基于机器学习的冠状动脉慢性完全闭塞患者术后不良心血管事件预测模型的构建

    Development of a machine-learning-based model for predicting major adverse cardiovascular events in patients with chronic total occlusion undergoing percutaneous coronary intervention

      摘要
    • 摘要:
      目的 本研究旨在建立一种利用机器学习识别冠状动脉慢性完全闭塞病变介入（‌chronic total occlusion percutaneous coronary intervention，CTO-PCI）‌‌治疗术后主要不良心血管事件（major adverse cardiovascular event, MACE）危险因素的临床预测模型。该模型可增强临床决策能力。
      方法 纳入2018～2021年在西京医院经皮冠状动脉介入（‌percutaneous coronary intervention，PCI）‌‌治疗的冠状动脉慢性完全闭塞（chronic total occlusion, CTO）患者843例。患者数据包括基本信息、病史、住院 MACE 事件、术前检查和干预细节。对所有患者进行为期一年的随访，记录患者MACE事件。患者被分为训练(80%)和验证(20%)数据集。使用最小绝对收缩和选择算子（least absolute shrinkage and selection operator，LASSO）回归筛选变量，训练了共计10个机器学习模型并进行5倍交叉折叠验证，最终选择随机森林方法进行预测模型构建。采用接受者操作特性（receiver operating characteristic，ROC）曲线 、决策曲线分析（decision curve analysis，DCA）和布里尔评分（Brier score，BS） 评估模型性能。并使用特征重要性分析和SHapley Additive exPlanations（SHAP）值依赖图，详细解释关键特征对模型预测结果的贡献和影响。
      结果 根据LASSO回归分析结果，共计纳入16个预测变量进行模型的构建，其中，随机森林模型在所有构建的模型中表现最佳，ROC曲线下面积（AUC）为0.759，BS为0.120，特征重要性分析显示，红细胞计数（RBC）、D-二聚体（D-Dimer）、血小板计数（PLT）、手术时间（operation time）和年龄（age）是预测CTO-PCI患者术后一年内发生MACE事件的关键特征。SHAP值依赖图进一步揭示了这些特征对模型预测结果的具体影响。
      结论 本研究成功构建了一个用于预测CTO-PCI患者术后一年内发生MACE事件的随机森林模型，该预后模型成功识别了CTO-PCI术后高危患者，有助于临床决策并降低 MACE 风险。

       

      Abstract:
      AIM  To develop a clinical prediction model using machine learning to identify risk factors for major adverse cardiovascular event (MACE) after chronic total occlusion percutaneous coronary intervention (CTO-PCI), so as to enhance the clinical decision-making capabilities.
      METHODS We included 843 CTO patients who underwent PCI treatment at Xijing Hospital from 2018 to 2021. Patient data included basic information, medical history, in-hospital MACE events, preoperative examinations and intervention details. All patients were followed up for one year to record MACE events. Patients were divided into training (80%) and validation (20%) datasets. Variables were selected using LASSO regression and a total of 10 machine learning models were trained and evaluated using 5-fold cross-validation. The Random Forest method was ultimately chosen for the prediction model. Model performance was assessed using receiver operating characteristic (ROC) curves, decision curve analysis (DCA) curves and Brier scores. Feature importance analysis and SHapley Additive exPlanations (SHAP) value dependence plots were used to detail the contribution and impact of key features on the model’s predictions.
      RESULTS Based on the results of the LASSO regression analysis, a total of 16 predictor variables were included in the model construction. Among all the models built, the random forest model performed the best, with an AUC of 0.759 and a Brier score of 0.120. Feature importance analysis revealed that RBC (red blood cell count), D-Dimer, PLT (platelet count), operation time and age are the key features in predicting the occurrence of MACE within one year after CTO-PCI. SHAP value dependence plots further illustrated the specific impacts of these features on the model’s prediction results.
      CONCLUSION This study has successfully constructed a random forest model to predict the occurrence of MACE within one year after CTO-PCI. The prognostic model effectively identifies high-risk patients post-CTO-PCI, which aids in clinical decision-making and reducing the risk of MACE.

       

      • HTML全文
      • 参考文献(30)
      • 相关文章
      • 施引文献
    • 资源附件(0)

    /

    返回文章
    返回