"High risk" groups for exacerbations of chronic obstructive pulmonary disease (COPD) in the 2011 Global Initiative for Chronic Obstructive Lung Disease (GOLD) proposal (i.e. groups C and D) [1] include: patients with a forced expiratory volume in 1 s FEV1 <50% reference and <2 exacerbations year -1 (subgroups C1 and D1); patients with ⩾2 exacerbations·year -1 and an FEV1 ⩾50% reference (subgroups C2 and D2); and patients with both FEV1 <50% ref. and ⩾2 exacerbations·year -1 (subgroups C3 and D3) [2][3][4][5]. We hypothesised that these high-risk subgroups will differ in other clinical, functional and biological characteristics and will be associated with different long-term outcomes. We explored this hypothesis in the ECLIPSE (Evaluation of COPD Longitudinally to Identify Predictive Surrogate Endpoints) cohort [6,7].The design and methodology of the ECLIPSE study (www.clinicaltrials.gov with identifier number NCT00292552; GSK study code SCO104960) has been published elsewhere in detail [6]. The study was approved by the Ethics Committees from the participating centres and all participants signed their informed consent [7]. Out of the 2164 GOLD grades II-IV patients included in the ECLIPSE study, 2101 (97%) had complete GOLD 2011 data [4] and were included in the current analysis; 1313 of them (62.5%) were classified as groups C (n=483, 36.8%) or D (n=830, 63.2%), using the modified Medical Research Council dyspnoea score to determine high and low COPD symptoms.These analyses are exploratory and may have low power, since they are based on a subset of the COPD subjects enrolled in ECLIPSE. Results are shown as mean±SD or n (%), as appropriate. Kruskal-Wallis tests and Cochran-Mantel-Haenszel tests were used to assess differences in subject characteristics among groups. Kaplan-Meier curves were constructed to describe the occurrence of the first event for the following outcomes: moderate-to-severe exacerbations of COPD, hospitalisations for COPD exacerbation, and all-cause mortality over the study period. A p-value <0.05 (two sided) was considered statistically significant. No adjustments were made for multiple comparisons. SAS (version 9; SAS Institute, Inc., Cary, NC, USA) was used to conduct all analyses and figures were created via S-PLUS (TIBCO Software, Boston, MA, USA).On the one hand, from the 483 group C patients, 336 (70%) 62 (13%) and 85 (18%) were classified in the subgroups C1, C2 and C3, respectively [4]. As expected, lung function was worse in subgroups C1 and C3 (both groups had clinically comparable airflow limitation), and exacerbations prior to the start of the study were more frequent in subgroups C2 and C3 (both groups had a comparable level of exacerbation history). The remaining variables were similar across groups (table 1). However, of note is that the C2 subgroup included more females and had less emphysema than the other two C subgroups.On the other hand, from the 830 group D patients, 522 (63%), 72 (9%) and 236 (28%) were classified in subgroups D1, D2 and D3, respectively [4] (table 1)....