Department of Biomedical Engineering, Zhejiang University, Hangzhou, China.
Department of Critical Care Medicine, Zhejiang Hospital, Hangzhou, China.
Blood Purif. 2024;53(9):704-715. doi: 10.1159/000539787. Epub 2024 Jun 12.
Continuous renal replacement therapy (CRRT) is a primary form of renal support for patients with acute kidney injury in an intensive care unit. Making an accurate decision of discontinuation is crucial for the prognosis of patients. Previous research has mostly focused on the univariate and multivariate analysis of factors in CRRT, without the capacity to capture the complexity of the decision-making process. The present study thus developed a dynamic, interpretable decision model for CRRT discontinuation.
The study adopted a cohort of 1,234 adult patients admitted to an intensive care unit in the MIMIC-IV database. We used the eXtreme Gradient Boosting (XGBoost) machine learning algorithm to construct dynamic discontinuation decision models across 4 time points. SHapley Additive exPlanation (SHAP) analysis was conducted to exhibit the contributions of individual features to the model output.
Of the 1,234 included patients with CRRT, 596 (48.3%) successfully discontinued CRRT. The dynamic prediction by the XGBoost model produced an area under the curve of 0.848, with accuracy, sensitivity, and specificity of 0.782, 0.786, and 0.776, respectively. The performance of the XGBoost model was far superior to other test models. SHAP demonstrated that the features that contributed most to the model results were the Sequential Organ Failure Assessment score, serum lactate level, and 24-h urine output.
Dynamic decision models supported by machine learning are capable of dealing with complex factors in CRRT and effectively predicting the outcome of discontinuation.
连续性肾脏替代治疗（CRRT）是重症监护病房急性肾损伤患者肾脏支持的主要形式。准确决策停止治疗对于患者的预后至关重要。先前的研究大多集中在 CRRT 中因素的单变量和多变量分析上，而没有能力捕捉决策过程的复杂性。因此，本研究开发了一种用于 CRRT 停止治疗的动态、可解释的决策模型。
本研究采用了 MIMIC-IV 数据库中 1234 名成年重症监护病房患者的队列。我们使用极端梯度提升（XGBoost）机器学习算法在 4 个时间点构建了动态停止决策模型。通过 SHapley Additive exPlanation（SHAP）分析展示了单个特征对模型输出的贡献。
在 1234 名接受 CRRT 的患者中，596 名（48.3%）成功停止了 CRRT。XGBoost 模型的动态预测产生了 0.848 的曲线下面积，准确率、敏感度和特异度分别为 0.782、0.786 和 0.776。XGBoost 模型的性能远优于其他测试模型。SHAP 表明，对模型结果贡献最大的特征是序贯器官衰竭评估评分、血清乳酸水平和 24 小时尿量。
机器学习支持的动态决策模型能够处理 CRRT 中的复杂因素，并有效预测停止治疗的结果。
