Table 2.

Predictive capacity and consistency of individual prediction models for the depletion of GSH with 4 different incubation conditions

Solvent	Test chemicals	Cut-off (%)	Sensitivity (%)	Specificity (%)	Accuracy (%)	Consistency (%)
Acetonitrile	15-fold	≥ 8.5	76.2 (16/21)	87.0 (20/23)	81.8 (36/44)	90.9 (40/44)
	10-fold	≥ 6.0	90.5 (19/21)	91.3 (21/23)	90.9 (40/44)	72.7 (32/44)
DMSO	15-fold	≥ 8.7	90.5 (19/21)	82.6 (19/23)	86.4 (38/44)	90.9 (40/44)
	10-fold	≥ 5.8	95.2 (20/21)	91.3 (21/23)	93.2 (41/44)	93.2 (41/44)

The cut-off values were determined by analyzing corresponding receiver-operating characteristic (ROC) curves. To determine the cut-off value for these prediction models, total three results of each trial were subjected to ROC analysis. Optimal cut-off values were obtained by plotting true positive values (sensitivity) in y-axis against false positive values (1-specificity) in x-axis. The value that provided the highest Youden index and the point with the lowest distance to the upper-left corner of the ROC curve was chosen as the best cut-off value