Department of Cardiology, Beijing Anzhen Hospital, Capital Medical University, 2nd Anzhen Road, Chaoyang District, Beijing 100029, China.
Department of Cardiology, Beijing Anzhen Hospital, Capital Medical University, 2nd Anzhen Road, Chaoyang District, Beijing 100029, China.
Curr Probl Cardiol. 2024 Dec;49(12):102870. doi: 10.1016/j.cpcardiol.2024.102870. Epub 2024 Sep 27.
Early identification of cardiogenic shock (CS) patients at risk for renal replacement therapy (RRT) is crucial for improving clinical outcomes. This study aimed to develop and validate a prediction model using readily available clinical variables.
A retrospective cohort study was conducted using data from 4,133 CS patients from the MIMIC and eICU-CRD databases. Patients from MIMIC databases were randomly divided into 80 % training and 20 % validation cohorts, while those from eICU-CRD constituted the test cohort. Feature selection involved univariate logistic regression (LR), LASSO, and Boruta methods. Prediction models for RRT were developed using stepwise selection by LR and five machine learning (ML) algorithms (naive bayes, support vector machines, k-nearest neighbors, random forest, extreme gradient boosting) in the training cohort. Model performance was evaluated in both validation and test cohorts. A nomogram was constructed based on LR model. Kaplan-Meier survival analysis assessed 28-day mortality.
The incidence of RRT was approximately 13 % across all cohorts. Ten variables were selected: age, anion gap, chloride, bun, creatinine, potassium, ast, lactate, estimated glomerular filtration rate (eGFR), and mechanical ventilation. Compared with ML models, the LR model showed superior predictive performance with an AUC of 0.731 in the validation cohort and 0.714 in the test cohort. Four variables that best predicted the need for RRT (age, lactate, mechanical ventilation, and creatinine) were used to generate the CMLA nomogram risk score. The CMLA model showed better predictive accuracy for RRT in the test cohort compared to the previous CALL-K model (AUC: 0.731 vs. 0.699, DeLong test P < 0.05). Calibration curves and decision curve analysis (DCA) indicated that the CMLA model also had good calibration (Hosmer-Lemeshow P=0.323) and clinical utility in the test cohort. Kaplan-Meier analysis indicated significantly higher 28-day mortality in the high-risk CMLA group.
A clinically applicable nomogram with four key variables was developed to predict RRT risk in CS patients. It demonstrated good performance, promising enhanced clinical decision-making.
早期识别心源性休克（CS）患者是否需要肾脏替代治疗（RRT）对于改善临床结局至关重要。本研究旨在使用易于获得的临床变量开发和验证预测模型。
使用来自 MIMIC 和 eICU-CRD 数据库的 4133 例 CS 患者的回顾性队列研究数据。MIMIC 数据库中的患者被随机分为 80%的训练和 20%的验证队列，而 eICU-CRD 中的患者则构成测试队列。特征选择涉及单变量逻辑回归（LR）、LASSO 和 Boruta 方法。使用 LR 和五种机器学习（ML）算法（朴素贝叶斯、支持向量机、k-最近邻、随机森林、极端梯度提升）的逐步选择在训练队列中开发 RRT 预测模型。在验证和测试队列中评估模型性能。根据 LR 模型构建列线图。Kaplan-Meier 生存分析评估 28 天死亡率。
所有队列中 RRT 的发生率约为 13%。选择了 10 个变量：年龄、阴离子间隙、氯、BUN、肌酐、钾、AST、乳酸、估计肾小球滤过率（eGFR）和机械通气。与 ML 模型相比，LR 模型在验证队列中的 AUC 为 0.731，在测试队列中的 AUC 为 0.714，显示出更好的预测性能。使用 4 个变量（年龄、乳酸、机械通气和肌酐）来生成 CMLA 列线图风险评分，该评分可最好地预测 RRT 的需要。与之前的 CALL-K 模型相比，CMLA 模型在测试队列中对 RRT 的预测准确性更高（AUC：0.731 与 0.699，DeLong 检验 P<0.05）。校准曲线和决策曲线分析（DCA）表明，CMLA 模型在测试队列中也具有良好的校准（Hosmer-Lemeshow P=0.323）和临床实用性。Kaplan-Meier 分析表明，CMLA 高危组的 28 天死亡率显著更高。
开发了一种具有四个关键变量的临床适用列线图，用于预测 CS 患者的 RRT 风险。它表现出良好的性能，有望增强临床决策。
