Liao Gang, Ye Baning, Li Jianquan, Wen Mingxiang
Department of Critical Care Medicine, Guizhou Provincial People's Hospital, No. 52, Zhongshan East Road, Guiyang City, Guizhou Province, China.
BMC Nephrol. 2025 Aug 13;26(1):459. doi: 10.1186/s12882-025-04371-1.
Acute kidney injury (AKI) is a common and serious complication of acute pancreatitis (AP), which greatly increases the economic burden and mortality. Measuring intra-abdominal pressure (IAP) is very important for the management of patients with AP, and intra-bladder pressure (IBP) is an indirect indicator of IAP. However, research on the relationship between IBP and AKI in patients with AP is limited. Therefore, the purpose of this study is to use machine learning (ML) to develop and verify a predictive model, in order to explore the relationship between bladder pressure and AKI in patients with pancreatitis.
The clinical data from 223 patients with AP were extracted from the MIMIC-IV v2.2 database. The relationship between IBP and AKI is analyzed using restricted cubic splines. The Boruta algorithm is used to evaluate the prediction ability of IBP and select characteristic variables, and divide the data into a training set and verification set. Then, the ML algorithm is used to establish the prediction model. The predictive performance was evaluated by receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA). SHAP (Shapley addition explanation) is used to explain the ML model.
A total of 223 patients with AP were included in this study. Restricted cubic splines showed that there was a “J-shaped” correlation between IBP and AKI, indicating that the increase in patients’ IBP was significantly related to the increase in the risk of AKI (IBP ≥ 18.03, OR = 1.13(1.03–1.28), = 0.05). Using the Boruta algorithm, ten variables were determined for model development: creatinine, blood urea nitrogen (BUN), bladder pressure, calcium, vasopressin, apsiii score, urine output, potassium, sofa score, and base_excess. According to the area under the ROC curve, calibration curve, and DCA results of the training set, the XGBoost model showed excellent performance. The F1-score (harmonic mean of precision and recall) in the test set was 0.95, and the area under the receiver operating characteristic curve (AUC) was 0.968. SHAP-based bar charts and waterfall charts are used to explain the XGBoost model both globally and locally.
The relationship between IBP and AKI in patients with pancreatitis is J-shaped. The XGBoost model has the best predictive performance and can be used to help clinicians identify high-risk patients and implement early interventions to reduce the incidence of AKI.
The online version contains supplementary material available at 10.1186/s12882-025-04371-1.
急性肾损伤（AKI）是急性胰腺炎（AP）常见且严重的并发症，极大地增加了经济负担和死亡率。测量腹内压（IAP）对AP患者的管理非常重要，膀胱内压（IBP）是IAP的间接指标。然而，关于AP患者中IBP与AKI之间关系的研究有限。因此，本研究的目的是使用机器学习（ML）开发并验证一个预测模型，以探讨胰腺炎患者膀胱压力与AKI之间的关系。
从MIMIC-IV v2.2数据库中提取223例AP患者的临床数据。使用受限立方样条分析IBP与AKI之间的关系。使用Boruta算法评估IBP的预测能力并选择特征变量，将数据分为训练集和验证集。然后，使用ML算法建立预测模型。通过受试者工作特征（ROC）曲线、校准曲线和决策曲线分析（DCA）评估预测性能。使用SHAP（Shapley加性解释）来解释ML模型。
本研究共纳入223例AP患者。受限立方样条显示IBP与AKI之间存在“J形”相关性，表明患者IBP升高与AKI风险增加显著相关（IBP≥18.03，OR = 1.13(1.03–1.28)，P = 0.05）。使用Boruta算法，确定了10个用于模型开发的变量：肌酐、血尿素氮（BUN）、膀胱压力、钙、血管加压素、急性生理与慢性健康状况评分系统III（apsiii）评分、尿量、钾、序贯器官衰竭评估（sofa）评分和碱剩余。根据训练集的ROC曲线下面积、校准曲线和DCA结果，XGBoost模型表现出色。测试集中的F1分数（精确率和召回率的调和均值）为0.95，受试者工作特征曲线下面积（AUC）为0.968。基于SHAP的柱状图和瀑布图用于从全局和局部解释XGBoost模型。
胰腺炎患者中IBP与AKI之间的关系呈J形。XGBoost模型具有最佳预测性能，可用于帮助临床医生识别高危患者并实施早期干预以降低AKI的发生率。
在线版本包含可在10.1186/s12882 - 025 - 04371 - 1获取的补充材料。
