Chronic obstructive pulmonary disease (COPD) is a heterogeneous disease with pulmonary and extra-pulmonary manifestations. COPD heterogeneity is poorly understood and insights into molecular mechanisms is required to provide better treatment with targeted therapies. The inventors collected serum samples from 241 COPD subjects in the COBRA cohort and measured the expression of 1305 proteins using SOMAscan proteomic platform. Modular analyses and clustering of the proteomics were applied to identify disease subtypes. Cluster discoveries were revalidated during a follow up visit, and confirmed in a different COPD cohort. Unsupervised clustering using protein modules identified two clusters within COPD subjects. One cluster presented a higher expression of proteins associated with enhanced immune and pro-survival activities, host defense and cell renewal. Subjects in this cluster had a lower incidence of exacerbations, unscheduled medical visits, emphysema and diabetes. These protein signatures were conserved during a follow up visit, and validated in another COPD cohort. Among the 96 proteins, the inventors identified a small specific signature consisting of 15 proteins that were able to accurately differentiate the two patient clusters. This signature would thus be useful for predicting the clinical outcome of patients suffering from COPD.