Table 2.
Predictive power of the ML model and routine clinical models using data of the training cohort.
| Model | Specificity | Sensitivity | Accuracy | NPV | PPV | Recall | Youden’s index |
|---|---|---|---|---|---|---|---|
| Machine learning model | 85.9% | 87.6% | 86.3% | 96.0% | 64.2% | 87.6% | 1.735 |
| ALBI | 66.1% | 49.6% | 62.4% | 82.0% | 29.7% | 49.6% | 1.157 |
| FIB-4 | 33.7% | 81.8% | 44.4% | 86.5% | 26.2% | 81.8% | 1.154 |
| APRI | 55.8% | 59.1% | 56.5% | 82.6% | 27.8% | 59.1% | 1.149 |
| MELD | 76.2% | 31.4% | 66.2% | 79.4% | 27.6% | 31.4% | 1.076 |
| CTP | 95.8% | 6.6% | 75.8% | 78.0% | 31.0% | 6.6% | 1.024 |
ML, machine learning; NPV, negative predictive value; PPV, positive predictive value; ALBI, albumin-bilirubin grade; FIB-4, fibrosis 4 score; APRI, aspartate transaminase to platelet ratio; MELD, model for end-stage liver disease; CTP, Child-Turcotte-Pugh.