Table 2.

Results of ROC Curves of ELISA Analyses.

	LCN2/NGAL	TIMP1	CXCL16	TL*	CTL†
con-FPC	Sensitivity	1	0.80	0.81	1	1
	Specificity	1	0.85	1	1	1
	AUC	1	0.885	0.934	1	1
	Cutoff	42.3–102	273	4.17
con-PC	Sensitivity	0.92	0.75	0.62	0.93	0.93
	Specificity	1	0.85	0.85	1	1
	AUC	0.956	0.812	0.778	0.967	0.971
	Cutoff	43.4	282	3.7
con-Endo	Sensitivity	0.73	0.60	0.70	0.82	0.73
	Specificity	1	0.69	0.54	0.82	1
	AUC	0.811	0.600	0.612	0.818	0.802
	Cutoff	69	231	2.9
con-CP	Sensitivity	.825	0.47	0.35	0.89	.89
	Specificity	1	0.69	1	1	1
	AUC	0.859	0.456	0.473	0.912	0.918
	Cutoff	48.4	243	4
CP-FPC	Sensitivity	0.90	0.80	0.95	0.92	0.92
	Specificity	0.92	0.71	0.68	1	1
	AUC	0.958	0.775	0.864	0.983	0.981
	Cutoff	176	273	3.4
FPC-Endo	Sensitivity	0.85	0.70	0.86	0.92	0.92
	Specificity	1	0.80	0.80	1	1
	AUC	0.968	0.810	0.871	0.985	0.985
	Cutoff	301	318	3.69

For the human samples, ROC curves were used to determine the ability of each individual marker to distinguish between the diagnostic groups. The discrimination was determined for each of the following pairs with the group to the right defined as affected: control < endocrine < CP < duAdCa or FPC. This allowed for the determination of sensitivity (true-positive rate) and specificity (true-negative rate) as well as the determination of the AUC as a performance index. For the combinations of two or three markers, a logistic regression for binary outcome, affected or not, was performed as a first step to obtain a model with regression coefficients, which reflect the strength of influence for each marker on the probability that one of the two outcomes is present in an individual. The resulting logistic model was applied to determine the probability for each person to have one of the two outcomes. These individual probabilities based on the marker combination were then used to create an ROC curve, similar to the serum protein levels when looking only at a single marker.

^*TIMP1 plus LCN2.

^†CXCL16 plus TIMP1 plus LCN2.