Data-Driven Clustering Approach to Identify Novel Phenotypes in Acute Exacerbations of Chronic Obstructive Pulmonary Disease: A Retrospective, Multicentre Cohort Study

Abstract RATIONALE Acute Exacerbations of Chronic Obstructive Pulmonary Disease (AECOPD) is a serious events of COPD characterized by progression heterogeneity, and currently, there is a lack of precise disease phenotypes to identify AECOPD patients with poorer quality of life and higher mortality. This study aimed to reveal novel phenotypes within AECOPD based on multiple biomarkers utilizing a data-driven clustering approach. METHODS The data were obtained from a retrospective, multicentre, 4-hospital registry study taking place in Guangdong, China between October 1, 2009 and October 31, 2022. 131 biomarkers were analyzed to identify novel phenotypes utilizing penalized Cox models and a latent class analysis (LCA) approach. RESULTS The patients (n=13449) who first diagnosed as AECOPD were included in this study. A 6-phenotype LCA model were developed using 12 features in this study. Phenotype 1 was associated with older age (≥73, years) and higher levels of direct bilirubin (Dbil) and lactate dehydrogenase (LDH). Phenotype 2 patients who had higher percentage of lymphocyte (LYM%) but lower percentage of neutrophils (NEU%). Phenotype 3 was characterized by a higher prevalence of generalized cardiovascular disease (gCVD) and lower NEU%. Phenotype 4 was related to higher NEU% but lower LYM%. Phenotype 5 was characterized by older age, a higher prevalence of gCVD, history of surgical trauma and higher levels of Dbil, LDH and NEU%. Patients in phenotype 6 had a higher prevalence of respiratory failure, pulmonary heart disease, gCVD and a higher level of pulse at admission. When controlling for confounders, the patients in phenotype 1 had a higher risk of recurrence (P<0.05) than in phenotype 2, phenotype 3 and phenotype 4. Additionally, in comparison with patients in phenotype 1, the risk of all-cause mortality in phenotype 6 was 1.64 (adjusted hazard ratio [AHR], 95% confidence intervals [CI] 1.16-2.33), and 2.45 (95%CI 1.49-4.01) in the training and validation sets, respectively. CONCLUSIONS Our findings are beneficial for implementing precision health management and personalized intervention measures for patients with AECOPD based on the novel phenotypes. Additionally, this study will lay the foundation for the application of LCA approach in disease prediction.