Machine learning-based prognostic model for 30-day mortality prediction in Sepsis-3

Area of Science:

• Medical informatics
• Machine learning in healthcare
• Critical care medicine

Background:

• Sepsis is a life-threatening condition in hospitalized patients, especially in the Intensive Care Unit (ICU).
• Early identification and prediction of Sepsis mortality are critical for improving patient survival rates.
• Machine learning models offer advanced capabilities for outcome prediction compared to traditional methods.

Purpose of the Study:

• To develop and validate a prognostic model for predicting 30-day mortality risk in Sepsis-3 patients.
• To utilize a Stacking-based Meta-Classifier for enhanced prediction accuracy.
• To analyze the MIMIC-III database for Sepsis patient data.

Main Methods:

• A cohort of 4,240 Sepsis-3 patients was analyzed, with 783 deaths within 30 days.
• Fifteen key biomarkers were selected using feature ranking methods (XGBoost, Random Forest, Extra Tree).
• A stacking-based meta-classifier, incorporating Logistic Regression, was employed for mortality prediction after dataset balancing (SMOTE-TOMEK LINK) and validated using cross-validation.

Main Results:

• The developed stacking classifier model demonstrated high performance, achieving 95.52% accuracy, 95.79% precision, 95.52% recall, 93.65% specificity, and a 95.60% F1-score.
• The Logistic Regression classifier within the stacking ensemble achieved an Area Under the Curve (AUC) of 0.99.
• A nomogram was generated to provide clinical insights into the significance of individual biomarkers.

Conclusions:

• The proposed stacking classifier, combined with a nomogram, effectively predicts 30-day mortality in Sepsis patients.
• This predictive model shows significant potential for facilitating early intervention strategies.
• The findings suggest a promising approach for improving treatment outcomes in Sepsis management.