Table 3 Performance of the mortality prediction models in Non-ST-segment elevation myocardial infarction using the traditional features.

From: Machine learning enhances the performance of short and long-term mortality prediction model in non-ST-segment elevation myocardial infarction

 

AUC (95% CI)

Specificity

Sensitivity

Accuracy

F1-score

In-hospital mortality

Machine learning algorithms

LR with Lasso

0.886 (0.855–0.918)

0.793

0.818

0.794

0.236

LR with Ridge

0.885 (0.852–0.918)

0.810

0.758

0.808

0.235

LR with Elastic net

0.886 (0.854–0.918)

0.791

0.803

0.792

0.230

RF

0.889 (0.856–0.923)

0.793

0.758

0.792

0.220

SVM

0.760 (0.707–0.813)

0.709

0.667

0.707

0.150

XGBoost

0.888 (0.857–0.919)

0.785

0.803

0.786

0.226

Traditional and modified traditional model

TIMI

0.669 (0.613–0.724)

0.686

0.576

0.682

0.123

GRACE

0.873 (0.840–0.906)

0.734

0.803

0.736

0.191

ACTION-GWTG

0.871 (0.836–0.907)

0.812

0.712

0.808

0.224

Modified TIMI*

0.709 (0.656–0.763)

0.506

0.788

0.516

0.112

Modified GRACE*

0.876 (0.841–0.912)

0.806

0.773

0.805

0.235

Modified ACTION-GWTG*

0.884 (0.851–0.916)

0.819

0.758

0.817

0.243

3 month mortality

Machine learning algorithms

LR with Lasso

0.849 (0.795–0.903)

0.728

0.833

0.731

0.127

LR with Ridge

0.826 (0.764–0.889)

0.719

0.833

0.722

0.124

LR with Elastic net

0.849 (0.795–0.904)

0.735

0.833

0.738

0.130

RF

0.799 (0.719–0.878)

0.681

0.778

0.683

0.104

SVM

0.715 (0.633–0.798)

0.557

0.778

0.562

0.077

XGBoost

0.824 (0.760–0.888)

0.654

0.861

0.659

0.106

Traditional and modified traditional model

TIMI

0.672 (0.592–0.751)

0.689

0.528

0.685

0.073

GRACE

0.777 (0.711–0.844)

0.705

0.694

0.704

0.100

ACTION-GWTG

0.795 (0.728–0.862)

0.726

0.750

0.727

0.114

Modified TIMI*

0.675 (0.596–0.754)

0.534

0.750

0.539

0.071

Modified GRACE*

0.774 (0.709–0.838)

0.623

0.778

0.627

0.089

Modified ACTION-GWTG*

0.782 (0.721–0.843)

0.759

0.639

0.756

0.110

12 month mortality

Machine learning algorithms

LR with Lasso

0.860 (0.825–0.895)

0.693

0.901

0.703

0.219

LR with Ridge

0.858 (0.821–0.894)

0.710

0.859

0.717

0.219

LR with Elastic net

0.859 (0.824–0.894)

0.721

0.845

0.727

0.222

RF

0.836 (0.796–0.876)

0.688

0.803

0.694

0.195

SVM

0.729 (0.675–0.784)

0.625

0.746

0.631

0.157

XGBoost

0.851 (0.817–0.884)

0.725

0.845

0.731

0.225

Traditional and modified traditional model

TIMI

0.675 (0.619–0.731)

0.695

0.549

0.688

0.140

GRACE

0.808 (0.764–0.852)

0.697

0.789

0.701

0.196

ACTION-GWTG

0.790 (0.740–0.839)

0.719

0.718

0.719

0.191

Modified TIMI*

0.729 (0.683–0.776)

0.545

0.845

0.559

0.150

Modified GRACE*

0.820 (0.779–0.861)

0.715

0.761

0.717

0.199

Modified ACTION-GWTG*

0.808 (0.768–0.848)

0.739

0.732

0.739

0.206

  1. AUC, area under the receiver operating characteristic curve; 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.
  2. *Traditional models were modified using the recalculated parameters for TIMI, GRACE, and ACTION-GWTG.
Back to article page