Table 3 Statistical methods used for cluster analysis and outcomes tested in the included studies of clinical phenotypes for COPD

Burgel et al. (2010)

8 variables, expert opinion

Cluster using k-means after variable reduction using PCA

BOD^§ index

Burgel et al. (2012)

8 variables, expert opinion

Cluster using k-means after variable reduction using PCA

Mortality rates after 3.35 years of follow-up

Burgel et al. (2012)

18 variables, expert opinion

Hierarchical clustering with Wards method after variables reduction using PCA and MCA

Mortality rates after 17.2 months of follow-up

Cho et al. (2010)

43 variables (including 12 SNPs from 5 genes), expert opinion, and selection of genes included in previous genetic association studies

Cluster using k-means after variable reduction using factor analysis

Exacerbations/year over 3.3 years (retrospective)

DiSantostefano et al. (2013)

36 variables analyzed, co-linear variables dropped, expert opinion

Tree-based supervised cluster analysis using modified recursive partitioning

Decreased annual rate of exacerbations with SFC compared to SAL

Garcia-Aymerich et al. (2011)

224 variables, all variables collected (after excluding those with additive relationships or resulting from categorizations)

Cluster using k-means clustering method (PKM)

Admissions and mortality rates during a 4-year follow-up

Spinaci et al. (1985)

4 variables, expert opinion

Cluster using k -means clustering method (PKM)

Analysis of contingency tables

Vanfleteren et al. (2013)

13 variables, based on “clinical relevance and methodological possibilities to objectify the comorbidities”

Self- organizing maps (SOMs, or Kohonen maps) used to order patients by their overall similarity with regards to comorbidities. Clusters generated using a hybrid algorithm that applied classical hierarchical method of Ward on top of the SOM topology.

Updated BODE^¶ index, Framingham 10-year risk