Specifically, the partitioning around medoids (pam) clustering algorithm [ 29 ] with two centers predicted one cluster with 52 good outcome and 56 poor outcome patients, and another cluster with 56 good outcome and only 19 poor outcome events, which was highly significant under Fisher's exact test ( P = 0.0004; Figure 3c ).
Similarly, relative to the CC + /IR + subclass, the SR + subgroup had an HR of 2.35 (95% CI 1.13 to 4.88; P = 0.02), whereas for the CC + subgroup it did not quite reach statistical significance (HR 1.80, 95% CI 0.83 to 3.89; P = 0.13).
For the other three genes ( HLA-F , SPP1 , and XCL2 ), P values did not reach statistical significance ( P about 0.2), yet their expression profile trends were entirely consistent with those found in the integrated cohort, thus confirming their role as members of a robust prognostic module (Additional data files 6 and 7).