Scholarlink

Background <p><?tk 3?>The transition from acute kidney injury (AKI) to acute kidney disease (AKD) is a frequent sequela after liver transplantation (LT), associated with increased short-term mortality, prolonged hospitalization, and increased likelihood of progression to chronic kidney disease. Early identification of high-risk patients remains a critical unmet need. This study aimed to develop a machine learning-based model for early prediction of AKI-to-AKD transition after LT.</p> Methods <p>Patients who underwent LT and developed postoperative AKI between January 2019 and December 2022 were enrolled as the development cohort (<i>n</i> = 359), randomly allocated into a training set and an internal validation set. Six machine learning algorithms were evaluated to construct prediction models. All models were subsequently assessed in a temporal validation cohort in the same medical center (<i>n</i> = 104 from January 2024 to December 2025). The Shapley Additive Explanations (SHAP) were used for model interpretation. An online risk calculator was developed based on the final model.</p> Results <p>The incidence of postoperative AKD in the training set was 57.1%. AKD progression was associated with 8 predictors: postoperative day 7 blood urea nitrogen (BUN), preoperative serum creatinine (Scr), postoperative day 7 sodium, calcineurin inhibitor overexposure, imipenem/cilastatin use, amphotericin B use, sex, and cefoperazone/sulbactam use. The logistic regression model demonstrated the best overall performance, with an AUC of 0.805 in the internal validation cohort and 0.809 in the temporal validation cohort, with good calibration. SHAP analysis indicated that postoperative BUN, preoperative Scr, and postoperative sodium were the most influential predictors. Decision curve analysis demonstrated net clinical benefit.</p> Conclusion <p>This study developed and temporally validated an interpretable predictive model for AKI-to-AKD transition after LT, which can facilitate early identification of high-risk patients and guide individualized nephroprotective strategies.</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

Machine learning prediction model of transition from acute kidney injury to acute kidney disease after liver transplantation: a retrospective study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月4日

查看全文
  • Siyang Zeng,
  • Yuheng Xu,
  • Yuheng Luo,
  • Tianqiang Sheng,
  • Peixian Xiong,
  • Jirong Yang,
  • Chaojin Chen

摘要

Background

The transition from acute kidney injury (AKI) to acute kidney disease (AKD) is a frequent sequela after liver transplantation (LT), associated with increased short-term mortality, prolonged hospitalization, and increased likelihood of progression to chronic kidney disease. Early identification of high-risk patients remains a critical unmet need. This study aimed to develop a machine learning-based model for early prediction of AKI-to-AKD transition after LT.

Methods

Patients who underwent LT and developed postoperative AKI between January 2019 and December 2022 were enrolled as the development cohort (n = 359), randomly allocated into a training set and an internal validation set. Six machine learning algorithms were evaluated to construct prediction models. All models were subsequently assessed in a temporal validation cohort in the same medical center (n = 104 from January 2024 to December 2025). The Shapley Additive Explanations (SHAP) were used for model interpretation. An online risk calculator was developed based on the final model.

Results

The incidence of postoperative AKD in the training set was 57.1%. AKD progression was associated with 8 predictors: postoperative day 7 blood urea nitrogen (BUN), preoperative serum creatinine (Scr), postoperative day 7 sodium, calcineurin inhibitor overexposure, imipenem/cilastatin use, amphotericin B use, sex, and cefoperazone/sulbactam use. The logistic regression model demonstrated the best overall performance, with an AUC of 0.805 in the internal validation cohort and 0.809 in the temporal validation cohort, with good calibration. SHAP analysis indicated that postoperative BUN, preoperative Scr, and postoperative sodium were the most influential predictors. Decision curve analysis demonstrated net clinical benefit.

Conclusion

This study developed and temporally validated an interpretable predictive model for AKI-to-AKD transition after LT, which can facilitate early identification of high-risk patients and guide individualized nephroprotective strategies.