External validation of a Cox prognostic model: principles and methods

Table 5 Calibration by risk group

Risk group	t	Kaplan-Meier								Predicted
	(yr)	Derivation				Validation				Validation
		N	Ev.	$\hat{S} (t)$	SE	N	Ev.	$\hat{S} (t)$	SE	$\bar{S} (t)$
1. Good	2	247	90	0.90	0.02	124	28	0.88	0.03	0.88
	5			0.71	0.03			0.68	0.05	0.73
2. Fairly good	2	526	270	0.82	0.02	277	103	0.85	0.02	0.78
	5			0.56	0.02			0.59	0.03	0.54
3. Fairly poor	2	526	401	0.60	0.02	225	123	0.61	0.03	0.63
	5			0.31	0.02			0.37	0.04	0.32
4. Poor	2	247	204	0.44	0.03	60	45	0.53	0.07	0.47
	5			0.20	0.03			0.11	0.05	0.16

Values shown are Kaplan-Meier estimates and standard errors of recurrence-free survival probabilties in the three groups at two times. Values labelled Predicted were predicted from the derivation dataset by applying the PI to the smooth baseline survival estimate at the individual level in the validation dataset, and averaging across each risk group. N and Ev. denote the number of patients and events in each group.

ISSN: 1471-2288