Table II.

Five-fold validation.

Data subset used for validation	Calibration subset (n)	Validation subset (n)	Splitting values					AUC calibration	AUC validation
			Equation A	t3 (h)	Equation B Low	Equation B High	cells66h
All data	3275		11.481	42.905	0.341	0.578	7.5	0.653
1	2620	655	11.481	42.885	0.441	0.577	7.5	0.655	0.648
2	2620	655	11.330	43.085	0.359	0.578	7.5	0.653	0.645
3	2620	655	11.481	42.905	0.341	0.578	7.5	0.656	0.640
4	2620	655	9.511	42.910	0.341	0.602	7.5	0.654	0.646
5	2620	655	11.999	42.905	0.441	0.578	7.5	0.656	0.642
	Mean		11.16	42.94	0.38	0.58	7.50
	SD		0.96	0.08	0.05	0.01	0.00

The stability of the algorithm structure was verified by performing five calibration procedures on the same variables and in the same order as the final output obtained by the rpart routine. Each split was first calibrated by excluding 1/5 of the data at a time (calibration subset). The calibrated parameters obtained from the 4/5 of the data which was not excluded were used to generate scores for both the calibration subset and the remaining 1/5 of the data (validation subset). The AUC values are shown both for the scored calibration subset (AUC calibration) and the validation subset (AUC validation). Equation A, (t3−tPNf); Equation B, (t5−t3)/(t5−t2); cells66h, number of cells at 66 h. AUC, area under the curve.