Table 2.

Discrimination measures and hazard ratios for prognostic groups based on percentiles of the prognostic index (PI) evaluated in the training and validation datasets

Measure	Training Estimate (SE)	Validation 1 Estimate (SE)	Validation 2 Estimate (SE)
Harrell c-index	0.659 (0.018)	0.631 (0.018)	0.644 (0.030)
Gönen and Keller K	0.628 (0.003)	0.629 (0.003)	0.635 (0.006)
Explained variation $\left(r_D^2\right)$	0.133 (0.029)	0.124 (0.033)	0.140 (0.055)
Heller’s estimated explained relative risk	0.178 (0.043)	0.173 (0.048)	0.196 (0.080)
	Training HR (95% CI)	Validation 1 HR (95% CI)	Validation 2 HR (95% CI)
Prognostic index category (<16th percentile)	Referant	Referant	Referant
16th to <50th percentile	1.48 (0.97–2.25)	1.49 (0.86–2.58)	0.93 (0.47–1.83)
50th to <84th percentile	2.91 (1.96–4.32)	2.27 (1.33–3.85)	2.32 (1.24–4.35)
84th percentile or greater	4.44 (2.86–6.89)	4.17 (2.40–7.24)	3.07 (1.56–6.02)

Standard errors (SE) for the c-index, K, and $r_D^2$ were estimated from 200 bootstrap samples. Hazard ratios (HR) are relative to the reference group of PI <16th percentile CI confidence interval