Objective: Asthma is divided into various distinct phenotypes on the basis of clinical characteristics, physiological findings, and triggers, and phenotyping is usually performed in a hypothesis-driven univariate manner. However, phenotyping can also be performed using computer algorithms to evaluate hypotheses-free relationships among many clinical and biological characteristics. We aimed to identify asthma phenotypes based on multiple demographic, clinical, and immunological characteristics.
Methods: Cluster analysis in R v3.5.0 was performed using asthma patient data. A total of 170 adult patients with asthma (diagnosed according to the GINA recommendations) were recruited to the study. All patients completed questionnaires about their smoking history and underwent physical examination, spirometry, skin-prick test, blood sample collection to evaluate peripheral blood cell counts and serum IgE, periostin, and interleukin (IL)-33 levels, as well as body mass index measurements. Data normality was checked with histograms and QQ plots. Hierarchical clustering was performed using Ward's linkage with Ward's clustering criterion. The optimal number of clusters was validated using the Dunn criterion as well as by comparing different clustering algorithms using the clValid package.
Results: Three clusters characterizing asthma phenotypes were identified: (1) early-onset, highly atopic, and eosinophilic asthma associated with male sex and high levels of IL-33 and periostin; (2) late-onset, eosinophilic asthma associated with female sex and low levels of IL-33 and periostin; and (3) late-onset, obese, neutrophilic asthma associated with female sex, persistent airway obstruction, and very low IL-33 and periostin levels.
Keywords: Asthma; cluster; interleukin-33; periostin; phenotype.