Scholarlink

Background <p>Acute kidney injury (AKI), a critical complication of childhood idiopathic nephrotic syndrome (INS), markedly increases the risk of chronic kidney disease (CKD) and mortality. This study developed an interpretable machine learning (ML) model for early AKI prediction in pediatric INS to enable proactive interventions and mitigate adverse outcomes.</p> Methods <p>A total of 3,390 patients and 356 hospitalized pediatric patients with INS were included in the derivation and external cohorts, respectively, from four hospitals across China. Logistic regression, Random Forest, K-nearest neighbors, Naïve Bayes, and Support Vector machines were integrated into a stacking ensemble model and optimized for class imbalance using SMOTE-Tomek. Model performance was assessed using the area under the curve (AUC), area under the precision-recall curve, sensitivity, specificity, and balanced accuracy. SHapley Additive Explanations (SHAP) analysis elucidated the importance of features, and a Random Forest model was developed to predict CKD progression in patients with AKI.</p> Results <p>Of the 3,390 patients with INS, 12.9% developed AKI. The stacking model outperformed the individual algorithms, achieving an AUC of 0.888 internally and 0.822 externally, which could be well explained by the SHAP algorithm. It was then deployed as a web-based calculator for real-time risk assessment. Key predictors included exposure to nephrotoxic antibiotics, exposure to cyclophosphamide, respiratory tract infection, urine pH, and admission times. In the AKI cohort, alanine transaminase, aspartate transaminase, and serum phosphate levels were primarily associated with the development of CKD.</p> Conclusions <p>This study presents an interpretable ML model for early AKI prediction in pediatric Chinese patients with INS, which could serve as a practical, efficient, and economical tool for preventing AKI by identifying modifiable risk factors to reduce AKI incidence and mitigate CKD progression.</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 external validation of a machine learning-based predictive model for acute kidney injury in hospitalized children with idiopathic nephrotic syndrome

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月7日

查看全文
  • Xuejun Yang,
  • De Zhang,
  • Yan Li,
  • Anshuo Wang,
  • Zongwen Chen,
  • Li Wang,
  • Li Xiao,
  • Sijie Yu,
  • Hongxing Chen,
  • Fanghong Zhang,
  • Mo Wang,
  • Shaojun Li,
  • Haiping Yang,
  • Qiu Li

摘要

Background

Acute kidney injury (AKI), a critical complication of childhood idiopathic nephrotic syndrome (INS), markedly increases the risk of chronic kidney disease (CKD) and mortality. This study developed an interpretable machine learning (ML) model for early AKI prediction in pediatric INS to enable proactive interventions and mitigate adverse outcomes.

Methods

A total of 3,390 patients and 356 hospitalized pediatric patients with INS were included in the derivation and external cohorts, respectively, from four hospitals across China. Logistic regression, Random Forest, K-nearest neighbors, Naïve Bayes, and Support Vector machines were integrated into a stacking ensemble model and optimized for class imbalance using SMOTE-Tomek. Model performance was assessed using the area under the curve (AUC), area under the precision-recall curve, sensitivity, specificity, and balanced accuracy. SHapley Additive Explanations (SHAP) analysis elucidated the importance of features, and a Random Forest model was developed to predict CKD progression in patients with AKI.

Results

Of the 3,390 patients with INS, 12.9% developed AKI. The stacking model outperformed the individual algorithms, achieving an AUC of 0.888 internally and 0.822 externally, which could be well explained by the SHAP algorithm. It was then deployed as a web-based calculator for real-time risk assessment. Key predictors included exposure to nephrotoxic antibiotics, exposure to cyclophosphamide, respiratory tract infection, urine pH, and admission times. In the AKI cohort, alanine transaminase, aspartate transaminase, and serum phosphate levels were primarily associated with the development of CKD.

Conclusions

This study presents an interpretable ML model for early AKI prediction in pediatric Chinese patients with INS, which could serve as a practical, efficient, and economical tool for preventing AKI by identifying modifiable risk factors to reduce AKI incidence and mitigate CKD progression.