Related Experiment Video
Updated: May 10, 2025
07:42
A Data-Driven Approach to Quantifying Immune States in Sepsis
Published on: February 7, 2025
119
Interpretable machine learning model for predicting delirium in patients with sepsis: a study based on the MIMIC
Jing Fu1,2, Aifeng He1,3, Lulu Wang2
1Northern Jiangsu People's Hospital Affiliated to Yangzhou University/Clinical Medical College, Yangzhou University, Yangzhou, Jiangsu Province, China.
BMC Infectious Diseases
|April 24, 2025
Summary
This study developed an interpretable machine learning model to predict delirium risk in sepsis patients. The model, utilizing XGBoost, identified key risk factors and demonstrated improved 28-day survival prediction in sepsis care.
Area of Science:
- Critical Care Medicine
- Medical Informatics
- Machine Learning in Healthcare
Background:
- Delirium is a common complication in sepsis patients.
- Predicting delirium risk and its impact on survival is crucial for patient outcomes.
Purpose of the Study:
- To develop and validate interpretable machine learning models for predicting delirium in sepsis patients.
- To assess the impact of delirium on 28-day survival rates in this population.
Main Methods:
- Utilized the MIMIC-IV database, enrolling 10,321 adult sepsis patients.
- Developed four machine learning models: XGBoost, SVM, Logistic Regression, and Random Forest.
- Identified key predictors for delirium risk using the best-performing model.
Main Results:
- 45.45% of sepsis patients developed delirium, with significantly higher 28-day mortality.
- The XGBoost model showed the best predictive performance (AUC 0.767).
- A nomogram was constructed using hypertension, SOFA score, chlorine, hemoglobin, and creatinine.
Conclusions:
- An XGBoost-based nomogram offers a practical tool for clinicians to assess sepsis-associated delirium risk.
- This predictive model aids in understanding delirium's impact on sepsis patient survival.
- The study provides a novel approach for delirium prediction research in sepsis.
