Related Experiment Video
Updated: May 11, 2025
07:42
A Data-Driven Approach to Quantifying Immune States in Sepsis
Published on: February 7, 2025
119
Explainable machine learning model for predicting septic shock in critically sepsis patients based on coagulation
Qing-Bo Zeng1, En-Lan Peng2, Ye Zhou3
1Intensive Care Unit, The 908th Hospital of Chinese PLA Logistic Support Force, Nanchang, 330002, China; Intensive Care Unit, Nanchang Hongdu Hospital of Traditional Chinese Medicine, Nanchang, 330038, China.
Summary
This study developed and compared machine learning (ML) models to predict septic shock in sepsis patients. A support vector machine (SVM) model using key clinical and laboratory data demonstrated superior predictive performance compared to other ML and traditional methods.
Area of Science:
- Critical Care Medicine
- Computational Biology
- Hematology
Background:
- Septic shock significantly increases mortality in sepsis patients with coagulopathy.
- Existing methods for predicting septic shock lack systematic comparison.
- Accurate prediction of septic shock is crucial for timely intervention and improved patient outcomes.
Purpose of the Study:
- To develop and compare various machine learning (ML) algorithms and a traditional statistical method for predicting septic shock in sepsis patients.
- To identify the most effective model for early identification of patients at high risk of developing septic shock.
Main Methods:
- A retrospective cohort study of 484 sepsis patients admitted to intensive care units.
- Development of 5 ML models and 1 traditional statistical model using comprehensive coagulation and clinical data.
- Performance evaluation using area under the receiver operating characteristic curve, calibration plots, and decision curve analysis; SHapley Additive exPlanations (SHAP) used for interpretability.
Main Results:
- 37.2% of patients developed septic shock.
- The support vector machine (SVM) model achieved the highest predictive performance (Area Under Curve: 0.962 in training, 0.744 in validation).
- Key predictors identified by the SVM model included age, tissue plasminogen activator-inhibitor complex, prothrombin time, international normalized ratio, white blood cells, and platelet counts.
Conclusions:
- The SVM algorithm demonstrates superior performance in predicting septic shock compared to other ML and traditional statistical models.
- SHAP analysis enhances physician understanding of predictive model reliability and variable importance.
- The developed SVM model offers a promising tool for early septic shock prediction in clinical practice.
