Skip to main content
. 2020 Jun 1;27(7):1019–1027. doi: 10.1093/jamia/ocaa060

Figure 2.

Figure 2.

Data transformation B: (A) Kaplan-Meier survival curve, (B) MDS plot, and (C) t-SNE plot for 6 unsupervised clusters of CLL patients. Unsupervised machine learning, using k-means clustering with Partitioning Around Medoids (PAM) and the Sokal&Michener distance yields 7 clinical phenotypes with significant differences in time-to-progression (TTP) (P = .0451). (Supplementary Figure 1) Clusters separated by MDS along the first dimension reflect order of overall survival (OS) outcomes.