Use of Electrocardiograms to Identify Coronary Artery Disease

Figures

• 

  Central Illustration. Cross-Validation With External Validation The diagram represents the overall flow of the clinical validation with 2 separate databases. The first for training and developing the primary model to be used with cross-validation through 10 experiments, the second for measuring performance against never-before-seen data matched with a gold standard coronary angiogram. ROC = receiver-operating characteristic.

• 

  Figure 1. Decision Curve Analysis for ECGio Versus Alternative Treatment Pathways The decision curve representing the increase in overall benefit from use of ECGio when compared against treating all patients or treating no patients based on an acceptable risk tolerance.

• 

  Figure 2. The Receiver Operating-Characteristic Curves: Cross-Validation A 2x5 matrix represents the AUROC for each fold in the cross-validation with associated 95% CI compared with the mean AUROC for the cross-validation. The bottom panels are by sex; age; and number of risk factors, respectively. ROC = receiver-operating characteristic.

• 

  Figure 3. The Receiver-Operating Characteristic Curves: External Validation This set of panels showcase the performance in the external validation set. (A) Overall; (B) by sex; (C) by age; and (D) by number of risk factors. ROC = receiver-operating characteristic.

• 

  Figure 4. The Confusion Matrix Representing the Cross-Validation and External Validation Performance ECGio’s prediction of clinically significant coronary disease can be characterized by the results of the confusion matrix: a sensitivity, specificity, positive predictive value, negative predictive value, accuracy, F-1 score, and AUROC of 81.6%, 86.9%, 91.7%, 72.8%, 83.5%, 86.3%, and 91.4%, respectively. Within the external validation, the model had results of 78.2%, 90.7%, 82.5%, 88.1%, 86.2%, 80.3%, and 92.4%, respectively. The risk factors for consideration in panel D were: hypertension, hyperlipidemia, family history of disease, smoking, and diabetes mellitus.