Abstract
Background: Asthma is now recognized as a heterogeneous collection of distinct phenotypes. We aimed to uncover potential asthma phenotypes through clustering of clinical variables in the pan-European U-BIOPRED cohort.
Methods: Clinical data from subjects classified as severe (n=209) or non-severe (n=51) asthmatics [Bel et al. Thorax 2011] were analyzed through hierarchical clustering, k-means or partitioning around medoids. Quality of the results was assessed by bootstrapping analysis, also known as consensus clustering [Monti et al.Machine Learning 2003].
Results: Using a reduced set of 16 clinical variables, stable preliminary results were obtained, indicative of potentially new asthma phenotypes. K-means analysis of the 260 mild to severe asthma subjects identified 4 stable clusters (Table), defined mainly by baseline lung function, bronchodilator response and asthma control.
Conclusion: This preliminary analysis shows that clinical variables from the U-BIOPRED cohort can be used to derive phenotypic asthma clusters that are similar but not identical to previously published ones. The significance and stability of these clusters are being benchmarked through topology data analysis and will be further examined in an unrelated asthma cohort, ADEPT.
Funded by the Innovative Medicines Initiative (U-BIOPRED n°115010; eTRIKS n°115446).
- © 2014 ERS