Table 2.
Performance measures for selected plasma models predicting cancer status
| ADC 1 (training) LOOCV | ADC2 (test) external validation | |||||||
|---|---|---|---|---|---|---|---|---|
| Accuracy (%) | Sensitivity (%) | Specificity (%) | AUC | Accuracy (%) | Sensitivity (%) | Specificity (%) | AUC | |
| Single metabolite classifiers | ||||||||
| 3-phosphoglycerate | 70.7 | 60.8 | 87.1 | 0.734 | 51.2 | 34.9 | 67.4 | 0.578 |
| Maltose | 74.4* | 82.4 | 61.3 | 0.701 | 57.0* | 62.8 | 51.2 | 0.607 |
| Pyrophosphate | 69.5 | 66.7 | 74.2 | 0.703 | 76.7 | 67.4 | 86.0 | 0.811 |
| Multi-metabolite classifiers—clustered metabolites | ||||||||
| Cluster 1a SVMb | 80.5* | 86.3 | 71.0 | 0.713 | 70.9* | 72.1 | 69.8 | 0.675 |
Asterisk represents best model accuracies according to LOOCV accuracy. Best model accuracies according to external validation accuracy are underlined
a3-Phosphoglycerate, pyrophosphate
bSupport vector machines