Scholarlink

<p>Acute kidney injury is a common and critical complication in patients with community-acquired pneumonia who are admitted to intensive care units, substantially increasing their risk of short-term mortality. To enhance early clinical decision-making, we developed and validated multiple machine learning-based survival models to predict 28-day mortality using data from the Medical Information Mart for Intensive Care (MIMIC IV and MIMIC III databases). Five models were evaluated: Random Survival Forests, Gradient Boosting Machine, Lasso-Cox, CoxBoost, and Survival-SVM. Among these, the CoxBoost model demonstrated superior predictive performance with an AUC of 0.737in internal validation cohort and an AUC of 0.671 in external validation cohort, outperforming established clinical scoring systems. Decision curve analysis indicated high net benefit across a clinically relevant range of predicted risks. Key predictive features identified by model interpretation included age, vasopressor use, NSAIDs use, hemoglobin level, hypertension, and blood urea nitrogen. To improve practical application, we developed a web application that allows for individualized, real-time mortality risk prediction at the bedside. This tool may help identify high-risk patients earlier and support timely, personalized treatment strategies in critical care environments.</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

A machine learning model for predicting 28-day mortality in ICU patients with community-acquired pneumonia and acute kidney injury

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月9日

查看全文
  • Wenwen Ji,
  • Guangdong Wang,
  • Tingting Liu,
  • Mengcong Li,
  • Na Wang,
  • Tingting Li,
  • Tinghua Hu,
  • Zhihong Shi

摘要

Acute kidney injury is a common and critical complication in patients with community-acquired pneumonia who are admitted to intensive care units, substantially increasing their risk of short-term mortality. To enhance early clinical decision-making, we developed and validated multiple machine learning-based survival models to predict 28-day mortality using data from the Medical Information Mart for Intensive Care (MIMIC IV and MIMIC III databases). Five models were evaluated: Random Survival Forests, Gradient Boosting Machine, Lasso-Cox, CoxBoost, and Survival-SVM. Among these, the CoxBoost model demonstrated superior predictive performance with an AUC of 0.737in internal validation cohort and an AUC of 0.671 in external validation cohort, outperforming established clinical scoring systems. Decision curve analysis indicated high net benefit across a clinically relevant range of predicted risks. Key predictive features identified by model interpretation included age, vasopressor use, NSAIDs use, hemoglobin level, hypertension, and blood urea nitrogen. To improve practical application, we developed a web application that allows for individualized, real-time mortality risk prediction at the bedside. This tool may help identify high-risk patients earlier and support timely, personalized treatment strategies in critical care environments.