Chronic obstructive pulmonary disease phenotypes. Past, present, and future.

Abstract

Although chronic obstructive pulmonary disease (COPD) was arbitrarily defined in the early 1960s as a single disease characterized by chronic airflow obstruction (1), it has long been recognized as a heterogeneous group of related disorders. In the 1950s, Dornhorst described two extreme phenotypes of patients with severe respiratory disease: pink puffers and blue bloaters (2). More recently, the heterogeneity of patients with COPD was elegantly demonstrated in the Evaluation of COPD Longitudinally to Identify Predictive Surrogate Endpoints (ECLIPSE) study. The authors showed that for a given level of airflow limitation, there were major differences in dyspnea, exercise capacity, rates of exacerbations, and health status (3), reinforcing interest in understanding the determinants of heterogeneous clinical presentation. At that time, the concept of COPD phenotypes was defined by group of experts as “a single or combination of disease attributes that describe differences between individuals with COPD as they relate to clinically meaningful outcomes (symptoms, exacerbations, response to therapy, rate of disease progression, or death)” (4). The goals of phenotyping are to help in the identification of specific biological pathways, to enrich clinical trials in patients at high risk for outcomes of interest, and to pave the way for provision of individualized care. Because multiple patient characteristics have been identified that might contribute to identifying potential phenotypes, novel mathematical methods (e.g., principal component and cluster analyses) have been introduced in the COPD field with the goal of analyzing large amounts of clinical, imaging, and biological data (5). In recent years, several studies have used these mathematical approaches to analyze cohorts of patients with COPD with the goal of identifying clinically relevant COPD phenotypes (reviewed in Reference 6). Some phenotypes appeared relatively reproducible across studies, including two severe phenotypes with poor prognosis: young patients with severe respiratory disease but few cardiovascular comorbidities and older patients with less severe respiratory disease but high rates of metabolic and cardiovascular comorbidities (5, 6). However, all previous cluster studies had limitations related to relatively small numbers of patients, lack of extensive data for patient characterization, and lack of prospective validation of phenotypes in some of the studies. In this month’s issue of AnnalsATS, Rennard and colleagues (pp. 303–312) sought to identify clinically relevant COPD phenotypes, using data generated by the ECLIPSE study (7). The strengths of their analysis include the relatively large number of patients; standardized collection of clinical, physiological, radiological, and biomarker data; and a 3-year standardized follow-up period to assess clinically relevant outcomes. However, the patients included in the ECLIPSE study were not consecutive patients with COPD in the community but, rather, a select group of patients at tertiary care centers willing and able to participate in a 3-year study. The majority of those patients were receiving COPD maintenance treatment, and the intensity and appropriateness of such treatment likely affected the various parameters used in the cluster analysis, as well as measured outcomes (e.g., symptoms, exacerbations). These potential confounders should be recognized when interpreting this study. Were the clusters derived from this analysis clinically meaningful? Cluster A, composed of patients with mild COPD who had excellent outcomes, and cluster D, composed of patients with severe airflow obstruction who fared poorly, confirm what we see in everyday clinical practice. Cluster C, the comorbid cluster, supports increasing awareness that comorbid conditions may be as important in determining COPD phenotypes as characteristics of the lung disease itself. These observations confirm the validity of clusters that have been identified in earlier studies (6). Cluster B was the smallest cluster but is intriguing in that it identified patients with mild to moderate emphysema who experienced rapid progression despite a rate of current smoking (31%) that was comparable to that of the other groups. The most dissatisfying cluster is Cluster E, the mixed group, which accounted for half of the patients. Essentially, this fifth cluster is composed of patients who could not be clustered otherwise. Cluster E represents the largest group, with variable outcomes. It is a reminder of the limitations of cluster analysis and the considerable heterogeneity of COPD. The authors of this study recognize that many of the variables included in the ECLIPSE study are not usually measured in the clinical setting, and the analytic method does not use well-defined cutoff values to