Expression profiling of blood samples from an SU5416 Phase III metastatic colorectal cancer clinical trial: a novel strategy for biomarker identification

Table 5 Effectiveness of LCN2, CD24, and Lactoferrin as a predictor set for discriminating between the control and SU5416 treatment arms.*

1.	All cases pooled (67 cases from both trials)
	Trial arm	Predicted trial arm
		Control	SU5416	% Correct
	Control	26	5	84
	SU5416	6	30	83
	Total	32	35	84
2.	Jackknifed classification matrix for all cases pooled (67 cases from both trials)
		Control	SU5416	% Correct
	Control	26	5	84
	SU5416	8	28	78
	Total	34	33	81
3.	Prediction subset (randomly selected 34 cases) from all cases pooled
		Control	SU5416	% Correct
	Control	13	1	93
	SU5416	4	16	80
	Total	17	17	85
4.	Validation subset (randomly selected 33 cases) from all cases pooled
		Control	SU5416	% Correct
	Control	11	6	65
	SU5416	5	11	69
	Total	16	17	67

*Cross-validation results are displayed for two different approaches. In section 2, one case is dropped at a time and its group membership predicted from the other cases. In sections 3 and 4, cross-validation is carried out by using a randomly selected half of the cases as a training set and the remaining half as a test set. Section 4 summarizes the prediction accuracy achieved when the group in section 3 is used as a training set. Note: When results of one trial were used in predictive classification of results from the other trial, the accuracy in cross-validation was 86% and 77% for the training and testing set, respectively (with a Cohen's kappa value of 0.5507).

ISSN: 1471-2407