Table 2.
Fusion model performance benchmarked against clinical scores
| Threshold source | Model | Threshold | TP | TN | FP | FN | Sensitivity | Specificity | PPV | NPV | McNemar test |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Cross-validation fold | Fusion model | 0.037 | 22 | 14 | 62 | 0 | 1.00 | 0.18 | 0.26 | 1.00 | (Base) |
| Fusion model (no D-dimer) | 0.040 | 22 | 14 | 62 | 0 | 1.00 | 0.18 | 0.26 | 1.00 | 1.000 | |
| Wells et al.26 | Wells’ criteria | 1 | 14 | 24 | 52 | 8 | 0.64 | 0.32 | 0.21 | 0.75 | 0.005 |
| Le Gal et al.27 | Revised Geneva Score | 4 | 15 | 20 | 56 | 7 | 0.68 | 0.26 | 0.21 | 0.74 | 0.041 |
| Kline et al.28 | PERC | 1 | 22 | 2 | 74 | 0 | 1.00 | 0.03 | 0.23 | 1.00 | 0.185 |
| Roy et al.29 | 4PEPS | 0 | 19 | 19 | 57 | 3 | 0.86 | 0.25 | 0.25 | 0.86 | 0.002 |
| Holdout test set | Fusion model | 0.091 | 22 | 43 | 33 | 0 | 1.00 | 0.57 | 0.40 | 1.00 | (Base) |
| Fusion model (no D-dimer) | 0.072 | 22 | 21 | 55 | 0 | 1.00 | 0.28 | 0.29 | 1.00 | <0.001 | |
| Wells’ criteria | 0 | 22 | 0 | 76 | 0 | 1.00 | 0.00 | 0.22 | 0.00 | <0.001 | |
| Revised Geneva Score | 0 | 22 | 0 | 76 | 0 | 1.00 | 0.00 | 0.22 | 0.00 | <0.001 | |
| PERC | 1 | 22 | 2 | 74 | 0 | 1.00 | 0.03 | 0.23 | 1.00 | <0.001 | |
| 4PEPS | −1 | 22 | 4 | 72 | 0 | 1.00 | 0.05 | 0.23 | 1.00 | <0.001 |
Encounters are classified as PE-positive if the clinical score or model likelihood is greater than or equal to the threshold value.
4PEPS, 4-Level Pulmonary Embolism Clinical Probability Score; FN, false negatives; FP, false positives; NPV, negative predictive value; PERC, Pulmonary Embolism Rule-Out Criteria; PPV, positive predictive value; TN, true negatives; TP, true positives.