Scholarlink

Background <p>Chronic total occlusion (CTO) of the coronary artery remains a major challenge in interventional cardiology. Although percutaneous coronary intervention (CTO–PCI) restores perfusion, patients remain at high risk of 1-year major adverse cardiovascular events (MACE). Traditional linear models fail to capture complex nonlinear and interactive effects among clinical and procedural factors. To address this, we developed an interpretable Naive Bayes model using a hybrid feature selection strategy combining expert knowledge with machine learning techniques.</p> Methods <p>We retrospectively analyzed 1069 CTO–PCI patients treated at Xijing Hospital (2018–2021). Three feature selection methods—manual (clinical), algorithmic (Lasso, random forest), and hybrid—were compared. After evaluating multiple machine learning (ML) algorithms, Naive Bayes achieved the best balance between accuracy and interpretability. Dimensionality reduction (principal component analysis [PCA], random forest importance) and model interpretation (SHapley Additive exPlanations [SHAP], Local Interpretable Model-agnostic Explanations [LIME]) were applied. Model performance was assessed via area under the curve (AUC), accuracy, recall, F1-score, and decision curve analysis (DCA).</p> Results <p>The hybrid selection strategy yielded superior performance (AUC 0.8281, accuracy 77.46%, recall 78.00%, F1 0.7647). DCA confirmed significant clinical utility. Key predictors included ejection fraction, cardiac biomarkers, procedural success, and metabolic indicators. Notably, combined elevations of alanine aminotransferase (ALT), triglycerides (TG), and total cholesterol (TC) emerged as a potent interactive risk factor, suggesting a “liver–metabolism–cardiovascular axis” influencing MACE outcomes.</p> Conclusions <p>This interpretable Naive Bayes model effectively predicts 1-year MACE after CTO–PCI, uncovering both established and novel nonlinear risk interactions to support individualized patient management.</p> <p><i>Clinical trial number</i><i>: </i>KY20172019-1.</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

A clinically explainable Naive Bayes model based on hybrid feature selection for predicting post-CTO–PCI MACE

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月4日

查看全文
  • Bohui Zhang,
  • Peng Han,
  • Che Wang,
  • Tiantong Yu,
  • Yan Chen,
  • Xi Zhang,
  • Boda Zhu,
  • Yuan Ning,
  • Chengxiang Li,
  • Qiling Liu,
  • Yan Chen,
  • Kun Lian

摘要

Background

Chronic total occlusion (CTO) of the coronary artery remains a major challenge in interventional cardiology. Although percutaneous coronary intervention (CTO–PCI) restores perfusion, patients remain at high risk of 1-year major adverse cardiovascular events (MACE). Traditional linear models fail to capture complex nonlinear and interactive effects among clinical and procedural factors. To address this, we developed an interpretable Naive Bayes model using a hybrid feature selection strategy combining expert knowledge with machine learning techniques.

Methods

We retrospectively analyzed 1069 CTO–PCI patients treated at Xijing Hospital (2018–2021). Three feature selection methods—manual (clinical), algorithmic (Lasso, random forest), and hybrid—were compared. After evaluating multiple machine learning (ML) algorithms, Naive Bayes achieved the best balance between accuracy and interpretability. Dimensionality reduction (principal component analysis [PCA], random forest importance) and model interpretation (SHapley Additive exPlanations [SHAP], Local Interpretable Model-agnostic Explanations [LIME]) were applied. Model performance was assessed via area under the curve (AUC), accuracy, recall, F1-score, and decision curve analysis (DCA).

Results

The hybrid selection strategy yielded superior performance (AUC 0.8281, accuracy 77.46%, recall 78.00%, F1 0.7647). DCA confirmed significant clinical utility. Key predictors included ejection fraction, cardiac biomarkers, procedural success, and metabolic indicators. Notably, combined elevations of alanine aminotransferase (ALT), triglycerides (TG), and total cholesterol (TC) emerged as a potent interactive risk factor, suggesting a “liver–metabolism–cardiovascular axis” influencing MACE outcomes.

Conclusions

This interpretable Naive Bayes model effectively predicts 1-year MACE after CTO–PCI, uncovering both established and novel nonlinear risk interactions to support individualized patient management.

Clinical trial number: KY20172019-1.