. 2019 Jul 26;14(7):e0219531. doi: 10.1371/journal.pone.0219531

Table 4. Performance of metabolomics as a diagnostic tool in Gulf War illness.

No. of Analytes	Classifier	2 x 2 Contingency Table Analysis				AUROC Performance¹		Validation
No. of Analytes	Classifier	False Negatives/ True Positive	False Positives/ True Negatives	Sensitivity (95% CI)	Specificity (95% CI)	Accuracy (area under the curve)	95% CI	rdCV¹	p²
1	12-HETE	8/12	8/12	0.60 (0.39–0.78)	0.60 (0.39–0.78)	0.69	0.43–0.90	0.624	ns
2	12-HETE, Taurine	6/14	7/13	0.70 (0.48–0.85)	0.65 (0.43–0.82)	0.72	0.47–0.93	0.658	0.088
3	12-HETE, Taurine, Ceramide(d18:1/20:0)	5/15	3/17	0.75 (0.53–0.89)	0.85 (0.64–0.95)	0.84	0.65–0.98	0.764	0.02
4	12-HETE, Taurine, Ceramide(d18:1/20:0), SM(d18:1/26:1 OH)	3/17	3/17	0.85 (0.64–0.95)	0.85 (0.64–0.95)	0.92	0.72–1.0	0.832	0.002
5	12-HETE, Taurine, Ceramide(d18:1/20:0), SM(d18:1/26:1 OH), Xanthosine	2/18	3/17	0.90 (0.7–0.98)	0.85 (0.64–0.95)	0.93	0.76–1.0	0.824	0.003
6	12-HETE, Taurine, Ceramide(d18:1/20:0), SM(d18:1/26:1 OH) Xanthosine, Plasmalogen(20:4/p18:1)	1/19	2/18	0.95 (0.76–1.0)	0.90 (0.7–0.98)	0.94	0.79–1.0	0.836	0.004

¹ Area under the receiver operator characteristic (AUROC) curve and repeated double cross validation (rdCV) results were calculated by random forest analysis and bootstrap resampling x 100.

² Empirical p value after 1000 permutations. The direction and magnitude change of the metabolites used in the classifiers are discussed in the text, illustrated in Fig 3, and listed in S1 Table.