A simple and easily implemented risk model to predict 1-year ischemic stroke and systemic embolism in Chinese patients with atrial fibrillation

• Published in: Chinese medical journal Vol. 134; no. 19; pp. 2293 - 2298
• Main Authors: Jiang, Chao, Chen, Tian-Ge, Du, Xin, Li, Xiang, He, Liu, Lai, Yi-Wei, Xia, Shi-Jun, Liu, Rong, Hu, Yi-Ying, Li, Ying-Xue, Jiang, Chen-Xi, Liu, Nian, Tang, Ri-Bo, Bai, Rong, Sang, Cai-Hua, Long, De-Yong, Xie, Guo-Tong, Dong, Jian-Zeng, Ma, Chang-Sheng
• Format: Journal Article
• Language: English
• Published: China Lippincott Williams & Wilkins 05.10.2021; Lippincott Williams & Wilkins Ovid Technologies; Wolters Kluwer
• Subjects: Ablation; Age; Anticoagulants; Atrial Fibrillation - drug therapy; Body mass index; Brain Ischemia; Cardiac arrhythmia; Catheters; China; Cohort Studies; Confidence intervals; Demographics; Diabetes; Ejection fraction; Embolism; Embolisms; Female; Heart failure; Humans; Hypertension; Ischemic Stroke; Machine learning; Medical prognosis; Original; Patients; Prospective Studies; Risk Assessment; Risk Factors; Standard deviation; Stroke; Stroke - etiology; Thromboembolism; Transient ischemic attack; Variables; China; Beijing China
• ISSN: 0366-6999; 2542-5641; 2542-5641
• DOI: 10.1097/CM9.0000000000001515

Accurate prediction of ischemic stroke is required for deciding anticoagulation use in patients with atrial fibrillation (AF). Even though only 6% to 8% of AF patients die from stroke, about 90% are indicated for anticoagulants according to the current AF management guidelines. Therefore, we aimed to develop an accurate and easy-to-use new risk model for 1-year thromboembolic events (TEs) in Chinese AF patients. From the prospective China Atrial Fibrillation Registry cohort study, we identified 6601 AF patients who were not treated with anticoagulation or ablation at baseline. We selected the most important variables by the extreme gradient boosting (XGBoost) algorithm and developed a simplified risk model for predicting 1-year TEs. The novel risk score was internally validated using bootstrapping with 1000 replicates and compared with the CHA2DS2-VA score (excluding female sex from the CHA2DS2-VASc score). Up to the follow-up of 1 year, 163 TEs (ischemic stroke or systemic embolism) occurred. Using the XGBoost algorithm, we selected the three most important variables (congestive heart failure or left ventricular dysfunction, age, and prior stroke, abbreviated as CAS model) to predict 1-year TE risk. We trained a multivariate Cox regression model and assigned point scores proportional to model coefficients. The CAS scheme classified 30.8% (2033/6601) of the patients as low risk for TE (CAS score = 0), with a corresponding 1-year TE risk of 0.81% (95% confidence interval [CI]: 0.41%-1.19%). In our cohort, the C-statistic of CAS model was 0.69 (95% CI: 0.65-0.73), higher than that of CHA2DS2-VA score (0.66, 95% CI: 0.62-0.70, Z = 2.01, P = 0.045). The overall net reclassification improvement from CHA2DS2-VA categories (low = 0/high ≥1) to CAS categories (low = 0/high ≥1) was 12.2% (95% CI: 8.7%-15.7%). In Chinese AF patients, a novel and simple CAS risk model better predicted 1-year TEs than the widely-used CHA2DS2-VA risk score and identified a large proportion of patients with low risk of TEs, which could potentially improve anticoagulation decision-making. www.chictr.org.cn (Unique identifier No. ChiCTR-OCH-13003729).