Fig. 1.

Schematic overview of the different steps of ClustALL approach (best viewed in colour). ClustALL takes clinical variables as input. First, data complexity is reduced by grouping the features into a dendrogram, assessing the resulting depths, and using Principal Component Analysis (PCA) (green panel). The output is an embedding for each possible depth. Then, stratification is computed considering the combination of different distance measures, clustering techniques, and cluster numbers (K) (purple panel). In the final step, non-robust stratifications are filtered, and the centroids derived from computing Jaccard (coloured green squares) similarity among the robust stratifications (green squares) are considered the final representatives of the stratifications (red panel)