Scholarlink

Background <p>Cirrhosis complicated by sepsis is a common reason for intensive care unit (ICU) admission and is associated with a markedly increased risk of short-term mortality. However, early assessment of 28-day mortality risk at ICU admission remains uncertain in this population. Existing scoring systems often rely on subjective assessments or clinical variables that may not be readily available at the time of admission, highlighting the need for a simple and objective model based on routinely obtainable data for early risk stratification.</p> Methods <p>Data from the Medical Information Mart for Intensive Care (MIMIC)-IV database were used for model development and internal validation, while MIMIC-III and the multicenter eICU Collaborative Research Database (eICU) were used for external validation. Predictors were selected using least absolute shrinkage and selection operator regression combined with Boruta feature selection and clinical judgment. Multiple machine learning algorithms were compared, and logistic regression was selected as the final model.</p> Results <p>In the MIMIC-IV cohort, 3,670 patients were screened, of whom 1,615 met the inclusion criteria and were used for model development and internal validation. External validation included 207 patients from MIMIC-III and an additional multicenter cohort from eICU. Six predictors were retained: age, white blood cell count, total bilirubin, glucose, international normalized ratio, and lactate. In internal validation, the logistic regression model achieved an area under the receiver operating characteristic curve (AUROC) of 0.785 (95% CI 0.741–0.830), outperforming SOFA 0.671 (95% CI 0.617–0.725) and MELD 0.728 (95% CI 0.679–0.777). In the MIMIC-III external validation cohort, the AUROC was 0.756 (95% CI 0.690–0.821), compared with 0.663 (95% CI 0.588–0.737) for SOFA and 0.731 (95% CI 0.663–0.799) for MELD. In the multicenter eICU cohort, the model demonstrated an AUROC of 0.745 (95% CI 0.657–0.833).</p> Conclusions <p>A logistic regression model based on six routinely available variables within the first 24&#xa0;h of ICU admission demonstrated stable discriminative performance for predicting 28-day mortality in patients with cirrhosis and sepsis. The model provides a simple and objective tool to support early mortality risk stratification at ICU admission in this high-risk population.</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

Interpretable model for early prediction of 28-day mortality in patients with cirrhosis and sepsis: a multi-cohort ICU study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月20日

查看全文
  • Xin Xu,
  • Jingjing Li,
  • Haijing Yu,
  • Jiaquan Huang

摘要

Background

Cirrhosis complicated by sepsis is a common reason for intensive care unit (ICU) admission and is associated with a markedly increased risk of short-term mortality. However, early assessment of 28-day mortality risk at ICU admission remains uncertain in this population. Existing scoring systems often rely on subjective assessments or clinical variables that may not be readily available at the time of admission, highlighting the need for a simple and objective model based on routinely obtainable data for early risk stratification.

Methods

Data from the Medical Information Mart for Intensive Care (MIMIC)-IV database were used for model development and internal validation, while MIMIC-III and the multicenter eICU Collaborative Research Database (eICU) were used for external validation. Predictors were selected using least absolute shrinkage and selection operator regression combined with Boruta feature selection and clinical judgment. Multiple machine learning algorithms were compared, and logistic regression was selected as the final model.

Results

In the MIMIC-IV cohort, 3,670 patients were screened, of whom 1,615 met the inclusion criteria and were used for model development and internal validation. External validation included 207 patients from MIMIC-III and an additional multicenter cohort from eICU. Six predictors were retained: age, white blood cell count, total bilirubin, glucose, international normalized ratio, and lactate. In internal validation, the logistic regression model achieved an area under the receiver operating characteristic curve (AUROC) of 0.785 (95% CI 0.741–0.830), outperforming SOFA 0.671 (95% CI 0.617–0.725) and MELD 0.728 (95% CI 0.679–0.777). In the MIMIC-III external validation cohort, the AUROC was 0.756 (95% CI 0.690–0.821), compared with 0.663 (95% CI 0.588–0.737) for SOFA and 0.731 (95% CI 0.663–0.799) for MELD. In the multicenter eICU cohort, the model demonstrated an AUROC of 0.745 (95% CI 0.657–0.833).

Conclusions

A logistic regression model based on six routinely available variables within the first 24 h of ICU admission demonstrated stable discriminative performance for predicting 28-day mortality in patients with cirrhosis and sepsis. The model provides a simple and objective tool to support early mortality risk stratification at ICU admission in this high-risk population.