Related Experiment Video
Updated: Oct 4, 2025
07:42
A Data-Driven Approach to Quantifying Immune States in Sepsis
Published on: February 7, 2025
326
A Simple Weaning Model Based on Interpretable Machine Learning Algorithm for Patients With Sepsis: A Research of
Wanjun Liu1,2, Gan Tao1, Yijun Zhang1,2
1The 2nd Department of Intensive Care Unit, The Second Affiliated Hospital of Anhui Medical University, Hefei, China.
Frontiers in Medicine
|February 4, 2022
Summary
Predicting weaning from mechanical ventilation in sepsis patients is crucial. A simplified XGBoost model with four variables accurately predicts weaning success, improving patient outcomes and reducing mortality.
Area of Science:
- Critical Care Medicine
- Pulmonology
- Data Science in Healthcare
Background:
- Invasive mechanical ventilation is vital for sepsis patient prognosis.
- No specific tools exist to assess readiness for weaning from mechanical ventilation in sepsis.
- Developing a predictive model for weaning in sepsis patients is essential.
Purpose of the Study:
- To develop and validate a practical model for predicting weaning from invasive mechanical ventilation in sepsis patients.
- To identify key clinical variables associated with weaning success in sepsis.
- To create a user-friendly tool for clinical application.
Main Methods:
- Utilized data from MIMIC-IV and eICU-CRD databases.
- Employed Kaplan-Meier curves to analyze 28-day mortality.
- Developed and validated weaning prediction models using XGBoost, including a simplified four-variable version.
Main Results:
- Included 5020 (MIMIC-IV) and 7081 (eICU-CRD) sepsis patients.
- Weaning was independently associated with reduced 28-day mortality and ICU stay.
- The simplified XGBoost model demonstrated strong predictive performance (AUROC 0.75-0.78) and was developed into a web tool.
Conclusions:
- Weaning success is a significant predictor of short-term mortality in sepsis.
- The simplified XGBoost model offers robust predictive capabilities and clinical utility for weaning assessment.
- A web-based tool facilitates the clinical application of this predictive model.
