Scholarlink

Background <p>Acute kidney injury (AKI) is a severe complication in patients with alcoholic cirrhosis, often necessitating renal replacement therapy (RRT). Early risk identification for RRT is critical for timely intervention and improved outcomes.</p> Methods <p>This retrospective study used the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Patients with alcoholic cirrhosis complicated by severe AKI were included, with those admitted from 2008 to 2019 (<i>n</i> = 937) assigned to the training cohort and those from 2020 to 2022 (<i>n</i> = 229) to the temporal validation cohort. Least absolute shrinkage and selection operator (LASSO) regression was applied for feature selection. Eight machine learning models were developed and validated. Model performance was assessed using area under the receiver operating characteristic curve (AUROC), precision-recall area under the curve (PRAUC), calibration, and decision curve analysis. Shapley additive explanations (SHAP) were used for model interpretation.</p> Results <p>Ten key predictors of RRT initiation were identified: AKI stage, oliguria, serum creatinine, total bilirubin, international normalized ratio, septic shock, acute respiratory failure, norepinephrine use, vasopressin use, and blood transfusion therapy. In temporal validation&#xa0;cohort, the support vector machine (SVM) model demonstrated the best overall performance with AUROC 0.864 and PRAUC 0.746, with good calibration (Brier score 0.143), and robust clinical utility. The SVM model showed significantly better discrimination than the Sequential Organ Failure Assessment (SOFA) score and comparable performance to the Model for End-Stage Liver Disease (MELD) score in the validation cohort. An optimal probability threshold of 0.139 provided high negative predictive value for identifying low-risk patients. SHAP confirmed the clinical plausibility of predictor contributions.</p> Conclusion <p>We developed and externally validated a SVM model to predict RRT initiation in patients with alcoholic cirrhosis complicated by severe AKI, providing a practical tool for individualized risk assessment. The model was implemented as an interactive web-based calculator to facilitate individualized risk assessment and support clinical decision-making.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development and validation of a machine learning model to predict the need for renal replacement therapy in alcoholic cirrhosis complicated by severe acute kidney injury

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月13日

查看全文
  • Xu Sun,
  • Jianhong Lu,
  • Minmin Fu,
  • Lei Zhong,
  • Yuan Li

摘要

Background

Acute kidney injury (AKI) is a severe complication in patients with alcoholic cirrhosis, often necessitating renal replacement therapy (RRT). Early risk identification for RRT is critical for timely intervention and improved outcomes.

Methods

This retrospective study used the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Patients with alcoholic cirrhosis complicated by severe AKI were included, with those admitted from 2008 to 2019 (n = 937) assigned to the training cohort and those from 2020 to 2022 (n = 229) to the temporal validation cohort. Least absolute shrinkage and selection operator (LASSO) regression was applied for feature selection. Eight machine learning models were developed and validated. Model performance was assessed using area under the receiver operating characteristic curve (AUROC), precision-recall area under the curve (PRAUC), calibration, and decision curve analysis. Shapley additive explanations (SHAP) were used for model interpretation.

Results

Ten key predictors of RRT initiation were identified: AKI stage, oliguria, serum creatinine, total bilirubin, international normalized ratio, septic shock, acute respiratory failure, norepinephrine use, vasopressin use, and blood transfusion therapy. In temporal validation cohort, the support vector machine (SVM) model demonstrated the best overall performance with AUROC 0.864 and PRAUC 0.746, with good calibration (Brier score 0.143), and robust clinical utility. The SVM model showed significantly better discrimination than the Sequential Organ Failure Assessment (SOFA) score and comparable performance to the Model for End-Stage Liver Disease (MELD) score in the validation cohort. An optimal probability threshold of 0.139 provided high negative predictive value for identifying low-risk patients. SHAP confirmed the clinical plausibility of predictor contributions.

Conclusion

We developed and externally validated a SVM model to predict RRT initiation in patients with alcoholic cirrhosis complicated by severe AKI, providing a practical tool for individualized risk assessment. The model was implemented as an interactive web-based calculator to facilitate individualized risk assessment and support clinical decision-making.

Graphical Abstract