Unsupervised Learning Technique Identifies Bronchiectasis Phenotypes With Distinct Clinical Characteristics and Prognosis

BACKGROUND: Unsupervised learning technique allows researchers to identify different phenotypes of diseases with complex manifestations.OBJECTIVES: To identify bronchiectasis phenotypes and characterise their clinical manifestations and prognosis.METHODS: We conducted hierarchical cluster analysis to identify clusters that best distinguished clinical characteristics of bronchiectasis. Demographics, lung function, sputum bacteriology, aetiology, radiology, disease severity, quality-of-life, cough scale and capsaicin sensitivity, exercise tolerance, health care use and frequency of exacerbations were compared.RESULTS: Data from 148 adults with stable bronchiectasis were analysed. Four clusters were identified. Cluster 1 (n = 69) consisted of the youngest patients with predominantly mild and idiopathic bronchiectasis with minor health care resource use. Patients in cluster 2 (n = 22), in which post-infectious bronchiectasis predominated, had the longest duration of symptoms, greater disease severity, poorer lung function, airway Pseudomonas aeruginosa colonisation and frequent health care resource use. Cluster 3 (n = 16) consisted of elderly patients with shorter duration of symptoms and mostly idiopathic bronchiectasis, and predominantly severe bronchiectasis. Cluster 4 (n = 41) constituted the most elderly patients with moderate disease severity. Clusters 2 and 3 tended to have a greater risk of bronchiectasis exacerbations (P = 0.06) than clusters 1 and 4.CONCLUSION: Identification of distinct phenotypes will lead to greater insight into the characteristics and prognosis of bronchiectasis.