Fig. 3

Evaluation of reproducibility for clinical phenotypes. a Hierarchical cluster analysis for principal component scores of centroids of each phenotype. Phenotype analysis, silhouette analysis, and latent class analysis were used. Patient data with phenotype information were combined into a single data sheet, PCA was performed, principal component scores for each phenotypic centroid were calculated, and hierarchical cluster analysis was performed. b PCA for the centroid of each phenotype. The horizontal axis is the first principal component axis, and the vertical axis is the second principal component axis. Plot size indicates the number of patients. The distribution of cluster 8, which is strongly associated with death, was clearly different from that of clusters 1–7. There was no dissociation in the position of the clusters between the derivation and validation cohorts, suggesting that homology between the cohorts was maintained. PCA, principal component analysis