Scholarlink

<p>Accurate prognostication in the intensive care unit (ICU) is essential for delivering personalized and ethically sound care, yet it remains a challenge for high-risk patients. This study aimed to develop and validate an exploratory machine learning model for predicting ICU mortality in critically ill patients requiring both mechanical ventilation and vasoactive support. We conducted a retrospective analysis of 4816 patient records from the AmsterdamUMCdb database. The XGBoost model was developed using 66 clinical and laboratory features collected within the first 24&#xa0;h of ICU admission. The model demonstrated robust discriminatory performance in predicting in-ICU mortality at 24h from admission, achieving in internal validation an area under the receiver operating characteristic curve (AUROC) of 0.831 (95% CI [0.799–0.860]), a macro F1 score of 0.735, and a Brier score of 0.130, indicating reliable calibration. Explainable AI methods identified the minimum Glasgow Coma Scale score, patient age, and minimum platelet count as the most significant predictors. In this exploratory study, a machine learning model using single-center early ICU data showed promising performance for mortality prediction in a high-risk population. Given the lack of external or prospective validation, the model should be considered investigational and requires further validation before clinical application.</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

Early prognostication for ICU patients with combined respiratory and circulatory failure: an interpretable machine learning approach

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月25日

查看全文
  • Adam Piasecki,
  • Kamil Adamczyk,
  • Marcin Śniegowski,
  • Raffaele Mandarano,
  • Beatrice Brunoni,
  • Harm-Jan de Grooth,
  • Paul W. G. Elbers

摘要

Accurate prognostication in the intensive care unit (ICU) is essential for delivering personalized and ethically sound care, yet it remains a challenge for high-risk patients. This study aimed to develop and validate an exploratory machine learning model for predicting ICU mortality in critically ill patients requiring both mechanical ventilation and vasoactive support. We conducted a retrospective analysis of 4816 patient records from the AmsterdamUMCdb database. The XGBoost model was developed using 66 clinical and laboratory features collected within the first 24 h of ICU admission. The model demonstrated robust discriminatory performance in predicting in-ICU mortality at 24h from admission, achieving in internal validation an area under the receiver operating characteristic curve (AUROC) of 0.831 (95% CI [0.799–0.860]), a macro F1 score of 0.735, and a Brier score of 0.130, indicating reliable calibration. Explainable AI methods identified the minimum Glasgow Coma Scale score, patient age, and minimum platelet count as the most significant predictors. In this exploratory study, a machine learning model using single-center early ICU data showed promising performance for mortality prediction in a high-risk population. Given the lack of external or prospective validation, the model should be considered investigational and requires further validation before clinical application.