Andreas Halner, Respiratory Medicine Unit, Nuffield Department of Medicine, University of Oxford, Oxford, UK
Mona Bafadhel, Respiratory Medicine Unit, Nuffield Department of Medicine, University of Oxford, Oxford, UK
Chronic obstructive pulmonary disease (COPD) affects millions of people worldwide. It is now clear that COPD is heterogeneous, different components of the disease being present in different patients. Yet, the diversity of COPD pathophysiology, severity and how this relates to disease prognosis and treatment outcomes is far from understood. In order to address this, mathematical techniques such as cluster analysis have been employed to identify subgroups or clusters of COPD patients with differing disease attribute profiles. However, significant methodological shortcomings call into question the validity of the COPD clusters identified in such studies. Furthermore, few published studies relate COPD clusters to underlying disease mechanisms and treatment outcomes. Where this has been addressed, progress has particularly been made for patients with an eosinophilic-predominant profile. In order to maximise the usefulness of COPD cluster analysis studies, we propose that future studies must implement more stringent methodologies and focus on COPD inflammatory biology.
Keywords: Chronic obstructive pulmonary disease. Cluster analysis. Machine learning.
