TABLE 4.
Prediction performance of machine learning approaches for predicting early death among bone metastatic breast cancer patients.
| Measure | Approach | |||
|---|---|---|---|---|
| Logistic regression | Gradient boosting tree | Decision tree | Random forest | |
| Mean predicted | 0.176 | 0.176 | 0.176 | 0.175 |
| Brier score | 0.112 | 0.109 | 0.117 | 0.111 |
| Intercept | 0.06 | 0.06 | 0.05 | 0.07 |
| Calibration slope | 1.01 | 1.06 | 0.96 | 1.20 |
| AUC (95% CI) | 0.819 (0.791–0.847) | 0.829 (0.802–0.856) | 0.797 (0.767–0.826) | 0.828 (0.801–0.855) |
| Discrimination slope | 0.240 | 0.258 | 0.216 | 0.223 |
| Specificity | 0.766 | 0.823 | 0.764 | 0.775 |
| Sensitivity (recall) | 0.745 | 0.704 | 0.707 | 0.752 |
| NPV | 0.931 | 0.926 | 0.921 | 0.933 |
| PPV (precision) | 0.416 | 0.472 | 0.402 | 0.427 |
| Youden | 1.511 | 1.527 | 1.471 | 1.526 |
| Accuracy | 0.762 | 0.801 | 0.754 | 0.770 |
| Threshold | 0.191 | 0.203 | 0.191 | 0.193 |
AUC, area under the curve; CI, confident interval; NPV, negative predictive value; PPV, positive predictive value.