- Email: [email protected]
Defining the “Frequent Exacerbator” Phenotype in COPD
Accepted Manuscript Defining the « frequent exacerbator » phenotype in COPD: a hypothesis-free approach Olivier Le Rouzic, MD, PhD, Nicolas Roche, MD, PhD, Alexis B. Cortot, MD, PhD, Isabelle Tillie-Leblond, MD, PhD, Frédéric Masure, MD, Thierry Perez, MD, Isabelle Boucot, MD, Latifa Hamouti, MD, Juliette Ostinelli, MD, Céline Pribil, MD, Christine Poutchnine, MD, Stéphane Schück, MD, Mathilde Pouriel, Bruno Housset, MD, PhD PII:
S0012-3692(17)32903-3
DOI:
10.1016/j.chest.2017.10.009
Reference:
CHEST 1390
To appear in:
CHEST
Received Date: 6 May 2017 Revised Date:
31 August 2017
Accepted Date: 2 October 2017
Please cite this article as: Le Rouzic O, Roche N, Cortot AB, Tillie-Leblond I, Masure F, Perez T, Boucot I, Hamouti L, Ostinelli J, Pribil C, Poutchnine C, Schück S, Pouriel M, Housset B, Defining the « frequent exacerbator » phenotype in COPD: a hypothesis-free approach, CHEST (2017), doi: 10.1016/ j.chest.2017.10.009. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT 1
Total word count (text): 2500 words / 2500
2
Total word count (abstract): 222 / 250
3
Title: Defining the « frequent exacerbator » phenotype in COPD: a hypothesis-free approach
4
Running title: Statistically defined COPD “frequent exacerbator”
RI PT
5
Olivier Le Rouzic, MD, PhD; Nicolas Roche, MD, PhD; Alexis B. Cortot, MD, PhD; Isabelle
7
Tillie-Leblond, MD, PhD: Frédéric Masure, MD; Thierry Perez, MD; Isabelle Boucot, MD;
8
Latifa Hamouti, MD; Juliette Ostinelli, MD; Céline Pribil, MD; Christine Poutchnine, MD;
9
Stéphane Schück, MD; Mathilde Pouriel, Bruno Housset, MD, PhD.
SC
6
10
Author affiliations
12 13
Drs Le Rouzic, Perez, Tillie-Leblond and Pr. Cortot: Univ. Lille, CHU Lille, Department of Respiratory Diseases, Lille, France
14 15
Pr. Roche: AP-HP, Hôpital Cochin, Service Pneumologie, EA2511, Université Paris Descartes, Sorbonne Paris Cité, Paris, France
16 17
Dr. Masure: Department of Respiratory Diseases of Groupe Medical Saint Remi, Reims, France
18
Dr. Boucot: GlaxoSmithKline, Brentford, United Kingdom
19
Dr. Hamouti: Boehringer Ingelheim, Paris, France
20
Dr. Ostinelli: AstraZeneca, Rueil-Malmaison, France
21
Dr. Pribil: GlaxoSmithKline, Marly Le Roi, France
22
Dr. Poutchnine: Pfizer, Paris, France
23
Dr. Schück and Ms Pouriel : Kappa Santé, Paris, France
24 25
Pr. Housset: Centre hospitalier intercommunal de Créteil, Service Pneumologie, UPEC, Université Paris-Est, UMR S955, Créteil, France
TE D
EP
AC C
26
M AN U
11
27
Corresponding author:
28
Olivier Le Rouzic
29
Department of Respiratory Diseases
30
University Hospital of Lille
31
1 bd Jules Leclercq
ACCEPTED MANUSCRIPT 1
59037 Lille, France
2
E-mail: [email protected]
3
Funding information
5
This study was conducted with financial support from AstraZeneca, Boehringer Ingelheim,
6
GlaxoSmithKline, Nycomed and Pfizer.
RI PT
4
7
Conflicts of interest
9
Dr. Le Rouzic reports non-financial support from AstraZeneca, Chiesi, GlaxoSmithKline,
10
Leo, MSD, Mundipharma, Santelys Association and Teva, and personal fees and non-
11
financial support from Boehringer Ingelheim, Lilly and Novartis, outside the submitted work.
12
Pr. Roche reports grants and personal fees from Boehringer Ingelheim, Novartis and Pfizer,
13
and personal fees from Teva, GlaxoSmithKline, AstraZeneca, Chiesi, Mundipharma, Cipla,
14
Sanofi, Sandoz, 3M and Zambon, outside the submitted work.
15
Pr. Cortot reports personal fees and non-financial support from Astra-Zeneca, Novartis and La
16
Roche Hoffmann, grants, personal fees and non-financial support from Boehringer-Ingelheim,
17
and personal fees from Bristol Myers Squibb, MSD and Pfizer, outside the submitted work.
18
Dr. Masure has nothing to disclose.
19
Dr. Perez reports personal fees from Boehringer Ingelheim, Novartis, GlaxoSmithKline,
20
Chiesi and Pierre Fabre, outside the submitted work.
21
Drs Boucot and Pribil report being a full time employee of GlaxoSmithKline.
22
Dr. Hamouti reports being a full time employee of Boehringer Ingelheim and having been an
23
employee of Chiesi and Schering Plough.
24
Dr. Ostinelli reports being a full time employee of AstraZeneca.
25
Dr. Poutchnine reports being a full time employee of Pfizer.
26
Dr. Schück and Ms Pouriel report being a full time employee of Kappa Santé.
27
Pr. Housset reports personal fees from GSK, Novartis, Astra-Zeneca, Boehringer Ingelheim,
28
Chiesi and Pfizer, outside the submitted work.
AC C
EP
TE D
M AN U
SC
8
29 30 31
This article has supplemental online data.
ACCEPTED MANUSCRIPT Abbreviations list
2
AUC
area under the curve
3
BMI
body-mass index
4
CI
confidence interval
5
COPD
chronic obstructive pulmonary disease
6
FEV1
forced expiratory volume in first second
7
FVC
forced vital capacity
8
GOLD
Global initiative for chronic Obstructive Lung Disease
9
mMRC
modified Medical Research Council dyspnoea score
10
NS
not significant
11
SD
standard deviation
12
VSRQ
visual simplified respiratory questionnaire
SC
M AN U
TE D EP AC C
13
RI PT
1
1
ACCEPTED MANUSCRIPT Abstract
2
Background: The COPD frequent exacerbator phenotype is usually defined by at least 2
3
treated exacerbations per year and is associated with a huge impact on patient health.
4
However, existence of this phenotype and corresponding threshold still need to be formally
5
confirmed by statistical methods analysing exacerbations profiles with no specific a priori
6
hypothesis.
7
Objective: To confirm the existence of the frequent exacerbator phenotype with an
8
innovative unbiased statistical analysis of prospectively recorded exacerbations.
9
Methods: Data of COPD patients from the French cohort EXACO were analysed using the
10
KmL method designed to cluster longitudinal data and ROC curve analysis to determine the
11
best threshold to allocate patients to identified clusters. Univariate and multivariate analyses
12
were performed to study characteristics associated with different clusters.
13
Results: Two clusters of patients were identified based on exacerbation frequency over time
14
with 2.89 exacerbations per year on average in the first cluster (n=348) and 0.71 in the second
15
(n=116). The best threshold to distinguish these clusters was 2 moderate to severe
16
exacerbations per year. Frequent exacerbators had more airflow limitation, symptoms and
17
health related quality of life impairment. A simple clinical score was derived to help
18
identifying patients at risk of exacerbations.
19
Conclusions: These analyses confirmed the existence and clinical relevance of a frequent
20
exacerbator subgroup of COPD patients, and the currently used threshold to define this
21
phenotype.
AC C
EP
TE D
M AN U
SC
RI PT
1
22 23
Keywords: Chronic obstructive pulmonary disease, exacerbation, cohort studies
24
1
ACCEPTED MANUSCRIPT INTRODUCTION Chronic obstructive pulmonary disease (COPD) is a chronic inflammatory disease of the airways characterized by an airflow limitation that is not fully reversible.1 Course of the disease is punctuated by acute exacerbations associated with high morbidity, mortality and
RI PT
costs.2,3 Exacerbations become more frequent and more severe as COPD progresses4 and their recurrence is associated with the decline in lung function and health status of patients.5,6 Exacerbations can occur across all stages of airflow limitation measured by forced expiratory
SC
volume in first second (FEV1), which emphasizes the need to identify other predictors of high exacerbation risk.4
M AN U
During the last years, authors have attempted to characterize a specific phenotype of COPD patients with frequent exacerbations, as identified using various thresholds mostly derived from the median exacerbation frequency in various cohorts, with subsequent confirmation of their association with health status and prognosis.7 In the ECLIPSE study, about 60% of
TE D
patients with 2 or more moderate to severe exacerbations during one year also had at least 2 exacerbations during the following year.4 In these patients, exacerbation frequency was considered relatively stable over the 3-year study period. Consequently, most authors have
EP
been using this cut-off to describe the frequent exacerbator phenotype.4,8 This definition is
AC C
now part of the GOLD (Global initiative for chronic Obstructive Lung Disease) guidelines.1 However, intra-subject variability of annual exacerbation frequency limits the ability of this criteria to predict the future risk of exacerbations at an individual level.4,9 The first objective of this study was to determine, using a hypothesis free analysis, whether patients cluster together based on prospectively recorded frequencies of moderate to severe exacerbations over 4 years of follow-up. The subsequent main objective was to define the best threshold(s) to allocate patients to these clusters. Secondary objectives were to describe
2
ACCEPTED MANUSCRIPT clinical characteristics associated with these clusters and to determine whether a simple score
AC C
EP
TE D
M AN U
SC
RI PT
can be derived to identify patients at risk of frequent exacerbations.
3
ACCEPTED MANUSCRIPT METHODS Study design and patients The EXACO (EXAcerbations of COPD patients) study is a prospective cohort study
RI PT
conducted in France and approved by the Institutional Review Board of the French-language Society of respiratory medicine (CEPRO 2012-026). One hundred and thirty-two respiratory physicians, covering office-based and hospital settings from the private and public sectors,
SC
included 835 consecutive COPD patients from October 2005 to January 2007 and followed them during four years. Inclusion criteria were a ≥2 GOLD stage of airflow limitation (post-
M AN U
bronchodilator FEV1/FVC <70% and FEV1 ≤80% of predicted), age ≥40 years, a current or past smoking history ≥15 pack-year, and no exacerbation in the month preceding enrolment. Patients with other respiratory diseases, diagnosed with a cancer in the preceding three years, or unable to complete the follow-up requirements were excluded. Selection and characteristics
Exacerbations
TE D
of physicians and data collected are detailed in e-Appendix.
EP
Exacerbation was identified as ≥2 consecutive days with sustained worsening of patient’s symptoms beyond day-to-day variations leading to a change in treatment (details in e-
AC C
Appendix).10 The severity of exacerbations has been classified according to health care use: mild in case of self-management, moderate if the patient was not hospitalized but received a prescription of systemic corticosteroids and/or antibiotics and severe if the patient has been hospitalized. Only moderate to severe exacerbations were taken into account and pooled for analysis. Different collecting tools were organized to avoid under-reporting. First, patients were asked to complete every month an auto-questionnaire for respiratory symptoms and regarding their respiratory status. In the event of deterioration, patients were asked to complete immediately 4
ACCEPTED MANUSCRIPT an additional questionnaire. Second, patients were contacted by telephone every three months to ensure that all events were collected. Finally, at each follow-up visit, respiratory physicians asked the patient to report the number of COPD exacerbations requiring a change in treatment
avoid double counting a single exacerbation. Data analysis
RI PT
or hospitalization since the last consultation and recorded it. Dates of symptoms were used to
SC
Two sets of analyses were performed. The main analysis used data from all patients with complete information on annual exacerbations during each of the 4 years of follow-up, to
M AN U
ensure maximal longitudinal robustness (Figure 1). The second, called “extended analysis”, was a confirmatory analysis using data from all patients with at least two years of follow-up. We used KmL analysis which is an implementation of a non-parametric algorithm designed to work specifically on longitudinal data (details in e-Appendix).11 It provides scope for dealing
TE D
with missing values and runs the algorithm several times, varying the starting conditions and/or the number of clusters sought; thereby it identifies clusters of patients with similar evolution of exacerbation frequency during the 4 years. To ensure maximal robustness of
EP
findings, three different indexes were used to choose the optimal number of clusters (Figure 2A): the Calinski & Harabatz index, the Ray & Turi index and the Davies & Bouldin index.11
AC C
Higher indexes correspond to larger between-cluster variances and smaller within-cluster variances. ROC curve analysis was performed to identify the best threshold to separate clusters.
Data are reported as means ± standard deviation (SD) or percentages as appropriate. Comparisons of quantitative data between groups were carried out using a Student’s t-test after a F test for quantitative data and a Fisher’s exact test was performed for qualitative data. Factors associated with the frequent exacerbator cluster were assessed by iteratively reweighted least squares method for univariate analysis and variables with p < 0.1 were 5
ACCEPTED MANUSCRIPT included in a bidirectional stepwise multivariate regression. A score was derived from the results of multivariate logistic regression to predict belonging to the frequent exacerbator cluster and its accuracy was tested using a ROC curve analysis. Data were analysed with R software version 3.3.2 using packages kml version 2.4.1, pROC version 1.8 and dplyr version
AC C
EP
TE D
M AN U
SC
RI PT
0.5.0.11,12
6
ACCEPTED MANUSCRIPT RESULTS Study and analysis populations Eight hundred thirty-five COPD patients were included in the cohort (Figure 1). They were
RI PT
mostly former smoker males with symptomatic COPD and moderate to severe airflow obstruction associated with frequent comorbidities (Table 1). Two hundred and fifty patients were lost to follow-up before the end of the study and 121 patients died, mainly of respiratory
SC
causes and cancers (Figure 1 and e-Table 1). Therefore, complete information on annual exacerbation number during the 4 years of the study was available for 464 patients. Patients
M AN U
lost to follow-up were not significantly different from patients who completed the study whereas patients who died were older and exhibited more severe airflow limitation (e-Table 2). The extended analysis was performed in 608 patients with at least two years of follow-up (Table 1). Patients followed at least one year and subsequently lost to follow-up or dead
TE D
reported significantly more exacerbations during the first year of the study (2.42 ± 1.9 and 2.4 ± 1.7, respectively) than patients who completed the study (1.55 ± 1.9, p<0.001).
EP
Identification of two clusters with homogeneous frequencies of exacerbations According to all criteria, the optimal number of clusters to separate patients in groups with
AC C
homogeneous numbers of moderate to severe exacerbations over time was two (Figure 2A). The first cluster (A) consisted of 348 patients (75%) who experienced few exacerbations (0.71 ± 0.54 exacerbations/patient/year) and the second (B) comprised 116 patients (25%) who experienced frequent exacerbations (2.89 ± 1.07 exacerbations/patient/year) over the four years of the study (Figure 2B and e-Figure 1). The ROC curve analysis found a good accuracy with an area under the curve (AUC) of 0.91 (CI: 0.88-0.94) and a best threshold to maximize (sensitivity + specificity) of at least 2 exacerbations in the first year (sensitivity 85% and specificity 79%) (e-Figure 2). The extended analysis found similar results with 451 patients
7
ACCEPTED MANUSCRIPT allocated to cluster A (74.2%) with few exacerbations (0.89 ± 0.61 exacerbations/patient/year) and 157 to cluster B (25.8%) with more frequent exacerbations (3.17 ± 1.05 exacerbations/patient/year). In this extended analysis, sensitivity and specificity of the threshold of at least 2 exacerbations in the first year were 91% and 74.4% with an AUC of
RI PT
0.92 (CI: 0.90-0.94). Stability of the number of exacerbations over time
SC
Among patients with full 4-year data, only 7% had at least 2 exacerbations per year and 36% less than 2 exacerbations per year during each of the whole four years of follow-up (Figure 3).
M AN U
Among the 171 patients with at least 2 exacerbations during the first year of the study, 47%, 31% and 19% persisted to have 2 or more exacerbations during the following one, two or three years, respectively. Similarly, among the 293 patients with less than 2 exacerbations during the first year of the study, 79%, 66% and 57% still had less than 2 exacerbations
TE D
during the first, second and third following years, respectively. Stability over time of annual exacerbation frequency in KmL-defined clusters was fair with intraclass correlation coefficients of 0,57 (CI: 0,50 - 0,64) in cluster A and 0,42 (CI: 0,25 - 0,56) in cluster B
EP
(Figure 2B).
AC C
Factors associated with the frequent exacerbator cluster Frequent exacerbators (cluster B) had more frequent history of severe respiratory infections during infancy, a longer disease duration and more severe airflow limitation (Table 2 and 3). Moreover, they had more respiratory symptoms and limitations of daily activities, a worse health-related quality of life and underwent pulmonary rehabilitation more frequently. Only five factors remained associated with the frequent exacerbator cluster in multivariate logistic regression analysis: a COPD diagnosis made more than 7 years ago, presence of daily sputum production, severe disability associated with breathlessness, severe airflow limitation and
8
ACCEPTED MANUSCRIPT hospitalization for exacerbation during previous years (Table 4). Results were similar in the extended analysis (data not shown). Clinical rule to predict the frequent exacerbator cluster
RI PT
We derived a score called ESOD (Exacerbation history, chronic Sputum production, GOLD stage of Obstruction and mMRC Dyspnoea stage) from results of the multivariate logistic regression analysis (Table 5). The AUC of 0.705 (CI: 0.65-0.75) suggests a fair accuracy of
SC
the score to predict belonging to the frequent exacerbator cluster with a best threshold for this score of ≥2 (e-Figure 3). The mean exacerbation rate and the probability of belonging to the
M AN U
frequent exacerbators cluster increased with the score (e-Table 3). With a score lower than 2, 86.2% of patients who completed the study and 87% of patients of the extended analysis belonged to the non-frequent exacerbator cluster (e-Figure 4, data not shown for extended
AC C
EP
TE D
analysis).
9
ACCEPTED MANUSCRIPT DISCUSSION In this prospective cohort study, KmL method identified 2 clusters of patients based on longitudinal 4-year follow-up of moderate-to-severe exacerbation frequency. This analysis
RI PT
confirmed that two exacerbations per year, the threshold adopted in GOLD guidelines to define the frequent exacerbator phenotype, is the most adequate threshold to distinguish frequent from non-frequent exacerbators.1 Although, a follow-up of a few years is required due to the variability of exacerbation rate over time. Finally, a simple clinical score based on
SC
multivariate analysis predicts the risk to belong to the frequent exacerbator phenotype. These
M AN U
results obtained in patients with a full 4-year follow-up were confirmed in a population extended to those with at least 2 years of follow-up.
During the four years of the study, 121 patients died and 250 were lost to follow-up. Dead patients were older with more severe airway obstruction at inclusion, which is consistent with
TE D
a higher mortality risk.13 In contrast, patients lost to follow-up were similar to patients who completed the four years of the study. Rate of lost patients was in the same range as in other published studies.4,14 Repeating analyses including lost and dead patients with at least two
EP
years of participation obtained similar results, suggesting that results appear applicable to the
AC C
entirety of the population.
The frequent exacerbator phenotype was defined several years ago as COPD patients experiencing a higher frequency of exacerbations associated with poorer outcomes.5,15 The strengths of the present work are (1) the use of the KmL method, which does not presuppose any number of clusters or threshold(s) and (2) the confirmation by three different statistical criteria that, in this population, the optimal number of clusters to describe the population is two. To our knowledge, this is the first study to confirm the existence of a specific phenotype with frequent exacerbations opposed to patients with few or no exacerbations using a
10
ACCEPTED MANUSCRIPT hypothesis-free method. The proportion of frequent exacerbators among COPD patients remains debated. The SPIROMICS cohort study reported a markedly lower proportion of stable frequent exacerbators over three years (2%) compared to ECLIPSE cohort (12%) and our cohort (11%).4,9 This may reflect inclusion of patients with milder airflow limitation in
RI PT
SPIROMICS (mean FEV1 63% versus 48% and 49%, respectively). Moreover, some studies suggested a third intermediate phenotype called infrequent exacerbators including patients who have one exacerbation per year, a phenotype positioned between frequent exacerbators
SC
and patients without exacerbations.4,9,16–18 Our analysis did not identify this intermediate
M AN U
phenotype, which may relate to an insufficient population size or number of time points. From a clinical perspective, an important challenge is to identify patients at risk of future exacerbations when no sufficient follow-up is available yet, in order to provide appropriate education and prescribe appropriate therapy to prevent exacerbations. In contrast, identifying patients with a low risk of exacerbation may prevent overtreatment. However, due to the
TE D
variability of exacerbation rate over time, this identification is not straightforward. Based on annual exacerbation frequency, one third of patients changed group between first and second
EP
year of our study, which is concordant with other cohorts.4,9 Therefore, tools to predict the frequent exacerbator phenotype are needed in clinical practice.
AC C
Some authors tried to develop models to predict the risk of COPD exacerbations using multivariate analysis but most of those scores have a poor practical applicability.19 Based on our population, a score applicable in clinical practice was developed. Since the reported duration of the disease is subject to recall errors, we decided not to include this variable in the score, although it was significantly associated with the risk of frequent exacerbation in multivariate analysis. Criteria have not been weighted, as it did not change the area under the ROC curve and as we wanted to develop a simple score for clinical practice. Although the performance of the score is lower to predict the risk of having ≥2 exacerbations the following 11
ACCEPTED MANUSCRIPT year, a score lower than 2 accurately predicts a low risk of belonging to the frequent exacerbator cluster. The AUC was in the same range as other established scores.19 Unfortunately, our study was designed before the publication of ECLIPSE study results, therefore the number of exacerbations the previous year was not collected and could not be
AC C
EP
TE D
M AN U
SC
and significantly associated with risk of future exacerbations.
RI PT
included in our model. However, the number of previous severe exacerbations was collected
12
ACCEPTED MANUSCRIPT CONCLUSION For the first time, the existence of a frequent exacerbator phenotype in COPD patients was demonstrated by a statistical method that does not rely on pre-specified hypotheses regarding
RI PT
the final number of clusters or the most reliable threshold(s). Interestingly, this method confirmed the currently used threshold of 2 exacerbations per year to identify these patients. To allow easy identification of at-risk patients in routine care before prolonged follow-up is available, we propose a new simple score to predict the risk of COPD exacerbations, which
AC C
EP
TE D
M AN U
SC
now needs to be prospectively tested.
13
ACCEPTED MANUSCRIPT ACKNOWLEDGEMENT In memory of Pr. Isabelle Tillie-Leblond who has taken an essential role in the elaboration, coordination, conduct and analysis of this study, and more widely in the field of respiratory medicine in France.
and whose dedication to research made the trial possible.
SC
AUTHOR CONTRIBUTIONS
RI PT
The authors thank respiratory physicians and COPD patients who participated in the study,
Dr. Le Rouzic, Prs Roche and Housset and Ms. Pouriel are guarantors of the manuscript and
M AN U
take responsibility for the integrity of the data and the accuracy of the data analysis. All authors participated to the study design, conduct or data analysis and interpretation, and
AC C
EP
TE D
approved the final version.
14
ACCEPTED MANUSCRIPT REFERENCES Vogelmeier CF, Criner GJ, Martinez FJ, et al. Global Strategy for the Diagnosis, Management, and Prevention of Chronic Obstructive Lung Disease 2017 Report: GOLD Executive Summary. Eur Respir J 2017;49(3):1700214.
2.
Suissa S, Dell’Aniello S, Ernst P. Long-term natural history of chronic obstructive pulmonary disease: severe exacerbations and mortality. Thorax 2012;67(11):957–963.
3.
Punekar YS, Shukla A, Müllerova H. COPD management costs according to the frequency of COPD exacerbations in UK primary care. Int J Chron Obstruct Pulmon Dis 2014;9:65–73.
4.
Hurst JR, Vestbo J, Anzueto A, et al. Susceptibility to exacerbation in chronic obstructive pulmonary disease. N Engl J Med 2010;363(12):1128–1138.
5.
Donaldson G, Seemungal T, Bhowmik A, Wedzicha J. Relationship between exacerbation frequency and lung function decline in chronic obstructive pulmonary disease. Thorax 2002;57(10):847–852.
6.
Vestbo J, Edwards LD, Scanlon PD, et al. Changes in Forced Expiratory Volume in 1 Second over Time in COPD. N Engl J Med 2011;365:1184–1192.
7.
Soler-Cataluña JJ, Rodriguez-Roisin R. Frequent chronic obstructive pulmonary disease exacerbators: how much real, how much fictitious? COPD 2010;7(4):276–284.
8.
McGarvey L, Lee AJ, Roberts J, Gruffydd-Jones K, McKnight E, Haughney J. Characterisation of the frequent exacerbator phenotype in COPD patients in a large UK primary care population. Respir Med 2015;109(2):228–237.
9.
Han MK, Quibrera PM, Carretta EE, et al. Frequency of exacerbations in patients with chronic obstructive pulmonary disease: an analysis of the SPIROMICS cohort. Lancet Respir Med 2017;5(8):619–626.
EP
TE D
M AN U
SC
RI PT
1.
AC C
10. Rodriguez-Roisin R. Toward a consensus definition for COPD exacerbations. Chest 2000;117(5 Suppl 2):398S–401S. 11. Genolini C, Alacoque X, Sentenac M, Arnaud C. kml and kml3d: R Packages to Cluster Longitudinal Data. J Stat Softw 2015;65(4):1–34. 12. Robin X, Turck N, Hainard A, et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics 2011;12:77. 13. Jensen HH, Godtfredsen NS, Lange P, Vestbo J. Potential misclassification of causes of death from COPD. Eur Respir J 2006;28(4):781–785. 14. Calverley PMA, Anderson JA, Celli B, et al. Salmeterol and Fluticasone Propionate and Survival in Chronic Obstructive Pulmonary Disease. N Engl J Med 2007;356(8):775– 789.
15
ACCEPTED MANUSCRIPT 15. Soler-Cataluña JJ, Martínez-García MÁ, Sánchez PR, Salcedo E, Navarro M, Ochando R. Severe acute exacerbations and mortality in patients with chronic obstructive pulmonary disease. Thorax 2005;60(11):925–931. 16. Beeh KM, Glaab T, Stowasser S, et al. Characterisation of exacerbation risk and exacerbator phenotypes in the POET-COPD trial. Respir Res 2013;14:116.
RI PT
17. Yang H, Xiang P, Zhang E, et al. Predictors of exacerbation frequency in chronic obstructive pulmonary disease. Eur J Med Res 2014;19:18.
18. Tomioka R, Kawayama T, Suetomo M, et al. “Frequent exacerbator” is a phenotype of poor prognosis in Japanese patients with chronic obstructive pulmonary disease. Int J Chron Obstruct Pulmon Dis 2016;11:207–216.
AC C
EP
TE D
M AN U
SC
19. Guerra B, Gaveikaite V, Bianchi C, Puhan MA. Prediction models for exacerbations in patients with COPD. Eur Respir Rev 2017;26(143):160061.
16
ACCEPTED MANUSCRIPT FIGURE LEGENDS Figure 1: Flow diagram of participants through each stage of the study.
SC
RI PT
Figure 2: KmL clustering analysis of prospectively recorded annual exacerbation frequency. A Criteria applied to choose the optimal number of clusters including (1-black) the Calinski & Harabatz index, (2-red) the Ray & Turi index and (3-green) the inverse of the Davies & Bouldin index. These criteria are normalized to be mapped together using kml package in R. Scores range from 0 to 1 for each tested number of clusters ranging from 2 to 6. The highest score corresponds to the most appropriate number of clusters. B Individual trajectories of annual exacerbation frequency during each year of the study. The red colour corresponds to cluster A or non-frequent exacerbators (75% of patients) and the blue colour to cluster B or frequent exacerbators (25% of patients). Thick lines labelled with the cluster letter represent the mean exacerbation frequency for each cluster.
AC C
EP
TE D
M AN U
Figure 3: Stability of the frequent-exacerbators phenotype (defined as ≥2 exacerbations per year) with KmL clusters included. The bars show the proportions of patients with no exacerbation (purple) or two or more exacerbations (red) in each years 1, 2, 3 and 4 from left to right, respectively. Each bar is subdivided in light colour for patients belonging to KmL cluster A (non-frequent exacerbators) and in dark colour for patients belonging to cluster B (frequent exacerbators). The numbers on the right denote the numbers of patients belonging to the B and A clusters for each trajectory.
17
ACCEPTED MANUSCRIPT TABLES
Table 1: Demographics and baseline characteristics of patients
mean ± SD % mean ± SD
66.3 ± 9.6 80,4 25.9 ± 5.2
66.5 ± 9.2 79,6 25.9 ± 5.2
% % %
14.3 73.5 12.1
Smoking status Active smoker % 24,7 Ex-smoker % 75,3 Smoking history (pack-years) mean ± SD 44.4 ± 19.8
AC C
FEV1 (% of predicted) FVC (% of predicted)
12.8 74.7 12.6
24.0 76.0 44.5 ± 20.1
22.8 77.2 44.4 ± 19.4
8.7 ± 7.2
9.0 ± 7.4
8.7 ± 7.0
% % %
42.6 32.8 24.6
43.6 33.1 23.4
45.7 33.2 21.1
47.9 ± 15.4 72.0 ± 18.8
48.7 ± 15.3 72.2 ± 18.3
EP
GOLD stage 2 3 4
12.6 74.7 12.6
mean ± SD
TE D
Time since COPD diagnosis (years)
66.1 ± 8.9 79,3 26.0 ± 5.1
SC
Occupational status Working Retired Unemployed
Completed (n=464)
RI PT
Extended (n=608)
M AN U
Age (year) Male sex BMI (kg/m2)
Full cohort (n=835)
mean ± SD 47.7 ± 15.6 mean ± SD 71.8 ± 18.9
This table summarizes demographic and baseline characteristics of all included patients (Full cohort), patients analysed in the extended analysis (Extended) and the 464 patients who completed the 4-year follow-up. All measures were recorded at screening. BMI: body-mass index, COPD: chronic obstructive pulmonary disease, GOLD: global initiative for chronic obstructive lung disease, FEV1: forced expiratory volume in the first second, FVC: forced vital capacity.
18
ACCEPTED MANUSCRIPT
B (n=116)
mean ± SD % mean ± SD
66.4 ± 8.8 80.2 26.3 ± 5.1
65.3 ± 9.0 76.7 25.3 ± 5.2
NS NS NS
% % %
13.3 75.3 11.3
NS
Smoking status Active smoker Ex-smoker Smoking history (pack-years)
10.5 73.0 16.4
M AN U
Occupational status Working Retired Unemployed
p-value A vs. B
RI PT
Age (year) Male sex BMI (kg/m2)
A (n=348)
SC
Table 2: Comparisons of demographics and baseline characteristics of patients who completed the 4-year follow-up separated in cluster A (non-frequent exacerbators) and B (frequent exacerbators)
% % mean ± SD
23.0 77.0 44.8 ± 19
22.4 77.6 43.4 ± 20.5
NS
mean ± SD
8.1 ± 6.8
10.3 ± 7.3
< 0.01
% % %
50.6 31.0 18.4
31.0 39.7 29.3
< 0.001
mean ± SD mean ± SD
50.3 ± 15.2 73.7 ± 18.1
43.9 ± 14.3 67.7 ± 18.0
< 0.001 < 0.01
Symptoms Daily sputum production Daily wheezing Daily coughing
% % %
52.9 27.6 61.5
70.7 38.8 75.0
< 0.001 < 0.05 < 0.05
mMRC score 0 1 2 3 4
% % % % %
12.6 44.0 31.6 10.9 0.9
5.2 37.9 25.0 30.2 1.7
< 0.001
TE D
Time since COPD diagnosis (years)
EP
GOLD stage 2 3 4
AC C
FEV1 (% of predicted) FVC (% of predicted)
NS
19
ACCEPTED MANUSCRIPT Daily activities limitations No impairment Mild impairment Moderate impairment Severe impairment Complete impairment
8.7 24.3 30.4 35.7 0.9
< 0.01
mean ± SD
46.2 ± 15.7
42.6 ± 15.5
< 0.05
% % %
15.2 16.6 7.6
26.3 27.6 4.3
% % % % % % % % % % % %
31.1 18.0 15.7 10,1 12.0 13.1 11.5 12.6 9.8 5.5 7.2 4.1
% %
4.7 3.5
TE D
Comorbidities Hypertension Osteoarthritis Ischemic heart disease Chronic alcohol consumption Depression Heart rythm disorder Lower limb arteriopathy Diabetes Sleep apnoea Clinical right heart failure Left heart failure Severe respiratory infection in infancy
< 0.05 < 0.05 NS
28.7 27.0 13.9 15.7 7.9 10.3 7.8 6.9 8.8 6.9 4.3 11.2
NS < 0.05 NS NS NS NS NS NS NS NS NS < 0.01
6.0 5.3
NS NS
EP
Pulmonary hypertension Stroke
SC
Treatments Respiratory rehabilitation Long-term oxygen therapy Non invasive ventilation
RI PT
15.4 27.5 38.8 16.9 1.5
M AN U
VSRQ global score
% % % % %
AC C
This table compares demographic and baseline characteristics of the two clusters obtains from the 464 patients who completed the 4-year follow-up, non-frequent exacerbators cluster (A) and frequent exacerbators cluster (B). All measures were recorded at screening. Comparisons between clusters A and B were performed using t-test or Fisher’s exact test as appropriate. p<0.05 was considered significant. BMI: body-mass index, COPD: chronic obstructive pulmonary disease, GOLD: global initiative for chronic obstructive lung disease, FEV1: forced expiratory volume in the first second, FVC: forced vital capacity, mMRC: modified Medical Research Council dyspnoea score, VSRQ global score: visual simplified respiratory questionnaire global score ranging from 80 (best health status) to 0 (poorest health status). NS: not significant.
20
ACCEPTED MANUSCRIPT Table 3: Univariate regression analysis of associations with belonging to the frequent exacerbator cluster (B) in patients with full 4-year follow-up Odds Ratio
Variables Demographic data
FVC
Symptoms Daily sputum production yes vs. no Daily wheezing yes vs. no Daily coughing yes vs. no mMRC dyspnoea score 1-2 vs. 0 3-4 vs. 0
TE D
Patient-reported outcomes Daily activities limitations Mild to moderate impairment Severe to complete impairment
VSRQ global score
per decrease of 4 points
2.33 2.57 1.64
[1.42 - 3.87] 0.001 [1.49 - 4.46] < 0.001 [1.26 - 2.14] < 0.001
1.15
[1.07 - 1.24] < 0.001
1.09
[1.03 - 1.17]
< 0.01
2.15 1.64 1.84 2.03 6.62
[1.38 - 3.41] 0.001 [1.05 - 2.54] 0.03 [1.16 - 2.99] 0.01 [0.9 - 5.49] NS [2.69 - 18.9] < 0.001
0.68
[0.31 - 1.37]
2.41 1.06
[1.49 - 3.9] < 0.001 [1.01 - 1.13] 0.04
1.91 0.54
[1.16 - 3.13] [0.18 - 1.34]
NS
0.01 NS
AC C
EP
Treatments Long-term oxygen therapy yes vs. no Non invasive ventilation yes vs. no
[0.69 - 1.11] NS [0.73 - 2.01] NS [0.92 - 1.00] NS [0.58 - 1.58] NS [1.06 - 1.43] < 0.01 [1.57 - 3.82] < 0.001
M AN U
per increase to next stage per 5% decrease in % predicted value per 5% decrease in % predicted value
FEV1
0.87 1.22 0.96 0.97 1.23 2.45
SC
4 vs. 2
p-value
RI PT
Age per 10-year increase Sex female vs. male BMI per increase of 1 point Smoking status current vs. former smoker Duration of COPD per 5-year increase Hospitalization previous year yes vs. no Lung function GOLD stage 3 vs. 2
95% CI
BMI: body-mass index, COPD: chronic obstructive pulmonary disease, GOLD: global initiative for chronic obstructive lung disease, FEV1: forced expiratory volume in the first second, FVC: forced vital capacity, mMRC: modified Medical Research Council dyspnoea score, VSRQ global score: visual simplified respiratory questionnaire global score. NS: not significant.
21
ACCEPTED MANUSCRIPT Table 4: Factors associated with the frequent exacerbator cluster (B) in the multivariate model in patients with full 4-year follow-up p-value
[1.06 - 2.79]
0.03
2.05
[1.25 - 3.42]
0.005
1.66 4.53 1.77
[0.67 - 5.1] [1.63 - 14.8] [1.08 - 2.93]
NS 0.006 0.03
1.99
[1.21 - 3.26]
RI PT
95% CI
SC
Time since COPD diagnosis (reference: less than 7 years) Daily sputum production (reference: no) Dyspnoea score (mMRC) (reference: mMRC 0) - mMRC 1 and 2 - mMRC 3 and 4 FEV1 (% of predicted) (reference: more than 50%) Hospitalization previous year (reference: none)
Odds Ratio 1.71
0.006
M AN U
Variables
AC C
EP
TE D
COPD: chronic obstructive pulmonary disease, mMRC: modified Medical Research Council dyspnoea score, FEV1: forced expiratory volume in the first second, 95%CI: 95% confidence interval. NS: not significant.
22
ACCEPTED MANUSCRIPT Table 5: ESOD score Points
Exacerbation (any hospitalization during previous 2 years) Sputum (chronic daily production)
Obstruction (%FEV1) Dyspnoea (mMRC score)
0
1
No
Yes
No ≥ 50% 0-1-2
Yes < 50% 3-4
RI PT
ESOD score
AC C
EP
TE D
M AN U
SC
The ESOD score is the sum of points related to each variable and ranges from 0 to 4. mMRC: modified Medical Research Council dyspnoea score, %FEV1: percent predicted forced expiratory volume in the first second.
23
ACCEPTED MANUSCRIPT
French Database n = 2511 respiratory physicians
RI PT
n=132 randomised respiratory physicians
SC
n=835 included patients
M AN U
n=37 lost to follow-up n=38 deaths
n=760 patients followed 1 year
n=675 patients followed 2 years
AC C
EP
n=608 patients with at least two years with full data
TE D
“Extended analysis”
n=53 lost to follow-up n=32 deaths
n=55 lost to follow-up n=25 deaths
n=595 patients followed 3 years
n=105 lost to follow-up n=26 deaths
n=464 patients followed 4 years
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT 0
20
43 (n=1 40
60
80
57 (n=1
47% (n=81)
0
20
40
RI PT
53% (n=90)
60
M AN U
68% (n=61)
EP
69% (n=2
20
20 (n=1 40
60
80
80 (n=4
TE D
0
70
31 (n=
SC
32% (n=29)
AC C
0
35% (n=28)
75 (n=1
25 (n= 32% (n=20)
68% (n=42)
29 (n=1 0
20
40
60
80
71 (n=3
21% (n=62)
79% (n=231)
0
20
40
60
80
100
32 (n=1
68
ACCEPTED MANUSCRIPT
e-Appendix 1 - SUPPLEMENTAL METHODS Physicians’ recruitment
RI PT
Physicians were selected at random from a complete database of 2511 French respiratory physicians covering office-based and hospital settings from the private and public sectors. An information letter was sent and they were phoned to enquire if they wanted to participate. If a physician refused, the next physician on the list was contacted until the predefined number of centers was reached. Altogether, 1202 respiratory physicians were
SC
asked to participate, 276 accepted and 132 actually included patients in the cohort. Characteristics of these active investigators were similar to that of the total French population of respiratory physicians except for sex (77% of males versus 68% in the total
M AN U
population of French respiratory physicians) and activity (46% of hospital-based physicians versus 22% in the total population of French respiratory physicians), and covered all French territory. Physicians had to include the 10 first patients who fulfilled all required conditions and accepted to participate. The mean number of patients actually recruited was 6.3 ± 4 per physician.
Physicians
completed
TE D
Data collection by physicians a
case-report
form
during
medical
consultations
recording
sociodemographic data (age, sex, marital status, education level, last profession, lodging type, urban/rural residence), history of occupational exposure to smoke, dust or gases, family history of COPD, height and weight, smoking status and history, comorbid medical
EP
conditions including cardiovascular disease and diabetes mellitus and other symptoms or syndromes potentially relevant to patients’ respiratory condition (atopy, gastro-oesophageal
AC C
reflux, nasal drip and sinusitis). A questionnaire asked details about patients’ COPD history (time since onset of breathlessness, cough, expectoration and time since COPD diagnosis, number of hospitalizations for COPD exacerbations the previous 2 years) and its clinical expression (presence of daily cough, expectoration, wheezing, and dyspnoea). Physicians also recorded current treatments for COPD, influenza and pneumococcal vaccinations and selected
non-pharmacological
management
of
COPD
(physiotherapy,
respiratory
rehabilitation programs, home oxygen, non-invasive nocturnal ventilation). Patients with ischemic heart disease, arrhythmias, left heart failure, peripheral artery disease, or with a previous cerebrovascular accident were defined as having cardiovascular disease. Patients reporting both daily cough and daily expectoration were defined as having a chronic bronchitis profile.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
Patients were asked to complete the modified Medical Research Council (mMRC) Dyspnoea Scale (Bestall Thorax 1999) and the Borg score (Silverman Am Rev Respir Dis 1988) as measures of breathlessness, the Visual Simplified Respiratory Questionnaire (VSRQ) (Perez
RI PT
Int J COPD 2009) as a measure of health status, and to classify their limitation in daily activities because of respiratory symptoms as absent, mild, moderate, severe or complete. Results of the most recent lung function tests, sputum microbiology, oxygen saturation and arterial blood gases were recorded. The mMRC score ranges from 0 (not troubled with breathlessness except with strenuous exercise) to 4 (too breathless to leave the house or status) to 0 (poorest health status).
M AN U
Monitoring of symptoms and exacerbations
SC
breathless when dressing or undressing) and the VSRQ global score from 80 (best health
Patients were asked to complete an auto-questionnaire for each respiratory symptom lasting more than 2 days and every month regarding their respiratory status. This questionnaire collected dates of start and end of these symptoms, apparition or increase of clinical symptoms (sputum volume and aspect, breathlessness, cough, fever, thoracic pain, wheezing and cold), dyspnoea auto-evaluation with Borg scale, decrease of daily activity, therapeutic modification, medical consultation, hospitalization and work stopping. Patients
TE D
were also contacted by telephone every three months by the Kappa Santé Society (Paris, France) to ensure this information was complete. In addition to patients’ direct reports of exacerbations, respiratory physicians recorded the number of COPD exacerbations requiring a change in treatment or hospitalization since the last consultation at each follow-up visit,
EP
(one to two visits per year).
In the event of deterioration, patients had to complete an additional questionnaire described
AC C
in the previous chapter. This questionnaire asked about the presence of eight symptoms (sputum volume and purulence, common cold, breathlessness, cough, fever, chest pain, wheezing) and included the Borg score as a measure of dyspnoea and a scale measuring limitation in daily activities. Dates of start and end of these symptoms were requested. Patients also reported if they had changed their own medication because of the worsened symptoms and if they had consulted a doctor, presented to a hospital emergency department, or been hospitalized in association with the episode, and when their respiratory state returned to normal. These data were collected for each exacerbation. KmL method KmL is a nonparametric clustering method to identify homogeneous trajectories in longitudinal data (Genolini and Falissard, 2011). This method was used to identify patients Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
with homogeneous yearly exacerbation rate over the 4 years of the study and to explore the existence of a frequent exacerbator phenotype based on exacerbation frequency. The advantages of this method are the following: (1) it does not require normality or any
RI PT
parametric assumption within clusters, (2) it does not require any assumption regarding the shape of the trajectory, (3) it is able to deal with missing values and (4) it does not require any assumption regarding the number of cluster. After running the algorithm, KmL most frequently uses the Calinski & Habaratz criterion (Calinski and Harabatz, 1974) to objectively choose the optimal number of clusters; two other criteria were applied to check
SC
consistency: the Ray & Turi criteria (Ray and Turi, 1999) and the Davies & Bouldin criteria (Davies and Bouldin, 1979). These non-parametric indices reflect both between-cluster and within-cluster variances. The best solution maximizes the between-cluster variance,
M AN U
meaning that trajectories of patients of different clusters are well-separated, and minimizes within-cluster variance, meaning that trajectories of patients within the same cluster are close to each other giving a compact cluster. For this study, KmL has been parameterized to search trajectories with up to 6 clusters and run the algorithm 20 times for each number of clusters. The three criteria were normalized to be mapped together (Figure 2A). The scores range from 0 to 1 for each tested number of clusters and the highest score corresponds to
AC C
EP
TE D
the most appropriate number of clusters.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
e-Table 1: Causes of death during follow-up Dead patients
Causes of death Respiratory failure
Cardiovascular
Respiratory infection
5
Pneumothorax
4
Pulmonary embolism
1
Myocardial ischemia
3
SC
Respiratory
29
Arteriosclerosis
2
Cardiac arrhythmia
2
Colorectal neoplasms Kidney neoplasms Neoplasms
Pancreatic neoplasm Oesophageal neoplasm Brain neoplasm
19 2 2 1 1 1
Otorhinolaryngologic neoplasm
1
Suicide
1
TE D
Others
1
M AN U
Heart arrest Lung neoplasms
Gastrointestinal haemorrhage
1
Cerebral haemorrhage
1 44
AC C
EP
Not defined
RI PT
(n=121)
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
e-Table 2: Baseline comparison with lost to follow-up
66.1 ± 8.9
65 ± 10.6
%
79.3
79.2
mean ± SD
26 ± 5.1
26 ± 5.7
2
%
46.8
3
%
31.4
4
%
21.8
mean ± SD
48.7 ± 15.3
Male sex BMI (kg/m2) GOLD stage
FEV1 (% predicted)
mean ± SD
45.2 ± 15.7
Previous severe exacerbation
mean ± SD
0.45 ± 0.9
2
25.2 ± 4.7
NS
1
1
p-value 1 vs 3
p-value 2 vs 3
NS
***
***
< 0.001
2
NS
***
***
41.2 ± 15.9
< 0.001
1
NS
***
***
40.0 ± 17.6
0.01
1
NS
***
NS
0.75 ± 1.2
0.01
1
NS
NS
NS
29.0
26.3
23.7
50.0
49.1 ± 15.5
0.70 ± 2.1
0.02 NS
23.7
44.6 ± 15.8
p-value p-value 3 groups 1 vs 2
86.8
47.3
TE D
VSRQ global score
69.6 ± 9.1
M AN U
mean ± SD
RI PT
Age (year)
Lost to Dead patients follow-up (N=121) - 3 (N=250) - 2
SC
Followed patients (N=464) - 1
AC C
EP
BMI: body-mass index, GOLD: global initiative for chronic obstructive lung disease, FEV1: forced expiratory volume in the first second, VSRQ global score: visual simplified respiratory questionnaire global score ranging from 80 (best health status) to 0 (poorest health status). 1 one-way analysis of variance (ANOVA) 2 chi-squared test. Post-hoc comparisons between each column were performed using ttest with Bonferroni adjustment. NS: not significant, *** = p < 0.001.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
e-Table 3: Mean exacerbation rate and probability to belong to the frequent
RI PT
exacerbator cluster (B) for each ESOD score
ESOD score 0
1
2
3
4
75
151
125
81
13
0.71
1.03
1.36
1.87
2.67
8
17.2
28
46.9
69.2
Mean annual exacerbation rate Probability to belong to B cluster Extended N Mean annual exacerbation rate
87
189
164
120
24
0.88
1.20
1.56
2.14
2.37
10.3
14.3
29.3
47.5
54.2
TE D
Probability to belong to B cluster
M AN U
N
SC
Completed
This table summarizes mean exacerbation rate and probability to belong to the frequent exacerbator cluster (B) for each ESOD score. Results are given for the patients who completed the 4-year follow-up (464) and patients analysed in the extended analysis
AC C
EP
(Extended). N: number of patients for each ESOD score.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
TE D
M AN U
SC
RI PT
AC C
EP
e-Figure 1: Overlap of exacerbations frequency for each year of the study in the 2 clusters. y-axis are numbers of exacerbations for each year of the study for clusters A (red) and B (blue). The white point is the median, the bold black lines represent the interquartile range (25th to 75th centiles) and the thin black lines represent the whiskers extended to the more extreme data point which is no more than 1.5 times the interquartile range from the box. The width represents the density for each value.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
Threshold performances Se = 99.1 % Sp = 47.7 % PPV = 38.7 % NPV = 99.4 %
RI PT
>0
M AN U
SC
>1
>2
Se = 68.1 % Sp = 93.4 % PPV = 77.5 % NPV = 89.8 %
>3
Se = 50.9 % Sp = 97.7 % PPV = 88.1 % NPV = 85.6 %
TE D
AUC = 0.91 (0.88 - 0.94)
Se = 85.3 % Sp = 79.3 % PPV = 57.9 % NPV = 94.2 %
AC C
EP
e-Figure 2: Receiver operating characteristic (ROC) curve illustrating the performances of using the number of exacerbations the first year of the study to identify the KmL-defined frequent exacerbator cluster. The grey area around the ROC curve represents 95% confidence intervals. Test performances are given for 0 to 3 moderate to severe exacerbations the first year of the study. AUC: area under the curve, Se: sensitivity, Sp: specificity, PPV: positive predictive value, NPV negative predictive value.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
Threshold performances Se = 94.8 % Sp = 21.2 % PPV = 28.6 % NPV = 94.4 %
RI PT
>0
M AN U
SC
>1
Se = 71.6 % Sp = 59.2 % PPV = 36.9 % NPV = 86.2 %
>2
Se = 40.5 % Sp = 86.2 % PPV = 49.5 % NPV = 81.3 %
>3
Se = 7.8 % Sp = 98.9 % PPV = 69.2 % NPV = 76.3 %
TE D
AUC = 0.705 (0.65 - 0.75)
AC C
EP
e-Figure 3: Receiver operating characteristic (ROC) curve illustrating the performances of the ESOD score to positively predict belonging to cluster B (frequent exacerbators according to KmL). The grey area around the ROC curve represents 95% confidence intervals. Test performances are given for each threshold of the score. AUC: area under the curve, Se: sensitivity, Sp: specificity, PPV: positive predictive value, NPV negative predictive value.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
Threshold performances Se = 92.4 % Sp = 22.9 % PPV = 41.1 % NPV = 83.8 %
RI PT
>0
SC
>1
Se = 59.6 % Sp = 58.0 % PPV = 45.3 % NPV = 71.1 %
>2
Se = 30.4 % Sp = 85.3 % PPV = 54.7 % NPV = 67.8 %
>3
Se = 5.8 % Sp = 99.0 % PPV = 76.9 % NPV = 64.3 %
TE D
M AN U
AUC = 0.64 (0.58 - 0.68)
AC C
EP
e-Figure 4: Receiver operating characteristic (ROC) curve illustrating the performances of the ESOD score to positively predict the risk of having at least 2 exacerbations in the following year. The grey area around the ROC curve represents 95% confidence intervals. Test performances are given for each threshold of the score. AUC: area under the curve, Se: sensitivity, Sp: specificity, PPV: positive predictive value, NPV: negative predictive value.
171104
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
S0012-3692(17)32903-3
DOI:
10.1016/j.chest.2017.10.009
Reference:
CHEST 1390
To appear in:
CHEST
Received Date: 6 May 2017 Revised Date:
31 August 2017
Accepted Date: 2 October 2017
Please cite this article as: Le Rouzic O, Roche N, Cortot AB, Tillie-Leblond I, Masure F, Perez T, Boucot I, Hamouti L, Ostinelli J, Pribil C, Poutchnine C, Schück S, Pouriel M, Housset B, Defining the « frequent exacerbator » phenotype in COPD: a hypothesis-free approach, CHEST (2017), doi: 10.1016/ j.chest.2017.10.009. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT 1
Total word count (text): 2500 words / 2500
2
Total word count (abstract): 222 / 250
3
Title: Defining the « frequent exacerbator » phenotype in COPD: a hypothesis-free approach
4
Running title: Statistically defined COPD “frequent exacerbator”
RI PT
5
Olivier Le Rouzic, MD, PhD; Nicolas Roche, MD, PhD; Alexis B. Cortot, MD, PhD; Isabelle
7
Tillie-Leblond, MD, PhD: Frédéric Masure, MD; Thierry Perez, MD; Isabelle Boucot, MD;
8
Latifa Hamouti, MD; Juliette Ostinelli, MD; Céline Pribil, MD; Christine Poutchnine, MD;
9
Stéphane Schück, MD; Mathilde Pouriel, Bruno Housset, MD, PhD.
SC
6
10
Author affiliations
12 13
Drs Le Rouzic, Perez, Tillie-Leblond and Pr. Cortot: Univ. Lille, CHU Lille, Department of Respiratory Diseases, Lille, France
14 15
Pr. Roche: AP-HP, Hôpital Cochin, Service Pneumologie, EA2511, Université Paris Descartes, Sorbonne Paris Cité, Paris, France
16 17
Dr. Masure: Department of Respiratory Diseases of Groupe Medical Saint Remi, Reims, France
18
Dr. Boucot: GlaxoSmithKline, Brentford, United Kingdom
19
Dr. Hamouti: Boehringer Ingelheim, Paris, France
20
Dr. Ostinelli: AstraZeneca, Rueil-Malmaison, France
21
Dr. Pribil: GlaxoSmithKline, Marly Le Roi, France
22
Dr. Poutchnine: Pfizer, Paris, France
23
Dr. Schück and Ms Pouriel : Kappa Santé, Paris, France
24 25
Pr. Housset: Centre hospitalier intercommunal de Créteil, Service Pneumologie, UPEC, Université Paris-Est, UMR S955, Créteil, France
TE D
EP
AC C
26
M AN U
11
27
Corresponding author:
28
Olivier Le Rouzic
29
Department of Respiratory Diseases
30
University Hospital of Lille
31
1 bd Jules Leclercq
ACCEPTED MANUSCRIPT 1
59037 Lille, France
2
E-mail: [email protected]
3
Funding information
5
This study was conducted with financial support from AstraZeneca, Boehringer Ingelheim,
6
GlaxoSmithKline, Nycomed and Pfizer.
RI PT
4
7
Conflicts of interest
9
Dr. Le Rouzic reports non-financial support from AstraZeneca, Chiesi, GlaxoSmithKline,
10
Leo, MSD, Mundipharma, Santelys Association and Teva, and personal fees and non-
11
financial support from Boehringer Ingelheim, Lilly and Novartis, outside the submitted work.
12
Pr. Roche reports grants and personal fees from Boehringer Ingelheim, Novartis and Pfizer,
13
and personal fees from Teva, GlaxoSmithKline, AstraZeneca, Chiesi, Mundipharma, Cipla,
14
Sanofi, Sandoz, 3M and Zambon, outside the submitted work.
15
Pr. Cortot reports personal fees and non-financial support from Astra-Zeneca, Novartis and La
16
Roche Hoffmann, grants, personal fees and non-financial support from Boehringer-Ingelheim,
17
and personal fees from Bristol Myers Squibb, MSD and Pfizer, outside the submitted work.
18
Dr. Masure has nothing to disclose.
19
Dr. Perez reports personal fees from Boehringer Ingelheim, Novartis, GlaxoSmithKline,
20
Chiesi and Pierre Fabre, outside the submitted work.
21
Drs Boucot and Pribil report being a full time employee of GlaxoSmithKline.
22
Dr. Hamouti reports being a full time employee of Boehringer Ingelheim and having been an
23
employee of Chiesi and Schering Plough.
24
Dr. Ostinelli reports being a full time employee of AstraZeneca.
25
Dr. Poutchnine reports being a full time employee of Pfizer.
26
Dr. Schück and Ms Pouriel report being a full time employee of Kappa Santé.
27
Pr. Housset reports personal fees from GSK, Novartis, Astra-Zeneca, Boehringer Ingelheim,
28
Chiesi and Pfizer, outside the submitted work.
AC C
EP
TE D
M AN U
SC
8
29 30 31
This article has supplemental online data.
ACCEPTED MANUSCRIPT Abbreviations list
2
AUC
area under the curve
3
BMI
body-mass index
4
CI
confidence interval
5
COPD
chronic obstructive pulmonary disease
6
FEV1
forced expiratory volume in first second
7
FVC
forced vital capacity
8
GOLD
Global initiative for chronic Obstructive Lung Disease
9
mMRC
modified Medical Research Council dyspnoea score
10
NS
not significant
11
SD
standard deviation
12
VSRQ
visual simplified respiratory questionnaire
SC
M AN U
TE D EP AC C
13
RI PT
1
1
ACCEPTED MANUSCRIPT Abstract
2
Background: The COPD frequent exacerbator phenotype is usually defined by at least 2
3
treated exacerbations per year and is associated with a huge impact on patient health.
4
However, existence of this phenotype and corresponding threshold still need to be formally
5
confirmed by statistical methods analysing exacerbations profiles with no specific a priori
6
hypothesis.
7
Objective: To confirm the existence of the frequent exacerbator phenotype with an
8
innovative unbiased statistical analysis of prospectively recorded exacerbations.
9
Methods: Data of COPD patients from the French cohort EXACO were analysed using the
10
KmL method designed to cluster longitudinal data and ROC curve analysis to determine the
11
best threshold to allocate patients to identified clusters. Univariate and multivariate analyses
12
were performed to study characteristics associated with different clusters.
13
Results: Two clusters of patients were identified based on exacerbation frequency over time
14
with 2.89 exacerbations per year on average in the first cluster (n=348) and 0.71 in the second
15
(n=116). The best threshold to distinguish these clusters was 2 moderate to severe
16
exacerbations per year. Frequent exacerbators had more airflow limitation, symptoms and
17
health related quality of life impairment. A simple clinical score was derived to help
18
identifying patients at risk of exacerbations.
19
Conclusions: These analyses confirmed the existence and clinical relevance of a frequent
20
exacerbator subgroup of COPD patients, and the currently used threshold to define this
21
phenotype.
AC C
EP
TE D
M AN U
SC
RI PT
1
22 23
Keywords: Chronic obstructive pulmonary disease, exacerbation, cohort studies
24
1
ACCEPTED MANUSCRIPT INTRODUCTION Chronic obstructive pulmonary disease (COPD) is a chronic inflammatory disease of the airways characterized by an airflow limitation that is not fully reversible.1 Course of the disease is punctuated by acute exacerbations associated with high morbidity, mortality and
RI PT
costs.2,3 Exacerbations become more frequent and more severe as COPD progresses4 and their recurrence is associated with the decline in lung function and health status of patients.5,6 Exacerbations can occur across all stages of airflow limitation measured by forced expiratory
SC
volume in first second (FEV1), which emphasizes the need to identify other predictors of high exacerbation risk.4
M AN U
During the last years, authors have attempted to characterize a specific phenotype of COPD patients with frequent exacerbations, as identified using various thresholds mostly derived from the median exacerbation frequency in various cohorts, with subsequent confirmation of their association with health status and prognosis.7 In the ECLIPSE study, about 60% of
TE D
patients with 2 or more moderate to severe exacerbations during one year also had at least 2 exacerbations during the following year.4 In these patients, exacerbation frequency was considered relatively stable over the 3-year study period. Consequently, most authors have
EP
been using this cut-off to describe the frequent exacerbator phenotype.4,8 This definition is
AC C
now part of the GOLD (Global initiative for chronic Obstructive Lung Disease) guidelines.1 However, intra-subject variability of annual exacerbation frequency limits the ability of this criteria to predict the future risk of exacerbations at an individual level.4,9 The first objective of this study was to determine, using a hypothesis free analysis, whether patients cluster together based on prospectively recorded frequencies of moderate to severe exacerbations over 4 years of follow-up. The subsequent main objective was to define the best threshold(s) to allocate patients to these clusters. Secondary objectives were to describe
2
ACCEPTED MANUSCRIPT clinical characteristics associated with these clusters and to determine whether a simple score
AC C
EP
TE D
M AN U
SC
RI PT
can be derived to identify patients at risk of frequent exacerbations.
3
ACCEPTED MANUSCRIPT METHODS Study design and patients The EXACO (EXAcerbations of COPD patients) study is a prospective cohort study
RI PT
conducted in France and approved by the Institutional Review Board of the French-language Society of respiratory medicine (CEPRO 2012-026). One hundred and thirty-two respiratory physicians, covering office-based and hospital settings from the private and public sectors,
SC
included 835 consecutive COPD patients from October 2005 to January 2007 and followed them during four years. Inclusion criteria were a ≥2 GOLD stage of airflow limitation (post-
M AN U
bronchodilator FEV1/FVC <70% and FEV1 ≤80% of predicted), age ≥40 years, a current or past smoking history ≥15 pack-year, and no exacerbation in the month preceding enrolment. Patients with other respiratory diseases, diagnosed with a cancer in the preceding three years, or unable to complete the follow-up requirements were excluded. Selection and characteristics
Exacerbations
TE D
of physicians and data collected are detailed in e-Appendix.
EP
Exacerbation was identified as ≥2 consecutive days with sustained worsening of patient’s symptoms beyond day-to-day variations leading to a change in treatment (details in e-
AC C
Appendix).10 The severity of exacerbations has been classified according to health care use: mild in case of self-management, moderate if the patient was not hospitalized but received a prescription of systemic corticosteroids and/or antibiotics and severe if the patient has been hospitalized. Only moderate to severe exacerbations were taken into account and pooled for analysis. Different collecting tools were organized to avoid under-reporting. First, patients were asked to complete every month an auto-questionnaire for respiratory symptoms and regarding their respiratory status. In the event of deterioration, patients were asked to complete immediately 4
ACCEPTED MANUSCRIPT an additional questionnaire. Second, patients were contacted by telephone every three months to ensure that all events were collected. Finally, at each follow-up visit, respiratory physicians asked the patient to report the number of COPD exacerbations requiring a change in treatment
avoid double counting a single exacerbation. Data analysis
RI PT
or hospitalization since the last consultation and recorded it. Dates of symptoms were used to
SC
Two sets of analyses were performed. The main analysis used data from all patients with complete information on annual exacerbations during each of the 4 years of follow-up, to
M AN U
ensure maximal longitudinal robustness (Figure 1). The second, called “extended analysis”, was a confirmatory analysis using data from all patients with at least two years of follow-up. We used KmL analysis which is an implementation of a non-parametric algorithm designed to work specifically on longitudinal data (details in e-Appendix).11 It provides scope for dealing
TE D
with missing values and runs the algorithm several times, varying the starting conditions and/or the number of clusters sought; thereby it identifies clusters of patients with similar evolution of exacerbation frequency during the 4 years. To ensure maximal robustness of
EP
findings, three different indexes were used to choose the optimal number of clusters (Figure 2A): the Calinski & Harabatz index, the Ray & Turi index and the Davies & Bouldin index.11
AC C
Higher indexes correspond to larger between-cluster variances and smaller within-cluster variances. ROC curve analysis was performed to identify the best threshold to separate clusters.
Data are reported as means ± standard deviation (SD) or percentages as appropriate. Comparisons of quantitative data between groups were carried out using a Student’s t-test after a F test for quantitative data and a Fisher’s exact test was performed for qualitative data. Factors associated with the frequent exacerbator cluster were assessed by iteratively reweighted least squares method for univariate analysis and variables with p < 0.1 were 5
ACCEPTED MANUSCRIPT included in a bidirectional stepwise multivariate regression. A score was derived from the results of multivariate logistic regression to predict belonging to the frequent exacerbator cluster and its accuracy was tested using a ROC curve analysis. Data were analysed with R software version 3.3.2 using packages kml version 2.4.1, pROC version 1.8 and dplyr version
AC C
EP
TE D
M AN U
SC
RI PT
0.5.0.11,12
6
ACCEPTED MANUSCRIPT RESULTS Study and analysis populations Eight hundred thirty-five COPD patients were included in the cohort (Figure 1). They were
RI PT
mostly former smoker males with symptomatic COPD and moderate to severe airflow obstruction associated with frequent comorbidities (Table 1). Two hundred and fifty patients were lost to follow-up before the end of the study and 121 patients died, mainly of respiratory
SC
causes and cancers (Figure 1 and e-Table 1). Therefore, complete information on annual exacerbation number during the 4 years of the study was available for 464 patients. Patients
M AN U
lost to follow-up were not significantly different from patients who completed the study whereas patients who died were older and exhibited more severe airflow limitation (e-Table 2). The extended analysis was performed in 608 patients with at least two years of follow-up (Table 1). Patients followed at least one year and subsequently lost to follow-up or dead
TE D
reported significantly more exacerbations during the first year of the study (2.42 ± 1.9 and 2.4 ± 1.7, respectively) than patients who completed the study (1.55 ± 1.9, p<0.001).
EP
Identification of two clusters with homogeneous frequencies of exacerbations According to all criteria, the optimal number of clusters to separate patients in groups with
AC C
homogeneous numbers of moderate to severe exacerbations over time was two (Figure 2A). The first cluster (A) consisted of 348 patients (75%) who experienced few exacerbations (0.71 ± 0.54 exacerbations/patient/year) and the second (B) comprised 116 patients (25%) who experienced frequent exacerbations (2.89 ± 1.07 exacerbations/patient/year) over the four years of the study (Figure 2B and e-Figure 1). The ROC curve analysis found a good accuracy with an area under the curve (AUC) of 0.91 (CI: 0.88-0.94) and a best threshold to maximize (sensitivity + specificity) of at least 2 exacerbations in the first year (sensitivity 85% and specificity 79%) (e-Figure 2). The extended analysis found similar results with 451 patients
7
ACCEPTED MANUSCRIPT allocated to cluster A (74.2%) with few exacerbations (0.89 ± 0.61 exacerbations/patient/year) and 157 to cluster B (25.8%) with more frequent exacerbations (3.17 ± 1.05 exacerbations/patient/year). In this extended analysis, sensitivity and specificity of the threshold of at least 2 exacerbations in the first year were 91% and 74.4% with an AUC of
RI PT
0.92 (CI: 0.90-0.94). Stability of the number of exacerbations over time
SC
Among patients with full 4-year data, only 7% had at least 2 exacerbations per year and 36% less than 2 exacerbations per year during each of the whole four years of follow-up (Figure 3).
M AN U
Among the 171 patients with at least 2 exacerbations during the first year of the study, 47%, 31% and 19% persisted to have 2 or more exacerbations during the following one, two or three years, respectively. Similarly, among the 293 patients with less than 2 exacerbations during the first year of the study, 79%, 66% and 57% still had less than 2 exacerbations
TE D
during the first, second and third following years, respectively. Stability over time of annual exacerbation frequency in KmL-defined clusters was fair with intraclass correlation coefficients of 0,57 (CI: 0,50 - 0,64) in cluster A and 0,42 (CI: 0,25 - 0,56) in cluster B
EP
(Figure 2B).
AC C
Factors associated with the frequent exacerbator cluster Frequent exacerbators (cluster B) had more frequent history of severe respiratory infections during infancy, a longer disease duration and more severe airflow limitation (Table 2 and 3). Moreover, they had more respiratory symptoms and limitations of daily activities, a worse health-related quality of life and underwent pulmonary rehabilitation more frequently. Only five factors remained associated with the frequent exacerbator cluster in multivariate logistic regression analysis: a COPD diagnosis made more than 7 years ago, presence of daily sputum production, severe disability associated with breathlessness, severe airflow limitation and
8
ACCEPTED MANUSCRIPT hospitalization for exacerbation during previous years (Table 4). Results were similar in the extended analysis (data not shown). Clinical rule to predict the frequent exacerbator cluster
RI PT
We derived a score called ESOD (Exacerbation history, chronic Sputum production, GOLD stage of Obstruction and mMRC Dyspnoea stage) from results of the multivariate logistic regression analysis (Table 5). The AUC of 0.705 (CI: 0.65-0.75) suggests a fair accuracy of
SC
the score to predict belonging to the frequent exacerbator cluster with a best threshold for this score of ≥2 (e-Figure 3). The mean exacerbation rate and the probability of belonging to the
M AN U
frequent exacerbators cluster increased with the score (e-Table 3). With a score lower than 2, 86.2% of patients who completed the study and 87% of patients of the extended analysis belonged to the non-frequent exacerbator cluster (e-Figure 4, data not shown for extended
AC C
EP
TE D
analysis).
9
ACCEPTED MANUSCRIPT DISCUSSION In this prospective cohort study, KmL method identified 2 clusters of patients based on longitudinal 4-year follow-up of moderate-to-severe exacerbation frequency. This analysis
RI PT
confirmed that two exacerbations per year, the threshold adopted in GOLD guidelines to define the frequent exacerbator phenotype, is the most adequate threshold to distinguish frequent from non-frequent exacerbators.1 Although, a follow-up of a few years is required due to the variability of exacerbation rate over time. Finally, a simple clinical score based on
SC
multivariate analysis predicts the risk to belong to the frequent exacerbator phenotype. These
M AN U
results obtained in patients with a full 4-year follow-up were confirmed in a population extended to those with at least 2 years of follow-up.
During the four years of the study, 121 patients died and 250 were lost to follow-up. Dead patients were older with more severe airway obstruction at inclusion, which is consistent with
TE D
a higher mortality risk.13 In contrast, patients lost to follow-up were similar to patients who completed the four years of the study. Rate of lost patients was in the same range as in other published studies.4,14 Repeating analyses including lost and dead patients with at least two
EP
years of participation obtained similar results, suggesting that results appear applicable to the
AC C
entirety of the population.
The frequent exacerbator phenotype was defined several years ago as COPD patients experiencing a higher frequency of exacerbations associated with poorer outcomes.5,15 The strengths of the present work are (1) the use of the KmL method, which does not presuppose any number of clusters or threshold(s) and (2) the confirmation by three different statistical criteria that, in this population, the optimal number of clusters to describe the population is two. To our knowledge, this is the first study to confirm the existence of a specific phenotype with frequent exacerbations opposed to patients with few or no exacerbations using a
10
ACCEPTED MANUSCRIPT hypothesis-free method. The proportion of frequent exacerbators among COPD patients remains debated. The SPIROMICS cohort study reported a markedly lower proportion of stable frequent exacerbators over three years (2%) compared to ECLIPSE cohort (12%) and our cohort (11%).4,9 This may reflect inclusion of patients with milder airflow limitation in
RI PT
SPIROMICS (mean FEV1 63% versus 48% and 49%, respectively). Moreover, some studies suggested a third intermediate phenotype called infrequent exacerbators including patients who have one exacerbation per year, a phenotype positioned between frequent exacerbators
SC
and patients without exacerbations.4,9,16–18 Our analysis did not identify this intermediate
M AN U
phenotype, which may relate to an insufficient population size or number of time points. From a clinical perspective, an important challenge is to identify patients at risk of future exacerbations when no sufficient follow-up is available yet, in order to provide appropriate education and prescribe appropriate therapy to prevent exacerbations. In contrast, identifying patients with a low risk of exacerbation may prevent overtreatment. However, due to the
TE D
variability of exacerbation rate over time, this identification is not straightforward. Based on annual exacerbation frequency, one third of patients changed group between first and second
EP
year of our study, which is concordant with other cohorts.4,9 Therefore, tools to predict the frequent exacerbator phenotype are needed in clinical practice.
AC C
Some authors tried to develop models to predict the risk of COPD exacerbations using multivariate analysis but most of those scores have a poor practical applicability.19 Based on our population, a score applicable in clinical practice was developed. Since the reported duration of the disease is subject to recall errors, we decided not to include this variable in the score, although it was significantly associated with the risk of frequent exacerbation in multivariate analysis. Criteria have not been weighted, as it did not change the area under the ROC curve and as we wanted to develop a simple score for clinical practice. Although the performance of the score is lower to predict the risk of having ≥2 exacerbations the following 11
ACCEPTED MANUSCRIPT year, a score lower than 2 accurately predicts a low risk of belonging to the frequent exacerbator cluster. The AUC was in the same range as other established scores.19 Unfortunately, our study was designed before the publication of ECLIPSE study results, therefore the number of exacerbations the previous year was not collected and could not be
AC C
EP
TE D
M AN U
SC
and significantly associated with risk of future exacerbations.
RI PT
included in our model. However, the number of previous severe exacerbations was collected
12
ACCEPTED MANUSCRIPT CONCLUSION For the first time, the existence of a frequent exacerbator phenotype in COPD patients was demonstrated by a statistical method that does not rely on pre-specified hypotheses regarding
RI PT
the final number of clusters or the most reliable threshold(s). Interestingly, this method confirmed the currently used threshold of 2 exacerbations per year to identify these patients. To allow easy identification of at-risk patients in routine care before prolonged follow-up is available, we propose a new simple score to predict the risk of COPD exacerbations, which
AC C
EP
TE D
M AN U
SC
now needs to be prospectively tested.
13
ACCEPTED MANUSCRIPT ACKNOWLEDGEMENT In memory of Pr. Isabelle Tillie-Leblond who has taken an essential role in the elaboration, coordination, conduct and analysis of this study, and more widely in the field of respiratory medicine in France.
and whose dedication to research made the trial possible.
SC
AUTHOR CONTRIBUTIONS
RI PT
The authors thank respiratory physicians and COPD patients who participated in the study,
Dr. Le Rouzic, Prs Roche and Housset and Ms. Pouriel are guarantors of the manuscript and
M AN U
take responsibility for the integrity of the data and the accuracy of the data analysis. All authors participated to the study design, conduct or data analysis and interpretation, and
AC C
EP
TE D
approved the final version.
14
ACCEPTED MANUSCRIPT REFERENCES Vogelmeier CF, Criner GJ, Martinez FJ, et al. Global Strategy for the Diagnosis, Management, and Prevention of Chronic Obstructive Lung Disease 2017 Report: GOLD Executive Summary. Eur Respir J 2017;49(3):1700214.
2.
Suissa S, Dell’Aniello S, Ernst P. Long-term natural history of chronic obstructive pulmonary disease: severe exacerbations and mortality. Thorax 2012;67(11):957–963.
3.
Punekar YS, Shukla A, Müllerova H. COPD management costs according to the frequency of COPD exacerbations in UK primary care. Int J Chron Obstruct Pulmon Dis 2014;9:65–73.
4.
Hurst JR, Vestbo J, Anzueto A, et al. Susceptibility to exacerbation in chronic obstructive pulmonary disease. N Engl J Med 2010;363(12):1128–1138.
5.
Donaldson G, Seemungal T, Bhowmik A, Wedzicha J. Relationship between exacerbation frequency and lung function decline in chronic obstructive pulmonary disease. Thorax 2002;57(10):847–852.
6.
Vestbo J, Edwards LD, Scanlon PD, et al. Changes in Forced Expiratory Volume in 1 Second over Time in COPD. N Engl J Med 2011;365:1184–1192.
7.
Soler-Cataluña JJ, Rodriguez-Roisin R. Frequent chronic obstructive pulmonary disease exacerbators: how much real, how much fictitious? COPD 2010;7(4):276–284.
8.
McGarvey L, Lee AJ, Roberts J, Gruffydd-Jones K, McKnight E, Haughney J. Characterisation of the frequent exacerbator phenotype in COPD patients in a large UK primary care population. Respir Med 2015;109(2):228–237.
9.
Han MK, Quibrera PM, Carretta EE, et al. Frequency of exacerbations in patients with chronic obstructive pulmonary disease: an analysis of the SPIROMICS cohort. Lancet Respir Med 2017;5(8):619–626.
EP
TE D
M AN U
SC
RI PT
1.
AC C
10. Rodriguez-Roisin R. Toward a consensus definition for COPD exacerbations. Chest 2000;117(5 Suppl 2):398S–401S. 11. Genolini C, Alacoque X, Sentenac M, Arnaud C. kml and kml3d: R Packages to Cluster Longitudinal Data. J Stat Softw 2015;65(4):1–34. 12. Robin X, Turck N, Hainard A, et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics 2011;12:77. 13. Jensen HH, Godtfredsen NS, Lange P, Vestbo J. Potential misclassification of causes of death from COPD. Eur Respir J 2006;28(4):781–785. 14. Calverley PMA, Anderson JA, Celli B, et al. Salmeterol and Fluticasone Propionate and Survival in Chronic Obstructive Pulmonary Disease. N Engl J Med 2007;356(8):775– 789.
15
ACCEPTED MANUSCRIPT 15. Soler-Cataluña JJ, Martínez-García MÁ, Sánchez PR, Salcedo E, Navarro M, Ochando R. Severe acute exacerbations and mortality in patients with chronic obstructive pulmonary disease. Thorax 2005;60(11):925–931. 16. Beeh KM, Glaab T, Stowasser S, et al. Characterisation of exacerbation risk and exacerbator phenotypes in the POET-COPD trial. Respir Res 2013;14:116.
RI PT
17. Yang H, Xiang P, Zhang E, et al. Predictors of exacerbation frequency in chronic obstructive pulmonary disease. Eur J Med Res 2014;19:18.
18. Tomioka R, Kawayama T, Suetomo M, et al. “Frequent exacerbator” is a phenotype of poor prognosis in Japanese patients with chronic obstructive pulmonary disease. Int J Chron Obstruct Pulmon Dis 2016;11:207–216.
AC C
EP
TE D
M AN U
SC
19. Guerra B, Gaveikaite V, Bianchi C, Puhan MA. Prediction models for exacerbations in patients with COPD. Eur Respir Rev 2017;26(143):160061.
16
ACCEPTED MANUSCRIPT FIGURE LEGENDS Figure 1: Flow diagram of participants through each stage of the study.
SC
RI PT
Figure 2: KmL clustering analysis of prospectively recorded annual exacerbation frequency. A Criteria applied to choose the optimal number of clusters including (1-black) the Calinski & Harabatz index, (2-red) the Ray & Turi index and (3-green) the inverse of the Davies & Bouldin index. These criteria are normalized to be mapped together using kml package in R. Scores range from 0 to 1 for each tested number of clusters ranging from 2 to 6. The highest score corresponds to the most appropriate number of clusters. B Individual trajectories of annual exacerbation frequency during each year of the study. The red colour corresponds to cluster A or non-frequent exacerbators (75% of patients) and the blue colour to cluster B or frequent exacerbators (25% of patients). Thick lines labelled with the cluster letter represent the mean exacerbation frequency for each cluster.
AC C
EP
TE D
M AN U
Figure 3: Stability of the frequent-exacerbators phenotype (defined as ≥2 exacerbations per year) with KmL clusters included. The bars show the proportions of patients with no exacerbation (purple) or two or more exacerbations (red) in each years 1, 2, 3 and 4 from left to right, respectively. Each bar is subdivided in light colour for patients belonging to KmL cluster A (non-frequent exacerbators) and in dark colour for patients belonging to cluster B (frequent exacerbators). The numbers on the right denote the numbers of patients belonging to the B and A clusters for each trajectory.
17
ACCEPTED MANUSCRIPT TABLES
Table 1: Demographics and baseline characteristics of patients
mean ± SD % mean ± SD
66.3 ± 9.6 80,4 25.9 ± 5.2
66.5 ± 9.2 79,6 25.9 ± 5.2
% % %
14.3 73.5 12.1
Smoking status Active smoker % 24,7 Ex-smoker % 75,3 Smoking history (pack-years) mean ± SD 44.4 ± 19.8
AC C
FEV1 (% of predicted) FVC (% of predicted)
12.8 74.7 12.6
24.0 76.0 44.5 ± 20.1
22.8 77.2 44.4 ± 19.4
8.7 ± 7.2
9.0 ± 7.4
8.7 ± 7.0
% % %
42.6 32.8 24.6
43.6 33.1 23.4
45.7 33.2 21.1
47.9 ± 15.4 72.0 ± 18.8
48.7 ± 15.3 72.2 ± 18.3
EP
GOLD stage 2 3 4
12.6 74.7 12.6
mean ± SD
TE D
Time since COPD diagnosis (years)
66.1 ± 8.9 79,3 26.0 ± 5.1
SC
Occupational status Working Retired Unemployed
Completed (n=464)
RI PT
Extended (n=608)
M AN U
Age (year) Male sex BMI (kg/m2)
Full cohort (n=835)
mean ± SD 47.7 ± 15.6 mean ± SD 71.8 ± 18.9
This table summarizes demographic and baseline characteristics of all included patients (Full cohort), patients analysed in the extended analysis (Extended) and the 464 patients who completed the 4-year follow-up. All measures were recorded at screening. BMI: body-mass index, COPD: chronic obstructive pulmonary disease, GOLD: global initiative for chronic obstructive lung disease, FEV1: forced expiratory volume in the first second, FVC: forced vital capacity.
18
ACCEPTED MANUSCRIPT
B (n=116)
mean ± SD % mean ± SD
66.4 ± 8.8 80.2 26.3 ± 5.1
65.3 ± 9.0 76.7 25.3 ± 5.2
NS NS NS
% % %
13.3 75.3 11.3
NS
Smoking status Active smoker Ex-smoker Smoking history (pack-years)
10.5 73.0 16.4
M AN U
Occupational status Working Retired Unemployed
p-value A vs. B
RI PT
Age (year) Male sex BMI (kg/m2)
A (n=348)
SC
Table 2: Comparisons of demographics and baseline characteristics of patients who completed the 4-year follow-up separated in cluster A (non-frequent exacerbators) and B (frequent exacerbators)
% % mean ± SD
23.0 77.0 44.8 ± 19
22.4 77.6 43.4 ± 20.5
NS
mean ± SD
8.1 ± 6.8
10.3 ± 7.3
< 0.01
% % %
50.6 31.0 18.4
31.0 39.7 29.3
< 0.001
mean ± SD mean ± SD
50.3 ± 15.2 73.7 ± 18.1
43.9 ± 14.3 67.7 ± 18.0
< 0.001 < 0.01
Symptoms Daily sputum production Daily wheezing Daily coughing
% % %
52.9 27.6 61.5
70.7 38.8 75.0
< 0.001 < 0.05 < 0.05
mMRC score 0 1 2 3 4
% % % % %
12.6 44.0 31.6 10.9 0.9
5.2 37.9 25.0 30.2 1.7
< 0.001
TE D
Time since COPD diagnosis (years)
EP
GOLD stage 2 3 4
AC C
FEV1 (% of predicted) FVC (% of predicted)
NS
19
ACCEPTED MANUSCRIPT Daily activities limitations No impairment Mild impairment Moderate impairment Severe impairment Complete impairment
8.7 24.3 30.4 35.7 0.9
< 0.01
mean ± SD
46.2 ± 15.7
42.6 ± 15.5
< 0.05
% % %
15.2 16.6 7.6
26.3 27.6 4.3
% % % % % % % % % % % %
31.1 18.0 15.7 10,1 12.0 13.1 11.5 12.6 9.8 5.5 7.2 4.1
% %
4.7 3.5
TE D
Comorbidities Hypertension Osteoarthritis Ischemic heart disease Chronic alcohol consumption Depression Heart rythm disorder Lower limb arteriopathy Diabetes Sleep apnoea Clinical right heart failure Left heart failure Severe respiratory infection in infancy
< 0.05 < 0.05 NS
28.7 27.0 13.9 15.7 7.9 10.3 7.8 6.9 8.8 6.9 4.3 11.2
NS < 0.05 NS NS NS NS NS NS NS NS NS < 0.01
6.0 5.3
NS NS
EP
Pulmonary hypertension Stroke
SC
Treatments Respiratory rehabilitation Long-term oxygen therapy Non invasive ventilation
RI PT
15.4 27.5 38.8 16.9 1.5
M AN U
VSRQ global score
% % % % %
AC C
This table compares demographic and baseline characteristics of the two clusters obtains from the 464 patients who completed the 4-year follow-up, non-frequent exacerbators cluster (A) and frequent exacerbators cluster (B). All measures were recorded at screening. Comparisons between clusters A and B were performed using t-test or Fisher’s exact test as appropriate. p<0.05 was considered significant. BMI: body-mass index, COPD: chronic obstructive pulmonary disease, GOLD: global initiative for chronic obstructive lung disease, FEV1: forced expiratory volume in the first second, FVC: forced vital capacity, mMRC: modified Medical Research Council dyspnoea score, VSRQ global score: visual simplified respiratory questionnaire global score ranging from 80 (best health status) to 0 (poorest health status). NS: not significant.
20
ACCEPTED MANUSCRIPT Table 3: Univariate regression analysis of associations with belonging to the frequent exacerbator cluster (B) in patients with full 4-year follow-up Odds Ratio
Variables Demographic data
FVC
Symptoms Daily sputum production yes vs. no Daily wheezing yes vs. no Daily coughing yes vs. no mMRC dyspnoea score 1-2 vs. 0 3-4 vs. 0
TE D
Patient-reported outcomes Daily activities limitations Mild to moderate impairment Severe to complete impairment
VSRQ global score
per decrease of 4 points
2.33 2.57 1.64
[1.42 - 3.87] 0.001 [1.49 - 4.46] < 0.001 [1.26 - 2.14] < 0.001
1.15
[1.07 - 1.24] < 0.001
1.09
[1.03 - 1.17]
< 0.01
2.15 1.64 1.84 2.03 6.62
[1.38 - 3.41] 0.001 [1.05 - 2.54] 0.03 [1.16 - 2.99] 0.01 [0.9 - 5.49] NS [2.69 - 18.9] < 0.001
0.68
[0.31 - 1.37]
2.41 1.06
[1.49 - 3.9] < 0.001 [1.01 - 1.13] 0.04
1.91 0.54
[1.16 - 3.13] [0.18 - 1.34]
NS
0.01 NS
AC C
EP
Treatments Long-term oxygen therapy yes vs. no Non invasive ventilation yes vs. no
[0.69 - 1.11] NS [0.73 - 2.01] NS [0.92 - 1.00] NS [0.58 - 1.58] NS [1.06 - 1.43] < 0.01 [1.57 - 3.82] < 0.001
M AN U
per increase to next stage per 5% decrease in % predicted value per 5% decrease in % predicted value
FEV1
0.87 1.22 0.96 0.97 1.23 2.45
SC
4 vs. 2
p-value
RI PT
Age per 10-year increase Sex female vs. male BMI per increase of 1 point Smoking status current vs. former smoker Duration of COPD per 5-year increase Hospitalization previous year yes vs. no Lung function GOLD stage 3 vs. 2
95% CI
BMI: body-mass index, COPD: chronic obstructive pulmonary disease, GOLD: global initiative for chronic obstructive lung disease, FEV1: forced expiratory volume in the first second, FVC: forced vital capacity, mMRC: modified Medical Research Council dyspnoea score, VSRQ global score: visual simplified respiratory questionnaire global score. NS: not significant.
21
ACCEPTED MANUSCRIPT Table 4: Factors associated with the frequent exacerbator cluster (B) in the multivariate model in patients with full 4-year follow-up p-value
[1.06 - 2.79]
0.03
2.05
[1.25 - 3.42]
0.005
1.66 4.53 1.77
[0.67 - 5.1] [1.63 - 14.8] [1.08 - 2.93]
NS 0.006 0.03
1.99
[1.21 - 3.26]
RI PT
95% CI
SC
Time since COPD diagnosis (reference: less than 7 years) Daily sputum production (reference: no) Dyspnoea score (mMRC) (reference: mMRC 0) - mMRC 1 and 2 - mMRC 3 and 4 FEV1 (% of predicted) (reference: more than 50%) Hospitalization previous year (reference: none)
Odds Ratio 1.71
0.006
M AN U
Variables
AC C
EP
TE D
COPD: chronic obstructive pulmonary disease, mMRC: modified Medical Research Council dyspnoea score, FEV1: forced expiratory volume in the first second, 95%CI: 95% confidence interval. NS: not significant.
22
ACCEPTED MANUSCRIPT Table 5: ESOD score Points
Exacerbation (any hospitalization during previous 2 years) Sputum (chronic daily production)
Obstruction (%FEV1) Dyspnoea (mMRC score)
0
1
No
Yes
No ≥ 50% 0-1-2
Yes < 50% 3-4
RI PT
ESOD score
AC C
EP
TE D
M AN U
SC
The ESOD score is the sum of points related to each variable and ranges from 0 to 4. mMRC: modified Medical Research Council dyspnoea score, %FEV1: percent predicted forced expiratory volume in the first second.
23
ACCEPTED MANUSCRIPT
French Database n = 2511 respiratory physicians
RI PT
n=132 randomised respiratory physicians
SC
n=835 included patients
M AN U
n=37 lost to follow-up n=38 deaths
n=760 patients followed 1 year
n=675 patients followed 2 years
AC C
EP
n=608 patients with at least two years with full data
TE D
“Extended analysis”
n=53 lost to follow-up n=32 deaths
n=55 lost to follow-up n=25 deaths
n=595 patients followed 3 years
n=105 lost to follow-up n=26 deaths
n=464 patients followed 4 years
AC C
EP
TE D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT 0
20
43 (n=1 40
60
80
57 (n=1
47% (n=81)
0
20
40
RI PT
53% (n=90)
60
M AN U
68% (n=61)
EP
69% (n=2
20
20 (n=1 40
60
80
80 (n=4
TE D
0
70
31 (n=
SC
32% (n=29)
AC C
0
35% (n=28)
75 (n=1
25 (n= 32% (n=20)
68% (n=42)
29 (n=1 0
20
40
60
80
71 (n=3
21% (n=62)
79% (n=231)
0
20
40
60
80
100
32 (n=1
68
ACCEPTED MANUSCRIPT
e-Appendix 1 - SUPPLEMENTAL METHODS Physicians’ recruitment
RI PT
Physicians were selected at random from a complete database of 2511 French respiratory physicians covering office-based and hospital settings from the private and public sectors. An information letter was sent and they were phoned to enquire if they wanted to participate. If a physician refused, the next physician on the list was contacted until the predefined number of centers was reached. Altogether, 1202 respiratory physicians were
SC
asked to participate, 276 accepted and 132 actually included patients in the cohort. Characteristics of these active investigators were similar to that of the total French population of respiratory physicians except for sex (77% of males versus 68% in the total
M AN U
population of French respiratory physicians) and activity (46% of hospital-based physicians versus 22% in the total population of French respiratory physicians), and covered all French territory. Physicians had to include the 10 first patients who fulfilled all required conditions and accepted to participate. The mean number of patients actually recruited was 6.3 ± 4 per physician.
Physicians
completed
TE D
Data collection by physicians a
case-report
form
during
medical
consultations
recording
sociodemographic data (age, sex, marital status, education level, last profession, lodging type, urban/rural residence), history of occupational exposure to smoke, dust or gases, family history of COPD, height and weight, smoking status and history, comorbid medical
EP
conditions including cardiovascular disease and diabetes mellitus and other symptoms or syndromes potentially relevant to patients’ respiratory condition (atopy, gastro-oesophageal
AC C
reflux, nasal drip and sinusitis). A questionnaire asked details about patients’ COPD history (time since onset of breathlessness, cough, expectoration and time since COPD diagnosis, number of hospitalizations for COPD exacerbations the previous 2 years) and its clinical expression (presence of daily cough, expectoration, wheezing, and dyspnoea). Physicians also recorded current treatments for COPD, influenza and pneumococcal vaccinations and selected
non-pharmacological
management
of
COPD
(physiotherapy,
respiratory
rehabilitation programs, home oxygen, non-invasive nocturnal ventilation). Patients with ischemic heart disease, arrhythmias, left heart failure, peripheral artery disease, or with a previous cerebrovascular accident were defined as having cardiovascular disease. Patients reporting both daily cough and daily expectoration were defined as having a chronic bronchitis profile.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
Patients were asked to complete the modified Medical Research Council (mMRC) Dyspnoea Scale (Bestall Thorax 1999) and the Borg score (Silverman Am Rev Respir Dis 1988) as measures of breathlessness, the Visual Simplified Respiratory Questionnaire (VSRQ) (Perez
RI PT
Int J COPD 2009) as a measure of health status, and to classify their limitation in daily activities because of respiratory symptoms as absent, mild, moderate, severe or complete. Results of the most recent lung function tests, sputum microbiology, oxygen saturation and arterial blood gases were recorded. The mMRC score ranges from 0 (not troubled with breathlessness except with strenuous exercise) to 4 (too breathless to leave the house or status) to 0 (poorest health status).
M AN U
Monitoring of symptoms and exacerbations
SC
breathless when dressing or undressing) and the VSRQ global score from 80 (best health
Patients were asked to complete an auto-questionnaire for each respiratory symptom lasting more than 2 days and every month regarding their respiratory status. This questionnaire collected dates of start and end of these symptoms, apparition or increase of clinical symptoms (sputum volume and aspect, breathlessness, cough, fever, thoracic pain, wheezing and cold), dyspnoea auto-evaluation with Borg scale, decrease of daily activity, therapeutic modification, medical consultation, hospitalization and work stopping. Patients
TE D
were also contacted by telephone every three months by the Kappa Santé Society (Paris, France) to ensure this information was complete. In addition to patients’ direct reports of exacerbations, respiratory physicians recorded the number of COPD exacerbations requiring a change in treatment or hospitalization since the last consultation at each follow-up visit,
EP
(one to two visits per year).
In the event of deterioration, patients had to complete an additional questionnaire described
AC C
in the previous chapter. This questionnaire asked about the presence of eight symptoms (sputum volume and purulence, common cold, breathlessness, cough, fever, chest pain, wheezing) and included the Borg score as a measure of dyspnoea and a scale measuring limitation in daily activities. Dates of start and end of these symptoms were requested. Patients also reported if they had changed their own medication because of the worsened symptoms and if they had consulted a doctor, presented to a hospital emergency department, or been hospitalized in association with the episode, and when their respiratory state returned to normal. These data were collected for each exacerbation. KmL method KmL is a nonparametric clustering method to identify homogeneous trajectories in longitudinal data (Genolini and Falissard, 2011). This method was used to identify patients Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
with homogeneous yearly exacerbation rate over the 4 years of the study and to explore the existence of a frequent exacerbator phenotype based on exacerbation frequency. The advantages of this method are the following: (1) it does not require normality or any
RI PT
parametric assumption within clusters, (2) it does not require any assumption regarding the shape of the trajectory, (3) it is able to deal with missing values and (4) it does not require any assumption regarding the number of cluster. After running the algorithm, KmL most frequently uses the Calinski & Habaratz criterion (Calinski and Harabatz, 1974) to objectively choose the optimal number of clusters; two other criteria were applied to check
SC
consistency: the Ray & Turi criteria (Ray and Turi, 1999) and the Davies & Bouldin criteria (Davies and Bouldin, 1979). These non-parametric indices reflect both between-cluster and within-cluster variances. The best solution maximizes the between-cluster variance,
M AN U
meaning that trajectories of patients of different clusters are well-separated, and minimizes within-cluster variance, meaning that trajectories of patients within the same cluster are close to each other giving a compact cluster. For this study, KmL has been parameterized to search trajectories with up to 6 clusters and run the algorithm 20 times for each number of clusters. The three criteria were normalized to be mapped together (Figure 2A). The scores range from 0 to 1 for each tested number of clusters and the highest score corresponds to
AC C
EP
TE D
the most appropriate number of clusters.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
e-Table 1: Causes of death during follow-up Dead patients
Causes of death Respiratory failure
Cardiovascular
Respiratory infection
5
Pneumothorax
4
Pulmonary embolism
1
Myocardial ischemia
3
SC
Respiratory
29
Arteriosclerosis
2
Cardiac arrhythmia
2
Colorectal neoplasms Kidney neoplasms Neoplasms
Pancreatic neoplasm Oesophageal neoplasm Brain neoplasm
19 2 2 1 1 1
Otorhinolaryngologic neoplasm
1
Suicide
1
TE D
Others
1
M AN U
Heart arrest Lung neoplasms
Gastrointestinal haemorrhage
1
Cerebral haemorrhage
1 44
AC C
EP
Not defined
RI PT
(n=121)
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
e-Table 2: Baseline comparison with lost to follow-up
66.1 ± 8.9
65 ± 10.6
%
79.3
79.2
mean ± SD
26 ± 5.1
26 ± 5.7
2
%
46.8
3
%
31.4
4
%
21.8
mean ± SD
48.7 ± 15.3
Male sex BMI (kg/m2) GOLD stage
FEV1 (% predicted)
mean ± SD
45.2 ± 15.7
Previous severe exacerbation
mean ± SD
0.45 ± 0.9
2
25.2 ± 4.7
NS
1
1
p-value 1 vs 3
p-value 2 vs 3
NS
***
***
< 0.001
2
NS
***
***
41.2 ± 15.9
< 0.001
1
NS
***
***
40.0 ± 17.6
0.01
1
NS
***
NS
0.75 ± 1.2
0.01
1
NS
NS
NS
29.0
26.3
23.7
50.0
49.1 ± 15.5
0.70 ± 2.1
0.02 NS
23.7
44.6 ± 15.8
p-value p-value 3 groups 1 vs 2
86.8
47.3
TE D
VSRQ global score
69.6 ± 9.1
M AN U
mean ± SD
RI PT
Age (year)
Lost to Dead patients follow-up (N=121) - 3 (N=250) - 2
SC
Followed patients (N=464) - 1
AC C
EP
BMI: body-mass index, GOLD: global initiative for chronic obstructive lung disease, FEV1: forced expiratory volume in the first second, VSRQ global score: visual simplified respiratory questionnaire global score ranging from 80 (best health status) to 0 (poorest health status). 1 one-way analysis of variance (ANOVA) 2 chi-squared test. Post-hoc comparisons between each column were performed using ttest with Bonferroni adjustment. NS: not significant, *** = p < 0.001.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
e-Table 3: Mean exacerbation rate and probability to belong to the frequent
RI PT
exacerbator cluster (B) for each ESOD score
ESOD score 0
1
2
3
4
75
151
125
81
13
0.71
1.03
1.36
1.87
2.67
8
17.2
28
46.9
69.2
Mean annual exacerbation rate Probability to belong to B cluster Extended N Mean annual exacerbation rate
87
189
164
120
24
0.88
1.20
1.56
2.14
2.37
10.3
14.3
29.3
47.5
54.2
TE D
Probability to belong to B cluster
M AN U
N
SC
Completed
This table summarizes mean exacerbation rate and probability to belong to the frequent exacerbator cluster (B) for each ESOD score. Results are given for the patients who completed the 4-year follow-up (464) and patients analysed in the extended analysis
AC C
EP
(Extended). N: number of patients for each ESOD score.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
TE D
M AN U
SC
RI PT
AC C
EP
e-Figure 1: Overlap of exacerbations frequency for each year of the study in the 2 clusters. y-axis are numbers of exacerbations for each year of the study for clusters A (red) and B (blue). The white point is the median, the bold black lines represent the interquartile range (25th to 75th centiles) and the thin black lines represent the whiskers extended to the more extreme data point which is no more than 1.5 times the interquartile range from the box. The width represents the density for each value.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
Threshold performances Se = 99.1 % Sp = 47.7 % PPV = 38.7 % NPV = 99.4 %
RI PT
>0
M AN U
SC
>1
>2
Se = 68.1 % Sp = 93.4 % PPV = 77.5 % NPV = 89.8 %
>3
Se = 50.9 % Sp = 97.7 % PPV = 88.1 % NPV = 85.6 %
TE D
AUC = 0.91 (0.88 - 0.94)
Se = 85.3 % Sp = 79.3 % PPV = 57.9 % NPV = 94.2 %
AC C
EP
e-Figure 2: Receiver operating characteristic (ROC) curve illustrating the performances of using the number of exacerbations the first year of the study to identify the KmL-defined frequent exacerbator cluster. The grey area around the ROC curve represents 95% confidence intervals. Test performances are given for 0 to 3 moderate to severe exacerbations the first year of the study. AUC: area under the curve, Se: sensitivity, Sp: specificity, PPV: positive predictive value, NPV negative predictive value.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
Threshold performances Se = 94.8 % Sp = 21.2 % PPV = 28.6 % NPV = 94.4 %
RI PT
>0
M AN U
SC
>1
Se = 71.6 % Sp = 59.2 % PPV = 36.9 % NPV = 86.2 %
>2
Se = 40.5 % Sp = 86.2 % PPV = 49.5 % NPV = 81.3 %
>3
Se = 7.8 % Sp = 98.9 % PPV = 69.2 % NPV = 76.3 %
TE D
AUC = 0.705 (0.65 - 0.75)
AC C
EP
e-Figure 3: Receiver operating characteristic (ROC) curve illustrating the performances of the ESOD score to positively predict belonging to cluster B (frequent exacerbators according to KmL). The grey area around the ROC curve represents 95% confidence intervals. Test performances are given for each threshold of the score. AUC: area under the curve, Se: sensitivity, Sp: specificity, PPV: positive predictive value, NPV negative predictive value.
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.
ACCEPTED MANUSCRIPT
Threshold performances Se = 92.4 % Sp = 22.9 % PPV = 41.1 % NPV = 83.8 %
RI PT
>0
SC
>1
Se = 59.6 % Sp = 58.0 % PPV = 45.3 % NPV = 71.1 %
>2
Se = 30.4 % Sp = 85.3 % PPV = 54.7 % NPV = 67.8 %
>3
Se = 5.8 % Sp = 99.0 % PPV = 76.9 % NPV = 64.3 %
TE D
M AN U
AUC = 0.64 (0.58 - 0.68)
AC C
EP
e-Figure 4: Receiver operating characteristic (ROC) curve illustrating the performances of the ESOD score to positively predict the risk of having at least 2 exacerbations in the following year. The grey area around the ROC curve represents 95% confidence intervals. Test performances are given for each threshold of the score. AUC: area under the curve, Se: sensitivity, Sp: specificity, PPV: positive predictive value, NPV: negative predictive value.
171104
Online supplements are not copyedited prior to posting and the author(s) take full responsibility for the accuracy of all data.