Table 2.
Model performance metrics.
| Model | Sensitivity, % (95% CI) |
Specificity, % (95% CI) |
Accuracy, % (95% CI) |
PPV, % (95% CI) |
F1 Score | ROC (95% CI) |
AUC PR (95% CI) |
p-Value * |
|---|---|---|---|---|---|---|---|---|
| Random Forest (MEWS++) | 78.9 (77.6–80.1) |
79.1 (78.9–79.3) |
79.1 (78.9–79.3) |
11.5 (11.1–11.9) |
0.2 | 87.9 (87.4–88.4) |
36.2 (34.7–37.7) |
<0.0001 |
| Linear SVM | 79.0 (77.6–80.3) |
77.9 (77.6–78.1) |
77.9 (77.7–78.2) |
11.0 (10.6–11.4) |
0.19 | 87.3 (86.8–87.9) |
28.7 (27.2–30.2) |
<0.00010.16 ** |
| LR | 61.4 (59.8–63.0) |
78.5 (78.3–78.8) |
77.9 (77.7–78.2) |
9.0 (8.6–9.4) |
0.16 | 79.1 (78.4–79.8) |
17.2 (16.0–18.5) |
<0.0001 |
| MEWS Score | 64.2 (62.7–65.7) |
66.2 (66.0–66.5) |
66.2 (65.9–66.4) |
6.1 (5.9–6.4) |
0.11 | 66.7 (65.9–67.6) |
7.0 (6.2–7.8) |
* p-value for difference between AUC ROC for respective ML model and MEWS Score. ** p-value = 0.16 for Random Forest vs. Linear SVM. AUCPR—Area Under Precision Recall Curve, LR—Linear Regression, SVM—Support Vector Machine, ROC—Receiver Operating Characteristic.