Table 1.
Performance measures for selected serum models predicting cancer status
| ADC1 (training) LOOCV | ADC2 (test) external validation | |||||||
|---|---|---|---|---|---|---|---|---|
| Accuracy (%) | Sensitivity (%) | Specificity (%) | AUC | Accuracy (%) | Sensitivity (%) | Specificity (%) | AUC | |
| Single metabolite classifiers | ||||||||
| Aspartic Acid | 62.5 | 40.8 | 96.8 | 0.698 | 79.1 | 62.8 | 95.3 | 0.862 |
| Cystine | 70.0* | 75.5 | 61.3 | 0.685 | 55.8* | 76.7 | 34.9 | 0.677 |
| Glutamic Acid | 62.5 | 42.9 | 93.5 | 0.687 | 76.7 | 65.1 | 88.4 | 0.846 |
| Oxalic Acid | 70.0* | 83.7 | 48.4 | 0.65 | 57.0* | 88.4 | 25.6 | 0.649 |
| Multi-metabolite classifiers—clustered metabolites | ||||||||
| Cluster 1a SVMb | 76.3* | 77.6 | 74.2 | 0.751 | 84.9* | 72.1 | 97.7 | 0.856 |
Asterisk represents best model accuracies according to LOOCV. Best model accuracies according to external validation accuracy are underlined
aAspartic acid, cystine, glutamic acid
bSupport vector machines