Table 2.
Performance in the mortality prediction models in ST-segment elevation myocardial infarction using the traditional features.
| AUC (95% CI) | Specificity | Sensitivity | Accuracy | F1-score | |
|---|---|---|---|---|---|
| In-hospital mortality | |||||
| Machine learning algorithms | |||||
| LR with Lasso | 0.890 (0.852–0.928) | 0.881 | 0.726 | 0.873 | 0.388 |
| LR with Ridge | 0.889 (0.850–0.927) | 0.766 | 0.871 | 0.772 | 0.298 |
| LR with Elastic net | 0.890 (0.852–0.928) | 0.888 | 0.677 | 0.876 | 0.378 |
| RF | 0.910 (0.879–0.941) | 0.817 | 0.823 | 0.817 | 0.333 |
| SVM | 0.819 (0.765–0.873) | 0.804 | 0.677 | 0.797 | 0.271 |
| XGBoost | 0.912 (0.884–0.939) | 0.845 | 0.839 | 0.845 | 0.376 |
| Traditional and modified traditional model | |||||
| TIMI | 0.855 (0.813–0.897) | 0.769 | 0.774 | 0.769 | 0.272 |
| GRACE | 0.896 (0.862–0.930) | 0.842 | 0.774 | 0.838 | 0.347 |
| ACTION-GWTG | 0.891 (0.855–0.927) | 0.837 | 0.758 | 0.832 | 0.335 |
| Modified TIMI* | 0.885 (0.849–0.920) | 0.826 | 0.806 | 0.825 | 0.339 |
| Modified GRACE* | 0.901 (0.870–0.932) | 0.826 | 0.823 | 0.826 | 0.345 |
| Modified ACTION-GWTG* | 0.859 (0.810–0.907) | 0.833 | 0.710 | 0.826 | 0.312 |
| 3-month mortality | |||||
| Machine learning algorithms | |||||
| LR with Lasso | 0.777 (0.682–0.871) | 0.673 | 0.857 | 0.677 | 0.101 |
| LR with Ridge | 0.779 (0.683–0.875) | 0.620 | 0.857 | 0.625 | 0.088 |
| LR with Elastic net | 0.777 (0.683–0.872) | 0.652 | 0.857 | 0.657 | 0.095 |
| RF | 0.763 (0.656–0.870) | 0.801 | 0.571 | 0.797 | 0.107 |
| SVM | 0.667 (0.525–0.810) | 0.852 | 0.381 | 0.842 | 0.092 |
| XGBoost | 0.784 (0.688–0.880) | 0.726 | 0.762 | 0.727 | 0.106 |
| Traditional and modified traditional model | |||||
| TIMI | 0.743 (0.650–0.837) | 0.610 | 0.810 | 0.614 | 0.082 |
| GRACE | 0.766 (0.670–0.862) | 0.652 | 0.857 | 0.657 | 0.096 |
| ACTION-GWTG | 0.709 (0.602–0.816) | 0.630 | 0.667 | 0.630 | 0.070 |
| Modified TIMI* | 0.704 (0.593–0.815) | 0.628 | 0.714 | 0.629 | 0.075 |
| Modified GRACE* | 0.602 (0.458–0.745) | 0.832 | 0.238 | 0.820 | 0.053 |
| Modified ACTION-GWTG* | 0.653 (0.528–0.778) | 0.731 | 0.476 | 0.726 | 0.068 |
| 12-month mortality | |||||
| Machine learning algorithms | |||||
| LR with Lasso | 0.835 (0.776–0.895) | 0.799 | 0.688 | 0.796 | 0.179 |
| LR with Ridge | 0.840 (0.784–0.896) | 0.720 | 0.844 | 0.724 | 0.165 |
| LR with Elastic net | 0.835 (0.781–0.889) | 0.776 | 0.719 | 0.775 | 0.171 |
| RF | 0.825 (0.749–0.901) | 0.697 | 0.875 | 0.703 | 0.160 |
| SVM | 0.684 (0.574–0.795) | 0.592 | 0.719 | 0.597 | 0.103 |
| XGBoost | 0.806 (0.743–0.869) | 0.782 | 0.656 | 0.778 | 0.160 |
| Traditional and modified traditional model | |||||
| TIMI | 0.793 (0.726–0.860) | 0.642 | 0.844 | 0.648 | 0.134 |
| GRACE | 0.826 (0.770–0.881) | 0.677 | 0.812 | 0.681 | 0.142 |
| ACTION-GWTG | 0.780 (0.709–0.850) | 0.770 | 0.562 | 0.763 | 0.134 |
| Modified TIMI* | 0.802 (0.736–0.868) | 0.786 | 0.688 | 0.783 | 0.170 |
| Modified GRACE* | 0.741 (0.663–0.820) | 0.771 | 0.625 | 0.766 | 0.148 |
| Modified ACTION-GWTG* | 0.659 (0.554–0.764) | 0.748 | 0.531 | 0.741 | 0.117 |
AUC, area under the receiver operating characteristic curve; CI, confidential interval; LR, Logistic regression; Lasso, L1 penalty; Ridge, L2 penalty; Elastic net, Elastic net penalty; RF, Random Forest; SVM, Support Vector Machine; XGBoost, Extreme Gradient Boosting; Thrombolysis in myocardial infarction, TIMI; Global registry of acute coronary events, GRACE; Acute coronary treatment and intervention outcomes network—Get With The Guidelines, ACTION-GWTG.
*Traditional models were modified using the recalculated parameters for TIMI, GRACE, and ACTION-GWTG.