Figure 1.

Good and poor prognosis group separated by the PI applied for the testing sample of size n. Concretely, $\left\{i;{\mathrm{PI}}_i\left(\widehat{\alpha }\right)\le c\right\}$ is a good prognosis (low-risk) group, and $\{i;{\mathrm{PI}}_i\left(\widehat{\alpha }\right)>c\}$ is a poor prognosis (high-risk) group. If the classification is successful, survival functions exhibit $S_{\mathrm{Good}}\left(t|c\right)>S_{\mathrm{Poor}}(t|c)$. The cut-off value $c$ can be the median of the PI values calculated from the training samples.