Table 2. Performance of scoring systems and ML approaches for the estimation of in-hospital mortality in patients with an out-of-hospital cardiac arrest.
Model | Predicted mortality | AUC (95% CI) | Brier score | Log loss |
---|---|---|---|---|
Actual mortality | 45.5% | |||
APACHE III risk of death | 52.8% | 0.80 (0.79–0.82) | 0.190 | 0.57 |
ANZROD | 39.9% | 0.81 (0.80–0.82) | 0.182 | 0.55 |
Logistic regression | 45.4% | 0.82 (0.81–0.83) | 0.170 | 0.51 |
Artificial neural network | 46.7% | 0.85 (0.84–0.86) | 0.158 | 0.48 |
Random forest | 45.7% | 0.86 (0.84–0.87) | 0.156 | 0.47 |
Support vector classifier | 45.4% | 0.86 (0.85–0.87) | 0.153 | 0.47 |
Ensemble | 45.5% | 0.87 (0.86–0.88) | 0.148 | 0.45 |
Gradient boosted machine | 45.3% | 0.87 (0.86–0.88) | 0.147 | 0.45 |
Results presented are based on test set (n = 3,957).
ANZROD, Australian and New Zealand Risk of Death; AUC, area under the curve; ML, machine learning.