Fig. 4

Hierarchical clustering and the prognostic value of the clusters. All LBAs were analyzed individually by means of hierarchical clustering by dividing the patient population into four groups. The clusters that differentiate the patients with favorable outcome (marked in red) from those with non-favorable outcome (marked in blue) were identified by permuting all cluster combinations to identify the cluster combination that resulted in the lowest p-value in log-rank analysis. Kaplan–Meier curves of the best permutation within each analyte are depicted and showed prognostic value (except for CTC mRNA-based clustering). A combination of the clustering results of all analytes, defined as the ELIMA.score, revealed good prognostic relevance (e)