Cardiology Research, ISSN 1923-2829 print, 1923-2837 online, Open Access
Article copyright, the authors; Journal compilation copyright, Cardiol Res and Elmer Press Inc
Journal website https://cr.elmerpub.com

Original Article

Volume 16, Number 5, October 2025, pages 385-393

Discriminative Accuracy of CHA2DS2-VASc Score, and Development of Predictive Accuracy Model Using Machine Learning for Ischemic Stroke Risk in Cardiac Amyloidosis and Atrial Fibrillation

Figures

Figure 1.
Figure 1. Proportion of major events and estimates of outcomes between CA-AF and no-CA-AF at index admission and 30-day readmission. CA: cardiac amyloidosis; AF: atrial fibrillation.
Figure 2.
Figure 2. Parsimony plot of the new model’s performance predicting stroke (as indicated by the area under the curve (AUC)) as a function of the model’s complexity (as noted in the number of variables included in the model). The AUC (accuracy of the model) increased with the addition of variables till HTN; no further increase in accuracy could be achieved with the addition of other variables beyond variable 8 (AKI). This indicates that the model’s highest accuracy was attained with the first seven variables. HTN: hypertension; ESRD: end-stage renal disease; DM: diabetes mellitus; AKI: acute kidney injury; PUD: peptic ulcer disease; COPD: chronic obstructive pulmonary disease; PVD: peripheral vascular disease.
Figure 3.
Figure 3. ROC curve for E-CHADS model showing a high AUC at 80% for a cut-off RF score of 52 for identifying 30-day risk of ischemic stroke in patients with CA-AF. AUC: area under the curve; CA: cardiac amyloidosis; AF: atrial fibrillation.
Figure 4.
Figure 4. The CHA2DS2-VASc model (on the left) was a poor predictor of stroke in AF patients with comorbid cardiac amyloidosis, with an AUC = 0.50. The new model E-CHADS score (on the right) appreciably improves in predictive accuracy for stroke in AF patients with cardiac amyloidosis, with an AUC = 0.80. Variables in each model’s scoring system are listed with their individual random forest scores; the high cumulative score represents a higher risk of ischemic stroke in CA-AF at 30-day readmission. CHF: congestive heart failure; HTN: hypertension; DM: diabetes mellitus; Vasc: vascular disease; AUC: area under the curve. CA: cardiac amyloidosis; AF: atrial fibrillation.

Table

Table 1. Model Performance at Specified Score Cutoffs and Predicted Risk of Stroke
 
Score cut-offPredicted risk% of patientsAccuracy (95% CI)Sensitivity (95% CI)Specificity (95% CI)PPV (95% CI)NPV (95% CI)
Machine learning model’s predicted risk of stroke, based on scores, was determined by eight selected variables (heart failure, ESRD, HTN, neurological disease, dementia, electrolyte abnormalities, DM, and any cancer). Also shown are the accuracy, sensitivity, specificity, PPV, NPV, and their respective 95% CIs for the model at each score cut-off point. Values for accuracy, sensitivity, specificity, PPV, NPV, and their respective confidence intervals are listed as percentages. ESRD: end-stage renal disease; HTN: hypertension; DM: diabetes mellitus; PPV: positive predictive value; NPV: negative predictive value; CI: confidence interval.
≥ 20≥ 0.80%959.8 (7.9 - 11.7)100 (100 - 100)5.5 (3.6 - 7.5)4.8 (4.7 - 4.9)100 (100 - 100)
≥ 40≥ 2.40%5548.9 (44.7 - 53.2)95.8 (87.5 - 100)46.6 (42.3 - 51.2)7.9 (7 - 8.7)99.6 (98.7 - 100)
≥ 60≥ 5.90%2279.2 (75.8 - 82.6)66.7 (49.9 - 83.3)79.8 (76.3 - 83.2)13.6 (9.7 - 17.5)98.1 (97 - 99)
≥ 75≥ 11.60%988.5 (86 - 90.8)20.8 (4.2 - 37.5)91.7 (89.3 - 94.1)10.4 (2.9 - 18.9)96 (95.3 - 96.9)
≥ 90≥ 21.60%294.3 (93.2 - 95.3)4.2 (0 - 12.5)98.6 (97.6 - 99.6)11.1 (0 - 42.9)95.6 (95.4 - 96)
≥ 100≥ 29%095.5 (94.9 - 96)4.2 (0 - 12.5)99.8 (99.4 - 100)50 (0 - 100)95.6 (95.4 - 96)