A clustering analysis of eosinophilic asthmatics

editorial reviewedIntroduction Eosinophilic asthma is well recognized asthma phenotype associated with disease severity. Yet it is believed that clinical heterogeneity may exist between patients displaying eosinophilic airway inflammation. Cluster analysis is a well-known unsupervised learning methodology that considers multiple variables in order to create coherent subsets among a large group of patients. Aim and objects The main purpose of this study was to perform a cluster analysis on a large group of eosinophilic asthmatics based on sputum analysis. Methods 426 participants were selected from the CHU Liege Asthma Clinic Data Base based a sputum eosinophil count ≥ 3%. Missing values in the original dataset were handled by multiple imputation. After imputation step, HCPC (Hierarchical Clustering on Principal Components) was applied to determine the efficient number of cluster and then divide patients to proper cluster. In the final step of clustering, consensus clustering was used based on 4M method to assign each individual to cluster. Clustering was also applied to subgroups of ICS naive patients one the hand and high dose ICS treated patients (≥1000 µg propionate fluticasone) on the other hand. None of the selected patients were receiving biologics at the time of the visit. Results On the whole cohort of eosinophilic asthmatics cluster analysis revealed two subgroups identified as phenotype 1 (n=276) and phenotype 2 (n=150). Phenotype 1 included younger patients (50 yrs), with a high proportion of atopic patients (67%), lower treatment burden (median ICS dose 500 µg eq beclomethasone) and preserved lung function (FEV1 93% predicted) a relatively good asthma control. Phenotype 2 included older patients (59 yrs) with a low proportion of atopic (36%), high treatment burden (ICS dose 2000 µg eq beclomethasone), more intense eosinophilic airway inflammation (sputum 16%), greater systemic inflammation, and impaired lung function (FEV165% predicted) and poorly controlled asthma. Clustering on ICC naïve and high dose ICS treated patients also yielded two clusters mainly structured by age, atopic status, intensity of granulocytic airway inflammation. Conclusion Our results points to different phenotypes in eosinophilic asthmatics with the atopic trait being generally associated with an achievable control of the disease with low dose ICS despite eosinophilic airway inflammation. In contrast there is an aggressive eosinophilic phenotype also usually featuring intense airway neutrophilic inflammation, not associated with atopic trait and leading to impaired lung function and poor asthma control despite heavy treatment burden