Stability of phenotypes defined by physiological variables and biomarkers in adults with asthma

Authors


  • Edited by: Michael Wechsler

Abstract

Background

Although asthma is characterized by variable airways obstruction, most studies of asthma phenotypes are cross-sectional. The stability of phenotypes defined either by biomarkers or by physiological variables was assessed by repeated measures over 1 year in the Pan-European BIOAIR cohort of adult asthmatics.

Methods

A total of 169 patients, 93 with severe asthma (SA) and 76 with mild-to-moderate asthma (MA), were examined at six or more visits during 1 year. Asthma phenotype clusters were defined by physiological variables (lung function, reversibility and age of onset of the disease) or by biomarkers (eosinophils and neutrophils in induced sputum).

Results

After 1 year of follow-up, the allocation to clusters was changed in 23.6% of all asthma patients when defined by physiological phenotypes and, remarkably, in 42.3% of the patients when stratified according to sputum cellularity (P = 0.034). In the SA cohort, 30% and 48.6% of the patients changed allocation according to physiological and biomarker clustering, respectively. Variability of phenotypes was not influenced by change in oral or inhaled corticosteroid dose, nor by the number of exacerbations. Lower stability of single and repeated measure was found for all evaluated biomarkers (eosinophils, neutrophils and FeNO) in contrast to good stability of physiological variables (FEV1), quality of life and asthma control.

Conclusion

Phenotypes determined by biomarkers are less stable than those defined by physiological variables, especially in severe asthmatics. The data also imply that definition of asthma phenotypes is improved by repeated measures to account for fluctuations in lung function, biomarkers and asthma control.

Initial attempts to define asthma phenotypes were based on the broad concepts of extrinsic (allergic) and intrinsic (nonallergic) asthma [1, 2]. With increased understanding of pathobiology, classifications identifying distinct inflammatory phenotypes have been attempted [3, 4]. Hastie et al. [5] proposed asthma phenotypes stratification using biomarkers of inflammation (namely eosinophils and neutrophils in induced sputum), whereas Moore et al. [6] suggested another approach based only on physiological variables and medical history. Although these attempts represent important steps towards a mechanism-based classification of asthma, a significant limitation of the studies published to date is that they often derive from cross-sectional measurements taken at a single time point or during a short interval of time [7].

The Pan-European BIOAIR study of severe asthma includes repeated measures of many outcomes, including sputum data [8, 9]. We therefore performed an analysis of the stability of phenotypes in adults with controlled and severe asthma using the BIOAIR cohort. The aim was to compare phenotypes defined either by biomarkers (‘the Hastie approach’) or by physiological variables (‘the Moore approach’) over 1 year of observations. Concerning inflammatory phenotypes defined by sputum analysis, there are conflicting data on its repeatability. Fleming et al. [10] found that sputum results were unstable in children with asthma, whereas previous studies in adults generally suggest reasonable stability and repeatability of sputum measurements [11-13].

Methods

Subjects

Patients aged 18–80 years were recruited for the study and diagnosed by pulmonary specialists according to standard criteria [14] at the 12 participating European centres. Inclusion in the severe asthma (SA) group required specialist treatment for at least 1 year and at least one exacerbation requiring oral steroid treatment in the past year, despite continuous treatment with high doses of inhaled corticosteroids (ICS; at least 1600 μg/day budesonide or beclomethasone, 800 μg/day fluticasone or equivalent), and long-acting β-agonists or oral theophylline for at least 1 year [8, 9, 15]. Subjects with less heavy medications and no exacerbations in the past year were included in the mild-to-moderate asthma group (MA) as detailed in the online depository.

Study design

All patients after screening (V1) underwent 4 weeks of treatment optimization period (V2) followed by a 2-week double-blind placebo-controlled steroid intervention (V3). After the screening, patients were allocated to SA and mild-to-moderate asthma (MA) cohorts, and thereafter followed for 12 months with control visits at four-monthly intervals (V4–V6), as well as at additional visits in case of exacerbations (Vex). The study flow chart with time points for consecutive visits is presented in Fig. S1 (online repository). Information regarding lung functions, biomarkers [induced sputum (with exception of V5), peripheral blood, exhaled NO], atopy, medical history, asthma control and quality of life was collected at baseline and during consecutive visits. The study (clinicaltrials.gov NCT00555607) was approved by the ethics review boards in the twelve participating centres. More specifications concerning the BIOAIR study design and methodology are available in the online depository.

Definition of asthma phenotypes

Asthma phenotype based on clinical and physiological variables was defined by the use of the algorithm proposed by Moore et al. [6]. Five clusters were identified by baseline lung function, reversibility and age of onset of the disease. Asthma phenotype based on biomarkers was defined as proposed by Hastie et al. [5]. Four subgroups were identified by percentage of eosinophils (<2% or ≥2%) and neutrophils (<40% or ≥40%) in induced sputum.

Statistical analysis

Data were entered into a central database through a web-based case record form (eCRF) system developed specifically for the BIOAIR study. Patient baseline characteristics are expressed as mean ± standard error of the mean (SEM). Continuous variables were analysed using Student's t-test or Mann–Whitney U-test for nonparametric values and categorical variables by chi-square test. Stability of eosinophil and neutrophil counts and lung function measurements was assessed using intraclass correlation coefficients (ICCs). An ICC was obtained for single measures (the stability of individual measurement) and average measures (the stability of a measurement relative to all measures in that individual). P < 0.05 was considered a statistically significant difference for all tests. The generalized estimating equations for logistic regression have been used to analyse stability of allocation to clusters (baseline vs at 1 year of follow-up and MA vs SA). All analyses were carried out using GraphPad Prism V.5 (GraphPad Software, Inc., La Jolla, CA, USA) or SPSS (V.19; IBM Corporation, Armonk, NY, USA).

Results

Baseline characteristics

A total of 169 asthmatic patients were included in the BIOAIR study, including 93 with severe and 76 with mild-to-moderate disease (Table 1). Of those, 138 patients (80 SA and 58 MA) completed 1 year of the follow-up study (Fig. S2 for flow chart and specifications).

Table 1. Demographic data and baseline characteristics of the study cohort [mean values (±SEM), unless stated differently]
 Severe asthmaMild-to-moderate asthma P
  1. ACQ, Asthma Control Questionnaire; SGRQ, St George's Respiratory Questionnaire; OCS, oral corticosteroids; ICS, inhaled corticosteroids; CRP, serum C-reactive protein; atopy defined as at least one positive skin prick test (ND, not determined, *Mann–Whitney U-test, †chi-square test).

Number of patients in the BIOAIR cohort (n)9376ND
Age (years, min–max)50.0 ± 1.3 (18–72)42.2 ± 1.5 (20–70)0.001*
Females (%)58610.982
FEV1 (% pred)70.4 ± 2.188.7 ± 2.1<0.0001*
FEV1 (l)2.04 ± 0.082.79 ± 0.08<0.0001*
FEV1/FVC0.67 ± 0.010.70 ± 0.010.093*
Reversibility9.4 ± 0.810.6 ± 0.70.192*
ICS [median (mean ± SD)] beclomethasone eq.1600 μg* (2064 ± 939.7)800 μg (614 ± 218.6)<0.0001*
OCS [median (mean ± SD; min–max)] prednisolone eq.10 mg (14.15 ± 11.8; 2–50)ND
BMI (kg/m2)28.5 ± 0.625.0 ± 0.4<0.0001*
ACQ (Juniper)2.03 ± 0.11.03 ± 0.7<0.0001*
QoL (SGRQ)45.9 ± 2.122.5 ± 2.0<0.0001*
CRP (mg/l)6.1 ± 0.93.5 ± 0.60.092*
Atopy (%)43480.642
FENO (ppb)46.3 ± 6.240.1 ± 4.10.962*
Sputum cells (×106)3.34 ± 1.02*1.83 ± 0.340.455*
Sputum eosinophils (%)16.7 ± 3.49*5.79 ± 1.710.018*
Sputum neutrophils (%)42.2 ± 3.744.2 ± 4.40.338*

Stability of a phenotype based on physiological variables

Data from 164 patients at baseline and 127 at 1 year of follow-up were complete and enabled allocation to clusters in line with an algorithm proposed by Moore et al. [6]. At baseline, 17.7% of patients were allocated to cluster 1, 43.9% to cluster 2, 12.2% to cluster 3, 14% to cluster 4 and 12.2% to cluster 5 (Fig. 1A). After 1 year of follow-up, the allocation to particular clusters was stable in 76.4% and changed in 23.6% of the whole asthma population and was stable in 70% and changed in 30% of severe asthmatics (Fig. 2A).

Figure 1.

Allocation to particular asthma phenotypes at baseline (V1) based on physiological parameters (A) (baseline lung function and reversibility) and medical history (age of onset of the disease) [11] and based on biomarkers (B) (eosinophils ‘eos’ and neutrophils ‘neu’ in induced sputum) [10].

Figure 2.

Stability of phenotypes defined by physiological parameters (A) and biomarkers (B) after 1 year of follow-up.

Stability of a phenotype based on sputum eosinophils as biomarkers

One hundred and three patients produced evaluable sputum samples at baseline and 52 at 1 year of follow-up, which enabled allocation to inflammatory phenotypes in line with an algorithm proposed by Hastie et al. [5]. At baseline, 24.3% of patients were allocated to cluster 1, 20.4% to cluster 2, 20.4% to cluster 3 and 35% to cluster 4 (Fig. 1B). After 1 year of follow-up, the allocation to particular clusters was stable in 57.7% and changed in 42.3% of the whole asthma population and was stable only in 51.4% and changed in 48.6% of severe asthmatics (Fig. 2B).

Comparison between stability of phenotypes based on physiological parameters or biomarkers

Allocation to particular phenotypes defined by physiological parameters and biomarkers at baseline and after 1 year of follow-up is presented in Fig. 3. Significantly more patients changed allocation to phenotype based on biomarkers (sputum eosinophils and neutrophils) in the whole asthma population (42.3% vs 23.6 biomarker vs physiological variables clustering, P = 0.034) and in subgroups: severe asthma cohort (48.6% vs 30.0%, biomarker vs physiological variables clustering) and mild-to-moderate asthma cohort (32.8% vs 15.5%, biomarker vs physiological variables clustering). There was a tendency (P = 0.097) to lower stability of phenotypes in severe asthmatics both for physiological variables clustering (30.0% vs 15.5% patients changed allocation in SA vs MA, respectively) and for clustering based on eosinophils and neutrophils in induced sputum (48.6% vs 32.8% patients changed allocation in SA vs MA, respectively).

Figure 3.

Allocation to particular phenotypes defined by physiological parameters (A) [11] and biomarkers (B) [10] at baseline and after 1 year of follow-up (percentage of the study group for those subjects where all data points were available, n = 127 for physiological parameters and n = 52 for biomarkers). Migration of particular patients (number of patients) is shown with arrows.

Longitudinal data

Changes in lung function (individual and mean values of FEV1% predicted) over the duration of the study for the severe and mild-to-moderate asthma cohorts are shown in Figs S5 and S6 (online repository). The ICC (95% CI) for FEV1 (single measures) was 0.585 (0.506–0.661, P < 0.001) and for average measures was 0.849 (0.804–0.887, P < 0.001). Changes in eosinophils and neutrophils over the duration of the study for the severe and mild-to-moderate asthma cohorts are shown in Figs S7 and S8 (online repository). The ICC (95% CI) for eosinophils (single measures) was 0.479 (0.386–0.573, P < 0.001) and for neutrophils 0.465 (0.297–0.638, P < 0.001). Intraclass correlation coefficient for eosinophils (average of replicate measures) was 0.786 (0.716–0.843, P < 0.001) and for neutrophils 0.777 (0.628–0.876, P < 0.001). These results consistently show improved reliability with repeated measures and that measurements of physiological variables (lung function) had better repeatability than measurements of cellular biomarkers (eosinophils and neutrophils in induced sputum).

The change in phenotype during 1 year of follow-up was compared with change in ICS dose during this time (Fig. S3). The phenotype change was not influenced by the change in the dose of inhaled corticosteroids used either in phenotypes defined by physiological variables (lung function, reversibility) and medical history (P = 0.275, Kruskal–Wallis test), or in phenotypes defined by biomarkers (eosinophils and neutrophils in induced sputum; P = 0.935; Fig. S9). Furthermore, 26 severe asthma patients were on oral corticosteroids (OCS) during the follow-up period. The phenotype change in this group was not influenced by the change in the dose of OCS (P = 0.563 for phenotypes defined by the physiological parameters and P = 0.479 for phenotypes defined by biomarkers, data not shown).

The change in phenotype during 1 year of follow-up was also compared with the number of exacerbations reported during this time (Fig. S4). The phenotype change was not influenced by the number of exacerbations either in phenotypes defined by physiological variables (lung function, reversibility) and by medical history (P = 0.55), or in phenotypes defined by biomarkers (eosinophils and neutrophils in induced sputum; P = 0.15). Changes in the level of asthma control (defined with the use of Asthma Control Questionnaire), quality of life (defined with the use of St George's Respiratory Questionnaire) and exhaled NO (FeNO) are presented in Figs S10, S11 and S12, respectively, of the online supplement. Finally, only five patients changed allocation using both evaluated methods of defining phenotypes during the follow-up year.

Discussion

To our knowledge, this is the first study in adults with asthma that compares stability over 1 year of follow-up of phenotypes defined by physiological variables or sputum cells as biomarkers. We found that allocation to phenotypes based on lung function and age of onset of the disease was more stable than allocation to phenotypes based on eosinophils and neutrophils in induced sputum. Moreover, phenotypes of mild-to-moderate asthmatics seem to be more stable than those of severe asthma patients. Although the proportion of subjects in the different groups remained similar after 1 year of follow-up, a significant number of patients shifted groups, meaning that variability in group allocation on the individual patient level was considerable. This is an important aspect if initial patient stratification should be used to allocate patients to a particular treatment. The variability in group assignment also raises concerns if patients should be studied for biomarkers and ‘omics to define endotypes. We interpret our findings in this analysis as a strong indication that repeated measures are required to obtain reliable phenotyping data on subjects with asthma, in particular when using biomarkers.

Previous studies in adults with asthma have indicated reasonable stability of inflammatory biomarkers in sputum (reviewed in [7]). The data in the literature are, however, conflicting. Many studies have been short term and in small cohorts of patients, with focus on eosinophils in induced sputum [11, 16]. van Veen et al. [17] and Simpson et al. [18] thus reported that sputum eosinophilia is a consistent feature overtime despite treatment with high doses of corticosteroids, and they found inflammatory subtypes of asthma to be reasonably stable. However, several other inflammatory phenotypes may be identified by sputum analysis in asthmatics [19]. Al-Samri et al. [20] analysed changes in sputum eosinophils and neutrophils over 1 year in 37 severe and 24 moderate adult asthmatics and found that sputum inflammatory profiles often changed independently of the severity of asthma. Fleming et al. [10] likewise found that inflammatory phenotype based on induced sputum measurements was unstable in children with asthma. McGrath et al. [13] reported on longitudinal sputum cytology data from 995 subjects with asthma enrolled in different clinical trials and found that a significant proportion of patients shifted groups. In addition, approximately half of mild-to-moderate asthmatics had persistent noneosinophilic inflammation, which may be associated with poor response to currently available anti-inflammatory therapy. Green et al. [21], in a study over 12 months, also found the noneosinophilic asthma phenotype to be relatively stable. Although the repeatability and reliability of sputum cellularity measurements in our study were reasonably high, it was nevertheless less reliable than physiological variables such as FEV1. It is known that FEV1 has lower variability than other lung function measures [22].

Several confounding factors, including change in medications, adherence to treatment, exacerbations, pollution and allergens' exposure, may have a potential influence on phenotype stability in asthma patients. It seems unlikely that phenotype variability in our study was related to differences in the dose of inhaled or oral corticosteroids as there were no differences in the level of steroid treatment between the studied clusters (Fig. S3). Similarly, there were no differences due to the number of exacerbations during 1 year of follow-up (Fig. S4). One limitation the BIOAIR project shares with other observational studies of patient cohorts is that the compliance with the prescribed medication has not been rigorously evaluated. It has been estimated that the overall compliance with inhaled controller therapy in asthma is no more than 50% [21]. In our study, an effort was, however, made to improve adherence by reminding the patients to take the inhaled corticosteroids every day when responding to the questions in the electronic daily card. Although we did not collect quantitative data on environmental triggers such as allergen exposure, infections or air pollution, it seems reasonable to assume that clinically important differences in environmental factors would be reflected as differences in the number of exacerbations between the clusters. However, the exacerbations were carefully monitored by daily recordings of lung function and symptoms [8], and there were no differences in exacerbations that could be associated with changes in phenotype cluster. As the dropout rate (18.3% of the included cohort, n = 31) was small, it was unlikely to influence the study results. Moreover, for the statistical analysis, only data from those subjects with complete data set were included. As data were expressed as percentage change, another possible limitation would be the smaller sample size for the inflammatory than for the physiological parameters (n = 52 vs n = 127), and this will need to be addressed in future studies.

We also evaluated whether it was the same subjects who changed allocation during 1 year of follow-up to phenotypes defined by physiological variables or by cellularity in induced sputum. However, only five patients consistently changed allocation in both applied algorithms. In contrast, most patients who were unstable according to one method of clustering were stable using the other strategy. Taken together, the findings support the interpretation that biomarkers (in this case, eosinophils and neutrophils in induced sputum) and clinical/physiological variables such as lung function, the number of exacerbations and the level of required asthma treatment [23, 24] reflect different processes that cannot be assumed to correlate, even if being partly dependent variables.

In summary, our comparison of two published strategies for clustering of asthma found that the inflammatory phenotypes determined by the cellularity in induced sputum were less stable than those defined by physiological variables. This is yet another confirmation that asthma is a variable inflammatory disease, and therefore, phenotyping to understand mechanisms of the disease needs to be long term, build on repeated measures and should ideally integrate several different outcomes.

Acknowledgments

We thank Matteo Bottai and Jonas Höijer of the Karolinska Institutet Biostatistical Core Facility for advice. The BIOAIR study was supported by The Fifth and Sixth Framework Programmes of the European Union, contract numbers: QLG1-CT-2000-01185 (BIOAIR) and FOOD-CT-2004-506378 (GA2LEN), and several national funding bodies (Sweden: Heart-Lung Foundation, MRC, Asthma and Allergy Foundation, the Stockholm County Council Funds (ALF), the Swedish Strategic Research Foundation (SSF) and the KIAZ-SciLifeLab ChAMP collaboration on Translational Science; Greece: unrestricted competitive research grant ‘The Herakleitos project 2002’, from the Hellenic Ministry of Education; UK: Patrick Mallia was supported by unrestricted grant from GSK). Maciej Kupczyk was a scholarship fellow of the Wenner-Gren Foundations and was supported by the Bernard Osher Initiative for Research on Severe Asthma at the Karolinska Institutet. The BIOAIR study also received unconditional support from AstraZeneca Sweden, Vitalograph Inc, Aerocrine AB and Amedon GmbH.

Author contributions

Primary data aggregation and statistical analyses were performed by MK and SED, and who then discussed and interpreted with the other co-authors at face-to-face meetings. MK wrote the first draft of the manuscript. Revisions were made by MK together with SED and in an interactive process including all co-authors. All authors participated in the study design, conduction of the trial, collection and interpretation of data, and critical review of the draft versions of the report.

Conflicts of interest

The authors declare that they have no conflicts of interest.

Note

  1. 1

    Deceased

Appendix 1

BIOAIR – Longitudinal Assessment of Clinical Course and BIOmarkers in Severe Chronic AIRway Disease Contributors to the BIOAIR study

Amsterdam: Els Weersink, Athens: Mina Gaga, Nikos Papadopoulos, Erasmia Oikonomidou, Eleftherios Zervas, Ferrara: Marco Contoli, Ghent: Romain A. Pauwels1, Guy F Joos, Isabelle de Rudder, Vanessa Schelfhout, Hamburg/Grosshansdorf: Kai Richter, Daisy Gerding, Helgo Magnussen Heraklion: Katerina Samara, Maria Plataki, Eva Papadopouli, Krakow: Andrzej Szczeklik1, Bozena Ziolkowska-Graca, Aleksander Kania, Agnieszka Gawlewicz-Mroczka, Mariusz Duplaga, Ewa Figiel, Leiden: Klaus F. Rabe, Pieter S. Hiemstra, Stefanie Gauw, Ilonka van Veen, Leuven: Johan C. Kips, London: Sebastian L. Johnston, Patrick Mallia, Deborah A. Campbell, Douglas S. Robinson, Luebeck: Frank Kanniess, Modena: Leo M. Fabbri, Micaela Romagnoli, Montpellier: Isabelle Vachier, Catherine Devautour, Lahouari Meziane, Palermo: A. Maurizio Vignola1, Elisabetta Pace, Mirella Profita, Southampton: Susan J. Wilson, Lorraine Hewitt, John Holoway, Stockholm: Roelinde JM Middelveld, Katarina Damm, Ingrid Delin, Marianne Eduards, Alexandra Ek, Tommy Ekström, Flora Gaber, Agneta Gülich, Anna James, Lovisa E. Johansson, Östen Karlsson, Maria Kumlin, Ingrid Martling, Marianne Olsson, Maria Skedinger, Shushila Haque.

Ancillary