Scholarlink

<p>Hypoglycemia and hyperglycemia are common complications among critically ill patients. Maintaining blood glucose levels in the normal range is crucial but challenging due to complex influencing factors. This study aimed to develop a machine learning model that predicts hypo- or hyperglycemia 6&#xa0;h in advance in patients admitted to the intensive care unit (ICU). We analyzed electronic health records of 8,853 ICU patients (1,350,097 records) from a single center in Japan (2010–2022). Hypoglycemia and hyperglycemia were defined as blood glucose levels ≤ 80&#xa0;mg/dL (4.4&#xa0;mmol/L) and ≥ 180&#xa0;mg/dL (10&#xa0;mmol/L), respectively. We developed prediction models using routinely collected ICU data, including demographic, physiological, laboratory, and treatment variables. Machine learning models were developed using eXtreme Gradient Boosting (XGBoost), random forest, neural networks, and logistic regression. The XGBoost model demonstrated the highest performance with an area under the curve (AUC) of 0.939 and an F1 score of 0.520 for predicting hypoglycemia and an AUC of 0.919 and an F1 score of 0.702 for predicting hyperglycemia. It also achieved high calibration and net benefit. The machine learning models, notably the XGBoost algorithm, accurately predicted glucose abnormalities in critically ill ICU patients. These findings support its potential as a tool for early detection and proactive management of dysglycemia in critical care.</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

Accurate prediction of hypoglycemia and hyperglycemia using machine learning in critically ill patients

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月18日

查看全文
  • Yu Ikeda,
  • Satoru Sugimoto,
  • Tetsuo Ishikawa,
  • Tatsuro Yokoyama,
  • Takehiko Oami,
  • Keisuke Tomita,
  • Eiryo Kawakami,
  • Taka-aki Nakada

摘要

Hypoglycemia and hyperglycemia are common complications among critically ill patients. Maintaining blood glucose levels in the normal range is crucial but challenging due to complex influencing factors. This study aimed to develop a machine learning model that predicts hypo- or hyperglycemia 6 h in advance in patients admitted to the intensive care unit (ICU). We analyzed electronic health records of 8,853 ICU patients (1,350,097 records) from a single center in Japan (2010–2022). Hypoglycemia and hyperglycemia were defined as blood glucose levels ≤ 80 mg/dL (4.4 mmol/L) and ≥ 180 mg/dL (10 mmol/L), respectively. We developed prediction models using routinely collected ICU data, including demographic, physiological, laboratory, and treatment variables. Machine learning models were developed using eXtreme Gradient Boosting (XGBoost), random forest, neural networks, and logistic regression. The XGBoost model demonstrated the highest performance with an area under the curve (AUC) of 0.939 and an F1 score of 0.520 for predicting hypoglycemia and an AUC of 0.919 and an F1 score of 0.702 for predicting hyperglycemia. It also achieved high calibration and net benefit. The machine learning models, notably the XGBoost algorithm, accurately predicted glucose abnormalities in critically ill ICU patients. These findings support its potential as a tool for early detection and proactive management of dysglycemia in critical care.