Table 4.

Validation of the prediction model of recurrence in TCGA

	Model with clinical/copy number variation (CNV)		Model with clinical/miRNA		Model with clinical/miRNA/CNV
Recurrence probability scale*	Cutoff=0.501		Cutoff=0.500		Cutoff=0.553
	Value	95% CI	Value	95% CI	Value	95% CI
Sensitivity	55%	0.48, 0.61	33%	0.18, 0.47	32%	0.13, 0.52
Specificity	81%	0.65, 0.94	81%	0.77, 0.85	80%	0.75, 0.85
Positive predictive value	16%	0.07, 0.25	19%	0.12, 0.25	18%	0.08, 0.26
Negative predictive value	89%	0.87, 0.91	89%	0.88, 0.91	90%	0.87, 0.92
Accuracy	74%	0.72, 0.77	75%	0.73, 0.77	74%	0.72, 0.77

Note: *Recurrence probability scale: 1/(exp(-score)+1), where score is the resulting value of the prediction model in logit scale. As detailed in methods, the threshold was selected for specificity around 80% and highest sensitivity and negative predictive value. The goal was to create models that would rule out at least 80 of non-recurrent patients but still capturing most of patients with recurrence.