantly with Scadding type I (n=318 (44.7%)), whereas higher frequencies of Scadding type III or IV were noted in older patients. Patients aged ≤40 years were significantly more prone to eye, intrathoracic lymph node and bronchial involvement, but less prone to heart involvement (supplementary table S3). Patients aged ≤40 years suffered significantly more from fever or subfebrile temperatures, but less from dyspnoea compared with patients aged >40 years.
Acute versus subacute onset
Patients with acute onset of disease compared with patients with subacute onset were significantly younger (44.6 versus 48.0 years; p<10–6), female (64.5% versus 56.7%; p<0.001) (table 1) and had a slightly but significantly better lung function (e.g. FEV1 % pred 96.5% versus 90.4%; p<10–6). Patients with acute sarcoidosis more often had a radiologically normal lung (67.7% versus 80.5%; p<10–6) (figure 2), but a higher percentage of histologically proven lung involvement (79.4% versus 70.0%; p<0.001). They more often had bronchial and musculoskeletal involvement, but less frequent cardiac, hepatic or splenic involvement (for frequencies and levels of significance, see table 1). Patients with acute sarcoidosis presented with a different spectrum of symptoms than patients with subacute sarcoidosis: they reported significantly more fatigue, fever, night sweats and arthralgia, but less cough and dyspnoea (supplementary table S4). No significant intergroup differences in terms of comorbidities were observed, including Inflammatory bowel disease: 12 (0.6%) of the sarcoidosis patients suffered from Crohn's disease and 10 (0.5%) had ulcerative colitis. For details, see supplementary table S4.
Frequency distributions of a) organ involvement and b) symptoms. CNS: central nervous system. *: p<0.05, comparing patients with acute versus subacute onset.
Treatment
1303 sarcoidosis patients (61.3%) were in need of treatment. Treatment regimens varied and were in the hands of the recruiting centres. Of the drug-treated patients, 92.7% received prednisolone or equivalents, 22.8% methotrexate, 11.2% azathioprine, 2.5% chloroquine, 1.3% infliximab, 0.8% cyclophosphamide and 0.3% cyclosporine; none received thalidomide. Reasons to start medication as stated by the treating physicians included clinical symptoms of cough and/or dyspnoea in 56.3%, clinical symptoms of fatigue, fever or weight loss in 63.1%, extrapulmonary organ involvement in 28.4%, and loss of pulmonary function in 23.5%.
Organ manifestation phenotypes
Organ involvement in sarcoidosis is characterised by a pronounced heterogeneity and seems to manifest randomly. In order to indentify underlying patterns of organ involvement, we performed MCA followed by HCPC analysis (figure 3). Patients with >50% missing data on organ involvement were excluded (supplementary table S5); the remaining 1932 patients were analysed in the MCA and HCPC analysis. The MCA led to the following main dimensions (supplementary table S6). The first and second dimensions, which explain 28% and 22% of the total variance, separated patients with arthritis and/or musculoskeletal involvement from patients with abdominal affection. The third dimension (19% explained variance) then segregated patients with of eye, skin, heart, central nervous system (CNS) or salivary gland involvement, with the fourth residual cluster of pulmonary–lymphonodal sarcoidosis patients. Subsequently, the HCPC identified five phenotypes (table 2 and see figure 4 for dendrogram), which are presented in detail, with frequencies always given for affected versus nonaffected patients:
Multiple correspondence analysis (MCA) and clusters of organ involvement that allow analysis of patterns of relationships between categorical variables (different organ involvement). CNS: central nervous system. a, b) Scatter plots of patients based on different dimensions derived from MCA coloured by cluster membership with explained variance given in brackets: a) dimension 1 versus dimension 2 and b) dimension 1 versus dimension 3. Cluster 1: abdominal; cluster 2: ocular–cardio–cutaneous–CNS; cluster 3: musculoskeletal–cutaneous; cluster 4: pulmonary–lymphonodal; cluster 5: extrapulmonary. Data are presented as dots for individual patients; open squares represent cluster centres. c, d) Scatter plots of variables contributing to different dimensions: c) dimension 1 versus dimension 2 and d) dimension 1 versus dimension 3. Labelling of variables close to the zero-point is omitted due to space constraints.
Characteristics (organ involvement and symptoms) of the patients# clustered into the five phenotypes
Dendrogram of the hierarchical clustering on principal components analysis leading to five clusters of organ involvement. Cluster 1: abdominal; cluster 2: ocular–cardio–cutaneous–central nervous system; cluster 3: musculoskeletal–cutaneous; cluster 4: pulmonary–lymphonodal; cluster 5: extrapulmonary. The height corresponds to the inertia gain.
1) Abdominal (renal/splenic/hepatic) involvement (n=133, JC 0.83). In patients with liver involvement, 36.7% also had spleen involvement (versus 2.1% without; p<10–6), 10% also had kidney involvement (versus 2.6% without; p<0.001) and more often intrathoracic lymph node involvement (87.5% versus 76.8%; p<0.05). These patients typically experienced weight loss (21.4% versus 7.8%; p<0.001) and night sweats (26.5% versus 11.4%; p<0.001). The most common organs involved in patients with splenic involvement were liver (45.2% versus 3.0%; p<10–6) and kidney (11.0% versus 2.7%; p<0.001). The most prevalent symptom in patients with splenic and renal involvement was weight loss (19.3% versus 8.2%; p<0.01 and 22.5% versus 8.4%; p<0.01, respectively). A similar spectrum was observed for renal involvement. Patients with splenic, hepatic or renal involvement also presented significantly more often with impaired lung function, e.g. with a reduction of about –5% FEV1 % pred.
2) Ocular–cardiac–cutaneous–CNS (“OCCC”; eye/heart/skin/salivary glands/CNS) involvement (n=240, JC 0.82). Patients with ocular involvement more frequently showed skin (37.5% versus 14.2%; p<10–6), CNS (13.8% versus 2.5%; p<10–6) and heart (8.9% versus 2.7%; p<10–4) sarcoidosis. Patients with neurosarcoidosis suffered more often from ocular (32.4% versus 7.1%; p<10–6) and cardiac (11.6% versus 2.8%; p<0.001) involvement. An increased frequency of ocular (18.8% versus 6%; p<10–6) and cardiac involvement (6.1% versus 2.5%; p<0.01) was observed in patients with skin sarcoidosis. Patients with eye or skin sarcoidosis also suffered more often from involvement of the salivary glands (both p<0.05; for eye: 11.3% versus 3.1%).
In cardiac, CNS and skin sarcoidosis, fatigue was more prevalent than in patients without these organ manifestations (all p<0.05). Interestingly, the rate of arthralgia was also increased in patients with eye, cardiac or skin involvement (65.4% versus 49.7%, 71.7% versus 50.5% and 64.6% versus 48.6%, respectively; all p<0.01). A significantly reduced frequency of fever or subfebrile temperature (12.2% versus 21.3%; p<0.05) was observed in patients with eye involvement.
3) Musculoskeletal–cutaneous involvement (n=189, JC 0.86). Patients with arthritis or musculoskeletal involvement presented significantly more often with an acute onset (both p<10–6), and suffered more from fever or subfebrile temperature, night sweats, weight loss and arthralgia (all at least p<0.05) than patients without (for frequencies and levels of significance, see supplementary table S7). Patients with arthritis or musculoskeletal involvement less often showed involvement of the lungs or bronchi, but more often of the skin, intrathoracic lymph nodes and kidneys (all at least p<0.05). Patients with musculoskeletal sarcoidosis also suffered significantly more often from eye involvement (p<10–3), whereas there was only a trend towards higher eye involvement in patients with arthritis. There was no significant difference between patients with or without arthritis or musculoskeletal involvement in terms of age, sex and lung function data.
4+5) Pulmonary–lymphonodal (n=1257, JC 0.99) and extrapulmonary sarcoidosis (n=113, JC 0.92). The remaining patients clustered in a group of no extrathoracic sarcoid involvement and a group of extrapulmonary sarcoidosis. Patients with lung involvement had worse lung function, and suffered more often from bronchial and intrathoracic lymph node involvement, but less often from skin or musculoskeletal involvement (all at least p<0.05; for frequencies and levels of significance, see supplementary table S8). They presented significantly more often with dyspnoea, cough, chest pain and fatigue (all at least p<0.01). Interestingly, the extrapulmonary phenotype had the highest frequencies for kidney involvement.
See table 2 for more details of the five new phenotypes.
To evaluate cluster stability, we performed bootstrapping using the JC values. Clusters with JC >0.75 are generally considered as valid, stable clusters, which all five clusters fulfil [27]. The pulmonary–lymphonodal, musculoskeletal–cutaneous and extrapulmonary clusters even had JC >0.85, indicative of high cluster stability.
Discussion
The GenPhenReSa study is a large-scale European multicentre sarcoidosis study of over 2000 patients that has documented patient characteristics in great detail. Phenotype analysis identified five novel distinct phenotype groups.
The sex distribution in our study showed a female predominance with a 3:2 female/male ratio, which is in line with the ACCESS study and a recent French study [2, 28]. However, in patients aged ≤40 years or with subacute onset, the ratio was almost 1:1. Age at study inclusion peaked in females during the fifth decade and in males during the fourth decade. This later peak and the predominance of females were similarly observed in the ACCESS and MUSC cohorts, as well as in a recent Swedish study [2, 15, 29]. As shown in the current study, sarcoidosis is not only a disease of young adults; it is also frequently diagnosed in middle-aged and elderly patients.
Pulmonary manifestation is by far the most common organ involvement in sarcoidosis, as shown in our study and a number of previous studies [2, 3, 30–32]. Compared with the ACCESS study [2], the distribution of organ involvement differs slightly. In the ACCESS study, more eye (11.8% versus 8.1%) and hepatic (11.5% versus 4.5%), yet less renal (0.7% versus 3.1%) and musculoskeletal (0.9% versus 7.3%) involvement was recorded than in the current study. These differences may be attributable to ethnic differences (e.g. African-Americans suffer more often from ocular involvement), as proven in the ACCESS study [2]. The distribution of the chest radiography types (Scadding) was similar to the ACCESS and WATL study populations [2, 3]. The higher percentage of Scadding type 0 in our study is explained by the fact that many recruiting centres are tertiary care institutions and in many countries patients with extrathoracic sarcoidosis are referred to pulmonologists for exclusion of pulmonary involvement. Notably, the rate of inflammatory bowel disease (0.5% ulcerative colitis and 0.6% Crohn's disease) was ∼5–10 times higher than the corresponding prevalence rates in Europe [33]. The simultaneous occurrence of sarcoidosis and inflammatory bowel disease might be due to overlapping genetic risk profiles [34]. In patients with subacute sarcoidosis, we found a slightly higher prevalence of cancer in the patients’ history, which is in accordance with a previous study by Askling et al. [35] and corroborates the notion that chronic inflammatory diseases, such as sarcoidosis and Crohn's disease, cause an increase in the incidence of neoplasia [36].
Many previous empirical studies distinguished pulmonary from extrapulmonary sarcoidosis [7, 37–39]. Some single associations between two involved organs have been published, e.g. liver and spleen or CNS and eye [40, 41]. Utilising MCA, we were able to comprehensively analyse associations in organ involvement and to identify five distinct clinical phenotypes of organ involvement: an abdominal, an “OCCC”, a musculoskeletal–cutaneous, a pulmonary–lymphonodal and an extrapulmonary phenotype. The classical Löfgren syndrome [12] is probably concealed behind the musculoskeletal–cutaneous phenotype, but we cannot distinguish between patients with arthritis, muscle, bone or joint sarcoidosis in our database. Skin involvement was part of the “OCCC” and musculoskeletal–cutaneous phenotypes. We have to assume that erythema nodosum predominantly arises in the musculoskeletal–cutaneous phenotype (classical Löfgren-like phenotype) and the other skin occurrences arise in the “OCCC” phenotype, but we have no data on skin involvement subtypes.
Cardiac involvement is part of the “OCCC” phenotype. A hypothetical conclusion might be that the clinically apparent cardiac sarcoidosis is mainly sarcoidosis of the electrical conduction system of the heart, as the conduction system is the structure, besides the myocardium itself, that is most often affected [42].
It is tempting to speculate that these five phenotypes are the sequelae of different aetiological agents causing sarcoidosis, which are either inhaled, ingested or sufficiently lipophilic to pass through the skin and/or the blood–brain barrier [7]. Furthermore, different genetic risk profiles might predispose to one of these five phenotypes. However, simple confounders such as access to diagnostic technologies might also lead to these clusters. Cluster analysis cannot deliver answers as to why these five phenotypes arise and thus further research is required.
This study also has several limitations. First, recruitment bias needs to be considered [43]. Investigators were mainly pulmonologists, so an overrepresentation of pulmonary sarcoidosis is likely. In addition, the centre recruiting the most patients was Belgrade (Serbia), resulting in an overrepresentation of Serbian patients. However, this subcohort (for basic characteristics, see supplementary table S9) did not shift the results of the MCA and did not influence the new phenotypes (supplementary figure S1). Second, all study centres were tertiary referral centres and hence an overrepresentation of patients with complicated, multiorgan sarcoidosis is possible. In this context, it has to be noted that an epidemiological study was not intended. Third, the assignment of the sarcoidosis patients to all variables was dependent on the use of a somewhat rigid phenotyping software tool, thus the full phenotypic spectrum of the disease “sarcoidosis” is potentially not covered in total, although over 200 traits have been documented. In addition, we phenotyped the patients according to the then up-to-date ACCESS criteria of organ involvement and methods of diagnosing them [18]. These criteria have since been updated [44] and new technologies will lead to better detection of organ involvement; therefore, frequencies and clusters of organ involvement will change over time. Fourth, it is well known that the extent and frequency of organ involvement in sarcoidosis varies in different ethnicities [15]; our newly described phenotypes are therefore only applicable in Caucasian cohorts and might be different in, for example, African or Asian populations. Fifth, our new clinical phenotypes do not constitute clearly defined groups of patients, but more likely represent the extremes of a continuum, albeit of scientific usefulness. Sixth, disease duration might influence disease phenotype; therefore, we evaluated the influence of disease duration on MCA and found it to be negligible (supplementary figure S2).
In conclusion, the phenotype of sarcoidosis in patients of Caucasian descent is highly variable, and depends on age, sex and type of onset. We defined an abdominal, an “OCCC”, a musculoskeletal–cutaneous, a pulmonary–lymphonodal and an extrapulmonary phenotype of organ affection. These new phenotypes can now be used in clinical and biomedical studies to obtain homogenous and clearly defined subcohorts of sarcoidosis patients. These results will be the basis of our efforts to link specific genotypes with phenotypes.
Supplementary material
Supplementary Material
Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.
Supplementary material ERJ-00991-2017_supplement
Disclosures
Supplementary Material
F. Bonella ERJ-00991-2017_Bonella
P. Bresser ERJ-00991-2017_Bresser
S. Freitag-Wolf ERJ-00991-2017_Freitag-Wolf
A. Günther ERJ-00991-2017_Gunther
S. Haraldsdottir ERJ-00991-2017_Haraldsdottir
J. Müller-Quernheim ERJ-00991-2017_Muller-Quernheim
S. O'Beirne ERJ-00991-2017_OBeirne
K. O'Reilly ERJ-00991-2017_OReilly
S. Pabst ERJ-00991-2017_Pabst
A. Prasse ERJ-00991-2017_Prasse
P. Rottoli ERJ-00991-2017_Rottoli
D. Valeyre ERJ-00991-2017_Valeyre
Acknowledgements
It is with great regret that we have to thank Philip Quanjer (Emeritus Professor of Physiology, Leiden University, The Netherlands) posthumously for his expertise on lung function data and his critical revision of the manuscript.
Footnotes
Published online Jan 25, 2018; republished Jan 30, 2018 with amendments to author forenames.
This article has supplementary material available from erj.ersjournals.com
This study is registered at the German Clinical Trials Register with identifier number DRKS00000045.
Support statement: This study was supported by German Federal Ministry for Education and Research grant 01EY1103 and German Research Foundation grant MU 692/8-1 and Clusterlab XI EXC 306/2. Funding information for this article has been deposited with the Crossref Funder Registry.
Conflict of interest: Disclosures can be found alongside this article at erj.ersjournals.com
- Received May 15, 2017.
- Accepted October 25, 2017.
- Copyright ©ERS 2018
References
