Pediatric/Adult Congenital Heart Disease -> Adult Congenital Heart Disease D-AB31 - ACHD and Pediatric EP (ID 29) Abstract

D-AB31-01 - Stroke Prediction In Adults With Congenital Heart Disease (ID 1439)

Abstract

Background: Traditional risk stratification models, like the CHA2DS2-VASC score, have not been validated in the adult congenital heart disease (ACHD) population. Patients with ACHD have a high incidence of atrial arrhythmias secondary to the cardiac repairs. Empirical use of anticoagulation is not without risks, especially in those patients with pulmonary hypertension who are particularly vulnerable to fatal hemorrhagic complications.
Objective: We aimed to define the real-world risk of stroke and systemic embolism (SSE) in a large cohort of patients with ACHD; and subsequently derive the most reliable model for event prediction in patients with and without atrial arrhythmias.
Methods: Using de-identified administrative claims data from the OptumLabs Data Warehouse, which includes medical and pharmacy claims and enrollment records for commercial and Medicare Advantage enrollees, we identified 59,192 patients with ACHD treated between 2009 and 2014. Their outcomes (SSE) were adjudicated at 1, 2, and 5 years. The performance of the CHA2DS2-VASC score in predicting the real-world risk of SSE was compared with three popular machine learning (ML) models: Cox Proportional Hazard Model (CPH), Random Survival Forest (RSF), and Gradient Boosting Machines (GBM).
Results: Of the 59,192 patients with ACHD, 9,555(16%) had atrial fibrillation. During the 5-year follow-up, SSE occurred in 2,024 (3.4%) patients. All three ML models outperformed the CHA2DS2-VASC score in predicting SSE. Of these, GBM provided the best fitted model, AUC 0.83(0.82, 0.83), compared to 0.75(0.73, 0.79) for the CHA2DS2-VASC score. Adding 2 points to the CHA2DS2-VASC in patients with an atrial septal defect (ASD) improved its ability to predict SSE (AUC increased to 0.78 (0.77, 0.82)), however, the prediction was still not as accurate as that of the ML algorithm.
Conclusion: This newly defined ML algorithm is validated in a large cohort of patients with ACHD, and provides better prediction of SSE than the traditional CHA2DS2-VASC score. The presence of an ASD is a novel risk factor for SSE in the ACHD population. Whether this represents an increased risk of paradoxical embolism as a stroke mechanism in this population or a risk factor for enlarged atria and atrial fibrillation requires further study.
Collapse