Fig. 4

Clustering of the study participants. Clustering of the training Innsbruck (IBK) cohort participants in respect to the survival-associated factors identified by Elastic Net modeling (Fig. 2) was investigated by PAM (partition around medoids) algorithm with cosine distance. Numeric clustering features were median centered prior to the clustering. Cluster assignment in the training Linz/Vienna cohort (LZ/W) was done by an inverse distance weighted 7-nearest neighbor classifier. Numbers of individuals assigned to the PAH clusters are presented in the plot captions or legends. A PAH cluster assignment overlaid on the 2-dimensional cosine-distance UMAP (Uniform Manifold Approximation and Projection for Dimension Reduction) layout plots. Percentages of variance associated with the components are indicated in the plot axes. B Differences in the clustering features between the PAH clusters were assessed by Mann–Whitney test corrected for multiple testing with Benjamini–Hochberg method. Normalized, median-centered values of the clustering factors are shown in violin plots. Points represent single observations. P values are indicated in the Y axes. CI: cardiac index; NT-pro-BNP: N terminal pro brain natriuretic peptide; RDW: red blood cell distribution width; PVR: pulmonary vascular resistance; RAA: right atrial area; SMWD: six minute walking distance