Table 3. Performance of the XGBoost and Meta-Classifier Models Compared With Logistic Regressiona.
| Model | Expanded LR, No. of patients (% observed mortality) | |||
|---|---|---|---|---|
| <1% | 1%-5% | >5% | All | |
| XGBoost vs LR | ||||
| Expanded XGBoost | ||||
| <1% | 65 193 (0.27) | 31 971 (0.65) | 422 (1.18) | 97 586 (0.40) |
| 1%-5% | 3384 (0.95) | 44 486 (2.21) | 13 155 (3.91) | 61 025 (2.51) |
| >5% | 68 (2.94) | 2899 (6.21) | 28 906 (20.79) | 31 873 (19.42) |
| All | 68 645 (0.30) | 79 356 (1.73) | 42 483 (15.37) | 190 484 (4.26) |
| Meta-classifier vs LR | ||||
| Expanded meta-classifier | ||||
| <1% | 65 694 (0.27) | 30 661 (0.65) | 175 (0.00) | 96 530 (0.39) |
| 1%-5% | 2930 (1.06) | 45 726 (2.17) | 9033 (3.55) | 57 689 (2.33) |
| >5% | 21 (0.00) | 2969 (6.03) | 33 275 (18.66) | 36 265 (17.61) |
| All | 68 645 (0.30) | 79 356 (1.73) | 42 483 (15.37) | 190 484 (4.26) |
Abbreviation: LR, logistic regression.
a
Pairwise comparisons of the same patients classified into low (<1%), medium (1-5%), and high (>5%) risk of death based on logistic regression and XGBoost (top) and meta-classifier (bottom).