A Simple and Easily Implemented Risk Model to Predict 1-Year Ischemic Stroke and Systemic Embolism in Chinese Patients with Atrial Fibrillation
Atrial fibrillation (AF) is a common cardiac arrhythmia associated with an increased risk of thromboembolic events (TEs), particularly ischemic stroke and systemic embolism. The management of AF patients often involves the use of oral anticoagulants (OACs) to reduce this risk. However, the decision to initiate anticoagulation therapy must balance the benefits of stroke prevention against the potential for bleeding complications. Current guidelines recommend the use of risk stratification tools, such as the CHA2DS2-VASc score, to guide anticoagulation decisions. Despite its widespread use, the CHA2DS2-VASc score has limitations, particularly in identifying patients at truly low risk of stroke, as it classifies over 90% of AF patients as candidates for anticoagulation. This study aimed to develop a more accurate and simpler risk model for predicting 1-year TEs in Chinese AF patients, with the goal of improving anticoagulation decision-making.
The study utilized data from the China Atrial Fibrillation (China-AF) Registry, a prospective, multicenter cohort study conducted in Beijing, China. The registry enrolled 23,108 AF patients from August 2011 to December 2017. For this analysis, patients who were already on OAC therapy or had undergone catheter ablation at baseline were excluded, leaving a cohort of 6,601 anticoagulation-naive AF patients. The primary outcome was the occurrence of a TE, defined as ischemic stroke or systemic embolism, within one year of enrollment. Patient-reported TEs were adjudicated by independent neurologists to ensure accuracy.
To develop the new risk model, the study employed the extreme gradient boosting (XGBoost) algorithm, a state-of-the-art machine learning technique, to identify the most important predictors of TEs from a pool of 44 candidate variables. The XGBoost algorithm was chosen for its ability to handle missing data and rank variables based on their contribution to the prediction model. Through this process, three key variables were identified as the most significant predictors of 1-year TEs: congestive heart failure or left ventricular dysfunction (LVEF 65 years, and prior stroke. These three variables accounted for 73.1% of the prognostic information provided by all clinical variables.
Using these three variables, the study developed a simplified risk model, termed the CAS model (Congestive heart failure or left ventricular dysfunction, Age, and prior Stroke). The CAS model was constructed using a multivariate Cox regression analysis, with point scores assigned to each variable based on their regression coefficients. Specifically, patients received 1 point for having congestive heart failure or LVEF <55%, 1 point for being older than 65 years, and 2 points for a history of prior stroke. The total CAS score was calculated by summing these points, with higher scores indicating a greater risk of TEs.
The CAS model was internally validated using bootstrapping with 1,000 replicates to assess its predictive accuracy. The model demonstrated good discrimination, with a C-statistic of 0.69 (95% confidence interval [CI]: 0.65–0.73), which was significantly higher than the C-statistic of the CHA2DS2-VA score (0.66, 95% CI: 0.62–0.70). The CAS model classified 30.8% of the patients (2,033 out of 6,601) as low risk (CAS score = 0), with a corresponding 1-year TE risk of 0.81% (95% CI: 0.41%–1.19%). In contrast, the CHA2DS2-VA score classified only 15.2% of the patients (1,002 out of 6,601) as low risk (CHA2DS2-VA score = 0), with a 1-year TE risk of 1.01% (95% CI: 0.36%–1.64%).
The CAS model’s ability to reclassify patients into more accurate risk categories was assessed using the net reclassification improvement (NRI) metric. The overall NRI from the CHA2DS2-VA score categories (low = 0, high ≥1) to the CAS categories (low = 0, high ≥1) was 12.2% (95% CI: 8.7%–15.7%). This improvement indicates that the CAS model more effectively identified patients at low risk of TEs, potentially reducing unnecessary anticoagulation therapy.
The study also compared the CAS model’s performance with the CHA2DS2-VA score in terms of identifying high-risk patients. When classifying a specific proportion of cases as high-risk, the CAS model consistently identified a higher proportion of patients who would actually experience TEs compared to the CHA2DS2-VA score. Additionally, to prevent a specific proportion of patients from experiencing TEs by treating them with OAC therapy, the CAS model classified a lower proportion of patients than the CHA2DS2-VA score, suggesting a more efficient use of anticoagulation therapy.
The findings of this study highlight the importance of using more precise risk stratification tools in clinical practice. The CAS model’s simplicity, with only three variables, makes it easy to implement in routine care. Its ability to identify a larger proportion of patients at truly low risk of TEs could help clinicians make more informed decisions about anticoagulation therapy, potentially reducing the risk of bleeding complications in patients who may not benefit from OACs.
The study’s limitations include its focus on a Chinese AF population, which may limit the generalizability of the findings to other ethnic groups. External validation of the CAS model in diverse populations is needed to confirm its applicability. Additionally, the study did not incorporate biomarkers, left atrial morphology and function, or AF burden into the risk prediction model, which may provide additional prognostic information. Future research could explore the integration of these factors to further refine the model.
In conclusion, the CAS risk model represents a significant advancement in the risk stratification of AF patients. By identifying a larger proportion of patients at low risk of TEs, the CAS model has the potential to improve anticoagulation decision-making and reduce unnecessary treatment. Its simplicity and accuracy make it a valuable tool for clinicians managing AF patients, particularly in the Chinese population. Further research and external validation are needed to confirm the model’s utility in broader clinical settings.
doi.org/10.1097/CM9.0000000000001515
Was this helpful?
0 / 0