Scholarlink

Objective <p>Sepsis is a life-threatening condition characterized by a dysregulated host response to infection. Although serum creatinine and albumin are well-established prognostic biomarkers, the clinical utility of the creatinine-to-albumin ratio (CAR) in sepsis remains insufficiently elucidated. This study aimed to investigate the nonlinear association between CAR and 28-day mortality and to develop a machine learning-based prediction model for risk stratification in critically ill patients with sepsis.</p> Methods <p>Clinical data from 5511 sepsis patients were extracted from the MIMIC-IV (v3.1) database, with external validation conducted in 545 patients from the Intensive Care Unit (ICU) of Lanzhou University Second Hospital. CAR was calculated at ICU admission, and patients were stratified into quartiles. Restricted cubic splines (RCS), Cox proportional hazards regression, and Kaplan–Meier curves were used to evaluate the CAR-mortality association. LASSO regression and the Boruta algorithm were employed for feature selection. Seven ML models were developed to predict 28-day mortality, with performance assessed via discrimination (AUC), calibration (Brier score), and decision curve analysis (DCA).</p> Results <p>CAR showed a nonlinear dose–response relationship with 28-day mortality (P for nonlinearity &lt; 0.001). In multivariable Cox regression analysis, higher CAR was associated with an increased risk of 28-day mortality (adjusted HR = 1.31, 95% CI 1.17–1.47, <i>P</i> &lt; 0.001). Both LASSO and Boruta identified CAR as a prognostic factor. The XGBoost model including CAR showed good predictive performance (AUC = 0.812, 95% CI 0.786–0.837). The calibration was acceptable (Brier score = 0.1341), and the model showed potential clinical value.</p> Conclusions <p>CAR was nonlinearly associated with 28-day and in-hospital mortality in patients with sepsis and remained associated with mortality risk. An XGBoost-based model incorporating CAR demonstrated good discriminative performance. These findings suggest that CAR may serve as a simple and readily available complementary marker for prognostic assessment in sepsis.</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

Prognostic value of the creatinine-to-albumin ratio in sepsis: a retrospective study based on machine learning prediction model development and external validation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月9日

查看全文
  • Jing Zhang,
  • Yating Lv,
  • Tongqin Li,
  • Yujing Jiang,
  • Jiaqi Wang,
  • Yamin Yuan,
  • Yanfei Meng,
  • Haiying Rui,
  • Xiaoxi Yan,
  • Miaobo Li,
  • Xiaorong Dong,
  • Bei Zhang,
  • Li Ma

摘要

Objective

Sepsis is a life-threatening condition characterized by a dysregulated host response to infection. Although serum creatinine and albumin are well-established prognostic biomarkers, the clinical utility of the creatinine-to-albumin ratio (CAR) in sepsis remains insufficiently elucidated. This study aimed to investigate the nonlinear association between CAR and 28-day mortality and to develop a machine learning-based prediction model for risk stratification in critically ill patients with sepsis.

Methods

Clinical data from 5511 sepsis patients were extracted from the MIMIC-IV (v3.1) database, with external validation conducted in 545 patients from the Intensive Care Unit (ICU) of Lanzhou University Second Hospital. CAR was calculated at ICU admission, and patients were stratified into quartiles. Restricted cubic splines (RCS), Cox proportional hazards regression, and Kaplan–Meier curves were used to evaluate the CAR-mortality association. LASSO regression and the Boruta algorithm were employed for feature selection. Seven ML models were developed to predict 28-day mortality, with performance assessed via discrimination (AUC), calibration (Brier score), and decision curve analysis (DCA).

Results

CAR showed a nonlinear dose–response relationship with 28-day mortality (P for nonlinearity < 0.001). In multivariable Cox regression analysis, higher CAR was associated with an increased risk of 28-day mortality (adjusted HR = 1.31, 95% CI 1.17–1.47, P < 0.001). Both LASSO and Boruta identified CAR as a prognostic factor. The XGBoost model including CAR showed good predictive performance (AUC = 0.812, 95% CI 0.786–0.837). The calibration was acceptable (Brier score = 0.1341), and the model showed potential clinical value.

Conclusions

CAR was nonlinearly associated with 28-day and in-hospital mortality in patients with sepsis and remained associated with mortality risk. An XGBoost-based model incorporating CAR demonstrated good discriminative performance. These findings suggest that CAR may serve as a simple and readily available complementary marker for prognostic assessment in sepsis.