Scholarlink

Background <p>The left atrial appendage (LAA) is the primary source of thromboembolic events in patients with non-valvular atrial fibrillation (NVAF). Therefore, detection of LAA thrombosis and severe spontaneous echo contrast (SEC) is crucial for risk stratification and management of patients with NVAF.</p> Methods <p>We retrospectively enrolled 327 patients with non-paroxysmal NVAF and collected 34 clinical and echocardiographic variables. Three machine learning approaches (SVM-RFE, Boruta, and LASSO) were employed for feature selection to construct logistic regression models. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, decision curve analysis (DCA), and clinical impact curves (CIC). Reclassification analysis was performed using net reclassification improvement (NRI) and integrated discrimination improvement (IDI).</p> Results <p>LAA thrombosis or severe SEC was detected in 40 (12.2%) patients. The three machine learning algorithms demonstrated comparable area under the curve (AUC) performance, with values of 0.88 for SVM-RFE, 0.89 for LASSO, and 0.89 for Boruta. The final nomogram was constructed using variables selected by all three machine learning models. It achieved an AUC of 0.88 (95% CI: 0.83–0.93), which was significantly higher than that of the CHA<sub>2</sub>DS<sub>2</sub>-VASc score (AUC = 0.68, <i>p</i> &lt; 0.001). DCA and CIC analyses showed high net benefits across a wide range of threshold probabilities. Reclassification analysis revealed an NRI of 0.957 and IDI of 0.254 (both <i>p</i> &lt; 0.001) when compared with the CHA<sub>2</sub>DS<sub>2</sub>-VASc score.</p> Conclusions <p>This exploratory study suggests that the proposed nomogram may help estimate the risk of LAA thrombosis/severe SEC in patients with non-paroxysmal NVAF. The model may be useful for risk stratification and may inform clinical decision-making regarding the necessity of TEE in patients with non-paroxysmal NVAF, but its clinical reliability requires further validation in larger, independent, prospective cohorts.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development of a nomogram with machine learning-assisted feature selection for predicting left atrial appendage thrombosis and severe spontaneous echo contrast in patients with non-paroxysmal atrial fibrillation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月10日

查看全文
  • Heng Wang,
  • Jiali Fan,
  • Bingyuan Zhou,
  • Changsheng Ma

摘要

Background

The left atrial appendage (LAA) is the primary source of thromboembolic events in patients with non-valvular atrial fibrillation (NVAF). Therefore, detection of LAA thrombosis and severe spontaneous echo contrast (SEC) is crucial for risk stratification and management of patients with NVAF.

Methods

We retrospectively enrolled 327 patients with non-paroxysmal NVAF and collected 34 clinical and echocardiographic variables. Three machine learning approaches (SVM-RFE, Boruta, and LASSO) were employed for feature selection to construct logistic regression models. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, decision curve analysis (DCA), and clinical impact curves (CIC). Reclassification analysis was performed using net reclassification improvement (NRI) and integrated discrimination improvement (IDI).

Results

LAA thrombosis or severe SEC was detected in 40 (12.2%) patients. The three machine learning algorithms demonstrated comparable area under the curve (AUC) performance, with values of 0.88 for SVM-RFE, 0.89 for LASSO, and 0.89 for Boruta. The final nomogram was constructed using variables selected by all three machine learning models. It achieved an AUC of 0.88 (95% CI: 0.83–0.93), which was significantly higher than that of the CHA2DS2-VASc score (AUC = 0.68, p < 0.001). DCA and CIC analyses showed high net benefits across a wide range of threshold probabilities. Reclassification analysis revealed an NRI of 0.957 and IDI of 0.254 (both p < 0.001) when compared with the CHA2DS2-VASc score.

Conclusions

This exploratory study suggests that the proposed nomogram may help estimate the risk of LAA thrombosis/severe SEC in patients with non-paroxysmal NVAF. The model may be useful for risk stratification and may inform clinical decision-making regarding the necessity of TEE in patients with non-paroxysmal NVAF, but its clinical reliability requires further validation in larger, independent, prospective cohorts.