Scholarlink

Background <p>Protein-energy wasting (PEW) is a common complication of patients on maintenance haemodialysis (MHD) and is strongly associated with poor clinical outcomes; early identification and timely nutritional interventions are essential. The aim of this study was to develop and validate an explainable machine learning model to identify the risk of PEW in patients on MHD.</p> Methods <p>From January 2024 to June 2024, a total of 908 patients on MHD were enrolled. The data were randomly divided into training (<i>n</i> = 635) and test (<i>n</i> = 273) sets at a 7:3 ratio. The optimal features were selected via least absolute shrinkage and selection operator (LASSO) regression. Seven machine learning (ML) algorithms were used to develop and validate models to predict PEW risk, and the performance of the models was evaluated via evaluation indicators such as the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, F1 score, and Brier score to compare the performance of the prediction models in the test set. The best-performing model was interpreted via Shapley additive explanations (SHAPs) and deployed as a web application via Shiny R.</p> Results <p>Among the seven ML models, the AUC values ranged from 0.710 to 0.827 in the test set. The XGBoost model had the best predictive performance; achieved the highest AUC (AUC = 0.827, 95% CI, 0.772–0.883); and demonstrated good sensitivity (0.727), specificity (0.762), accuracy (0.755), and F1 score (0.544) and a lower Brier score (0.125). The SHAP method ranks the feature importance in descending order: predialysis creatinine level, handgrip strength, non-high-density lipoprotein cholesterol (non-HDL-C) level, urea clearance index (Kt/V), and high-sensitivity C-reactive protein.</p> Conclusion <p>The explainable machine learning model developed in this study is a practical tool for healthcare professionals to identify the risk of PEW in patients on MHD and provide decision support for early implementation of personalized nutrition management 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

Explainable machine learning models for predicting of protein-energy wasting in patients on maintenance haemodialysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月14日

查看全文
  • Genlian Cai,
  • Yujiao Zhang,
  • Mengyan Pan,
  • Suhan Zhou,
  • Xuejia Xiang,
  • Jinping Ying

摘要

Background

Protein-energy wasting (PEW) is a common complication of patients on maintenance haemodialysis (MHD) and is strongly associated with poor clinical outcomes; early identification and timely nutritional interventions are essential. The aim of this study was to develop and validate an explainable machine learning model to identify the risk of PEW in patients on MHD.

Methods

From January 2024 to June 2024, a total of 908 patients on MHD were enrolled. The data were randomly divided into training (n = 635) and test (n = 273) sets at a 7:3 ratio. The optimal features were selected via least absolute shrinkage and selection operator (LASSO) regression. Seven machine learning (ML) algorithms were used to develop and validate models to predict PEW risk, and the performance of the models was evaluated via evaluation indicators such as the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, F1 score, and Brier score to compare the performance of the prediction models in the test set. The best-performing model was interpreted via Shapley additive explanations (SHAPs) and deployed as a web application via Shiny R.

Results

Among the seven ML models, the AUC values ranged from 0.710 to 0.827 in the test set. The XGBoost model had the best predictive performance; achieved the highest AUC (AUC = 0.827, 95% CI, 0.772–0.883); and demonstrated good sensitivity (0.727), specificity (0.762), accuracy (0.755), and F1 score (0.544) and a lower Brier score (0.125). The SHAP method ranks the feature importance in descending order: predialysis creatinine level, handgrip strength, non-high-density lipoprotein cholesterol (non-HDL-C) level, urea clearance index (Kt/V), and high-sensitivity C-reactive protein.

Conclusion

The explainable machine learning model developed in this study is a practical tool for healthcare professionals to identify the risk of PEW in patients on MHD and provide decision support for early implementation of personalized nutrition management strategies.