Department of Cardiology, The First Affiliated Hospital of Nanjing Medical University, Nanjing, Jiangsu, China.
Department of Neurology, The Second Affiliated Hospital of Jianghan University (Wuhan City Fifth Hospital), Wuhan, Hubei, China.
Clin Exp Nephrol. 2024 Aug;28(8):811-821. doi: 10.1007/s10157-024-02472-z. Epub 2024 Mar 27.
This study aimed to develop machine learning models for risk prediction of continuous renal replacement therapy (CRRT) following coronary artery bypass grafting (CABG) surgery in intensive care unit (ICU) patients.
We extracted CABG patients from the electronic medical record system of the hospital. The endpoint of this study was the requirement for CRRT after CABG surgery. The Boruta method was used for feature selection. Seven machine learning algorithms were developed to train models and validated using 10 fold cross-validation (CV). Model discrimination and calibration were estimated using the area under the receiver operating characteristic curve (AUC) and calibration plot, respectively. We used the SHapley Additive exPlanations (SHAP) method to illustrate the effects of the features attributed to the model and analyze the effects of individual features on the output of the mode.
In this study, 72 (37.89%) patients underwent CRRT, with a higher mortality compared to those patients without CRRT. The Gaussian Naïve Bayes (GNB) model with the highest AUC were considered as the final predictive model and performed best in predicting postoperative CRRT. The analysis of importance revealed that cardiac troponin T, creatine kinase isoenzyme, albumin, low-density lipoprotein cholesterol, NYHA, serum creatinine, and age were the top seven features of the GNB model. The SHAP force analysis illustrated how created model visualized individualized prediction of CRRT.
Machine learning models were developed to predict CRRT. This contributes to the identification of risk variables for CRRT following CABG surgery in ICU patients and enables the optimization of perioperative managements for patients.
本研究旨在开发一种机器学习模型，用于预测重症监护病房（ICU）行冠状动脉旁路移植术（CABG）后的连续肾脏替代治疗（CRRT）风险。
我们从医院的电子病历系统中提取 CABG 患者。本研究的终点是 CABG 手术后需要 CRRT。使用 Boruta 方法进行特征选择。开发了七种机器学习算法来训练模型，并使用 10 折交叉验证（CV）进行验证。使用受试者工作特征曲线下面积（AUC）和校准图分别评估模型的判别能力和校准能力。我们使用 SHapley Additive exPlanations（SHAP）方法来说明模型特征的影响，并分析单个特征对模型输出的影响。
本研究中，72 例（37.89%）患者接受了 CRRT，与未接受 CRRT 的患者相比，死亡率更高。具有最高 AUC 的高斯朴素贝叶斯（GNB）模型被认为是最终的预测模型，在预测术后 CRRT 方面表现最佳。重要性分析表明，肌钙蛋白 T、肌酸激酶同工酶、白蛋白、低密度脂蛋白胆固醇、NYHA、血清肌酐和年龄是 GNB 模型的前七个最重要特征。SHAP 力分析说明了创建的模型如何可视化个体化的 CRRT 预测。
开发了机器学习模型来预测 CRRT。这有助于识别 ICU 患者 CABG 术后发生 CRRT 的风险变量，并为患者的围手术期管理提供优化。
