Deliu M, Yavuz ST, Sperrin M, Sahiner U, Sackesen C, Custovic A, Kalayci O

Hierarchical clustering identifies novelsubgroups of childhood asthma

Deliu M, Yavuz ST, Sperrin M, Sahiner U, Sackesen C, Custovic A, Kalayci O

Institute of Population Health, University ofManchester, Manchester, United Kingdom

PaediatricAllergy, GATA School of Medicine, Ankara, Turkey

Paediatric Allergy,Guven Hospital, Ankara, Turkey

Paediatric Allergy and Asthma Unit, HacettepeUniversity School of Medicine, Ankara, Turkey

Imperial College London, London, United Kingdom

Introduction: Childhood asthma is aheterogeneous condition. Previous unsupervised cluster analyses have indicated thepresence of asthma subtypes in childhoodthat differ from the adult population.

Methods: We recruited 263 asthmatic children aged 6 – 17 years from the GATAUniversity Military Hospital, Ankara, Turkey. Hierarchical clustering using Ward’smethod was performed on 68 binarisedvariables including demographic data, atopic sensitization, lung function, medicationuse, peripheral eosinophilia, and markersof asthma severity. A dendrogram wasgenerated to identify the number of resulting clusters.

Results: The optimal solution identified sixclusters of asthma school aged children.Children in Cluster 1 (n = 60) had earlyonset asthma, poor lung function, highermedication use, and more than 2 asthmaattacks per year. Cluster 2 (n = 47) comprised of non-atopic children with normallung function and well controlled asthma.Cluster 3 (n = 33) comprised of childrenwith later onset asthma, high peripheraleosinophils, sensitization to dog and grass,high medication use and poor lung function. Cluster 4 (n = 26) comprised of olderchildren with later onset and severe poorlycontrolled asthma. Cluster 5 (n = 49)included older aged children with mildsymptoms despite poor lung function.Finally, children in Cluster 6 (n = 48) hadmild atopic asthma and normal lung function. Predictors of cluster membership wereage of onset, atopic status, lung function,medication use, and asthma severity. Theclustering model was repeated multipletimes by varying the distance and linkagemeasures.

Conclusion: Novel clusters of childhoodasthma have been identified in our cohortalong with similar ones already published.Further validation is needed to test the stability of clusters in other populations.Identifying distinct phenotypes can lead tobetter treatment targeting