Scholarlink

Background <p>Atrial fibrillation (AF) is the most common arrhythmia worldwide, with catheter ablation being an effective yet recurrence-prone treatment. Given the limited accuracy of conventional risk scores in identifying patients at high risk of recurrence after catheter ablation, this study sought to develop and validate a machine learning (ML) model for predicting AF recurrence using a wide array of clinical and laboratory variables.</p> Methods <p>Of the 438 patients with AF included in this study who underwent catheter ablation between 2016 and 2023. Comprehensive demographic, clinical, echocardiographic, laboratory, medication, and risk score data were collected. The primary endpoint was AF recurrence, defined as documented AF, atrial flutter, or atrial tachycardia ≥ 30&#xa0;s occurring ≥ 3 months post-procedure. The dataset was randomly divided into training set and validation set in a 6:4 ratio. Univariate and multivariate logistic regression were used to identify independent risk factors for the risk of recurrence after catheter ablation of AF. Eleven ML algorithms were established on the training set—including random forest (RF), gradient boosting machine(GBM), logistic regression (LR), support vector machine(SVM) and XGBoost. Model performance was evaluated using receiver operating characteristic (ROC) curves, precision-recall (PR) curves, and calculating the area under the curve (AUC). A calibration curve assessed the model’s accuracy, while decision curve analysis (DCA) evaluated its clinical applicability. In addition, to avoid overfitting, we conducted an internal validation of best model using Bootstrap. Finally, Shapley additive explanations (SHAP) were employed to interpret the importance of predictor variables.</p> Results <p>Of the 438 patients with AF included in this study who underwent catheter ablation, 147 experienced recurrence during follow-up. The median age of the total population was 63 years, with 64 years in the non-recurrence group and 63 years in the recurrence group (<i>P</i> = 0.303). The proportion of females was 36.1% in the recurrence group vs. 52.6% in the non-recurrence group (<i>P</i> = 0.018). The RF model demonstrated superior performance, achieving an AUC of 0.878 in the training set and 0.925 in the validation set. It also showed excellent calibration (Brier score: 0.186) and clinical utility across a wide risk threshold range. Key predictors included alcohol consumption [OR = 2.12 (1.15–3.91), <i>P</i> = 0.017)], fibrin degradation products [FDP, OR = 1.22 (1.02–1.46), <i>P</i> = 0.027], and hypertension [OR = 0.47 (0.26–0.85), <i>P</i> = 0.012].</p> Conclusion <p>An interpretable ML model based on RF accurately predicts AF recurrence post-ablation and outperforms conventional risk scores. This tool may enhance individualized patient counseling, follow-up strategy design, and resource allocation in clinical practice.</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

Comparative study of 11 machine learning algorithms for predicting recurrence risk after atrial fibrillation catheter ablation based on a real-world cohort: a retrospective study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月14日

查看全文
  • Yicheng Wang,
  • Yi-Ming Peng,
  • Zi-Ao Fan,
  • Xiao-Lin Liao,
  • Hong-Yi Yang,
  • Jian-Quan Chen,
  • Jian-Cheng Zhang

摘要

Background

Atrial fibrillation (AF) is the most common arrhythmia worldwide, with catheter ablation being an effective yet recurrence-prone treatment. Given the limited accuracy of conventional risk scores in identifying patients at high risk of recurrence after catheter ablation, this study sought to develop and validate a machine learning (ML) model for predicting AF recurrence using a wide array of clinical and laboratory variables.

Methods

Of the 438 patients with AF included in this study who underwent catheter ablation between 2016 and 2023. Comprehensive demographic, clinical, echocardiographic, laboratory, medication, and risk score data were collected. The primary endpoint was AF recurrence, defined as documented AF, atrial flutter, or atrial tachycardia ≥ 30 s occurring ≥ 3 months post-procedure. The dataset was randomly divided into training set and validation set in a 6:4 ratio. Univariate and multivariate logistic regression were used to identify independent risk factors for the risk of recurrence after catheter ablation of AF. Eleven ML algorithms were established on the training set—including random forest (RF), gradient boosting machine(GBM), logistic regression (LR), support vector machine(SVM) and XGBoost. Model performance was evaluated using receiver operating characteristic (ROC) curves, precision-recall (PR) curves, and calculating the area under the curve (AUC). A calibration curve assessed the model’s accuracy, while decision curve analysis (DCA) evaluated its clinical applicability. In addition, to avoid overfitting, we conducted an internal validation of best model using Bootstrap. Finally, Shapley additive explanations (SHAP) were employed to interpret the importance of predictor variables.

Results

Of the 438 patients with AF included in this study who underwent catheter ablation, 147 experienced recurrence during follow-up. The median age of the total population was 63 years, with 64 years in the non-recurrence group and 63 years in the recurrence group (P = 0.303). The proportion of females was 36.1% in the recurrence group vs. 52.6% in the non-recurrence group (P = 0.018). The RF model demonstrated superior performance, achieving an AUC of 0.878 in the training set and 0.925 in the validation set. It also showed excellent calibration (Brier score: 0.186) and clinical utility across a wide risk threshold range. Key predictors included alcohol consumption [OR = 2.12 (1.15–3.91), P = 0.017)], fibrin degradation products [FDP, OR = 1.22 (1.02–1.46), P = 0.027], and hypertension [OR = 0.47 (0.26–0.85), P = 0.012].

Conclusion

An interpretable ML model based on RF accurately predicts AF recurrence post-ablation and outperforms conventional risk scores. This tool may enhance individualized patient counseling, follow-up strategy design, and resource allocation in clinical practice.