Unsupervised versus Supervised Identification of Prognostic Factors in Patients with Localized Retroperitoneal Sarcoma: A Data Clustering and Mahalanobis Distance Approach

Figure 3

Decision tree picture of the supervised- and unsupervised-based partitioning. Panels (a) and (c) depict the first three branches (splits) of the decision tree obtained by the numeric, supervised coding (scales reported in Table 2) of the 5 best performing variables in Table 3 (histology, grading, DFI, relapse pattern, and 1st-type treatment at recurrence). Panels (b) and (d) refer to the same data coded as alphanumeric symbols, hence loosing any quantitative specificity assigned by supervisors. The rectangular boxes in panels (c) and (d) contain the values, namely, an indication of the % of explained variability. Ideally, repeated partitioning should eventually produce a total = 1. Modeling has been carried out by the Partition Platform of JMP, version 13.

(a)

(b)

(c)

(d)