目的:本研究目的是寻求处理AMI-MC这类高度不平衡数据的方法以及快速准确预测模型的构建。背景:急性心梗后机械并发症是严重威胁患者生命的疾病,早识别、早处理是改善患者预后的关键。但实际工作中对于该类患者的识别往往存在滞后性,因...目的:本研究目的是寻求处理AMI-MC这类高度不平衡数据的方法以及快速准确预测模型的构建。背景:急性心梗后机械并发症是严重威胁患者生命的疾病,早识别、早处理是改善患者预后的关键。但实际工作中对于该类患者的识别往往存在滞后性,因此快速预测高危人群是亟待解决的临床问题。方法:基于山西省心血管病医院的509例急性心肌梗死患者数据,通过多种机器学习分类器,建立了预测模型,并进行了系统的模型训练和评估。此外,针对数据集中存在的不平衡问题,我们采用了先进的不平衡处理策略,以优化模型的预测性能。结果:509例AMI-MC患者数据中,仅有45例MC数据,经不平衡处理后,各算法准确率以及召回率等评价指标得分均得到了明显提升。最高准确率(0.987)、最高敏感性(0.999)、最高精确率(0.982)、最高F1分数(0.991)、最高敏感性(0.967)。并且各个模型在处理后数据上的AUROC得分较原数据而言均有明显提升(0.69521→0.96495)。LVEF是最影响机械并发症的特征,结合其他6个重要特征构建模型,在保持高召回率(1.0)的同时,准确率也达到了0.941、敏感性为0.935、F1分数为0.966。结论:本研究为提早预测急性心肌梗死的机械并发症提供了新的视角和方法,发现了LVEF在其中的关键影响,以及不平衡处理在提升模型性能中的显著作用。Objective: The purpose of this study was to seek methods for handling highly unbalanced data such as AMI-MC and the construction of a fast and accurate prediction model. Background: Mechanical complications after acute cardiac infarction are serious life-threatening diseases, and early recognition and management are the key to improving patients’ prognosis. However, in practice, there is often a lag in the identification of such patients, so rapid prediction of high-risk groups is a clinical problem that needs to be solved urgently. Methods: Based on the data of 509 patients with acute myocardial infarction in Shanxi Provincial Cardiovascular Disease Hospital, a prediction model was established by using various machine learning classifiers and systematic model training and evaluation was performed. In addition, for the imbalance problem existing in the dataset, we adopted an advanced imbalance processing strategy to optimize the prediction performance of the model. Results: In the data of 509 AMI-MC patients, there are only 45 cases of MC data. After imbalance processing, the accuracy, recall rate, and other evaluation metrics of each algorithm were significantly improved. The highest accuracy (0.987), the highest sensitivity (0.999), the highest precision (0.982), the highest F1 score (0.991), and the highest sensitivity (0.967). Moreover, the AUROC scores of various models on the processed data have significantly improved compared to the original data (0.69521→0.96495). LVEF is the feature that most affects mechanical complications, and the model was constructed by combining the other six important features, which achieved an accuracy of 0.941, a sensitivity of 0.935, and an F1 score of 0.966, while maintaining a high recall (1.0). Conclusion: This study provides new perspectives and methods for early prediction of mechanical complications in acute myocardial infarction, identifies the key influence of LVEF in this, and the significant role of imbalance treatment in enhancing model performance.展开更多
文摘目的:本研究目的是寻求处理AMI-MC这类高度不平衡数据的方法以及快速准确预测模型的构建。背景:急性心梗后机械并发症是严重威胁患者生命的疾病,早识别、早处理是改善患者预后的关键。但实际工作中对于该类患者的识别往往存在滞后性,因此快速预测高危人群是亟待解决的临床问题。方法:基于山西省心血管病医院的509例急性心肌梗死患者数据,通过多种机器学习分类器,建立了预测模型,并进行了系统的模型训练和评估。此外,针对数据集中存在的不平衡问题,我们采用了先进的不平衡处理策略,以优化模型的预测性能。结果:509例AMI-MC患者数据中,仅有45例MC数据,经不平衡处理后,各算法准确率以及召回率等评价指标得分均得到了明显提升。最高准确率(0.987)、最高敏感性(0.999)、最高精确率(0.982)、最高F1分数(0.991)、最高敏感性(0.967)。并且各个模型在处理后数据上的AUROC得分较原数据而言均有明显提升(0.69521→0.96495)。LVEF是最影响机械并发症的特征,结合其他6个重要特征构建模型,在保持高召回率(1.0)的同时,准确率也达到了0.941、敏感性为0.935、F1分数为0.966。结论:本研究为提早预测急性心肌梗死的机械并发症提供了新的视角和方法,发现了LVEF在其中的关键影响,以及不平衡处理在提升模型性能中的显著作用。Objective: The purpose of this study was to seek methods for handling highly unbalanced data such as AMI-MC and the construction of a fast and accurate prediction model. Background: Mechanical complications after acute cardiac infarction are serious life-threatening diseases, and early recognition and management are the key to improving patients’ prognosis. However, in practice, there is often a lag in the identification of such patients, so rapid prediction of high-risk groups is a clinical problem that needs to be solved urgently. Methods: Based on the data of 509 patients with acute myocardial infarction in Shanxi Provincial Cardiovascular Disease Hospital, a prediction model was established by using various machine learning classifiers and systematic model training and evaluation was performed. In addition, for the imbalance problem existing in the dataset, we adopted an advanced imbalance processing strategy to optimize the prediction performance of the model. Results: In the data of 509 AMI-MC patients, there are only 45 cases of MC data. After imbalance processing, the accuracy, recall rate, and other evaluation metrics of each algorithm were significantly improved. The highest accuracy (0.987), the highest sensitivity (0.999), the highest precision (0.982), the highest F1 score (0.991), and the highest sensitivity (0.967). Moreover, the AUROC scores of various models on the processed data have significantly improved compared to the original data (0.69521→0.96495). LVEF is the feature that most affects mechanical complications, and the model was constructed by combining the other six important features, which achieved an accuracy of 0.941, a sensitivity of 0.935, and an F1 score of 0.966, while maintaining a high recall (1.0). Conclusion: This study provides new perspectives and methods for early prediction of mechanical complications in acute myocardial infarction, identifies the key influence of LVEF in this, and the significant role of imbalance treatment in enhancing model performance.
