Prediction of mortality in sepsis patients using stacked ensemble machine learning algorithm.

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Medicine
• Healthcare Data Science

Background:

• Machine learning (ML) models are increasingly explored for predicting patient outcomes in sepsis.
• Identifying reliable predictors of mortality in sepsis is crucial for timely intervention.

Purpose of the Study:

• To evaluate the utility of a stacked ensemble machine learning algorithm for predicting mortality in sepsis patients.
• To compare the performance of individual ML models against a stacked ensemble approach.

Main Methods:

• A cohort of 1,453 adult sepsis patients was analyzed.
• The Boruta algorithm identified key predictors: inotrope use and assisted ventilation.
• A stacked ensemble model was developed using Random Forest, SVM, Elastic Net, and GBM as weak learners, with SVM as the meta-learner.

Main Results:

• The Random Forest model achieved the highest Area Under the Curve (AUC) of 97.91% on training data.
• Individual models showed high AUCs: SVM (95.21%), GBM (93.67%), GLM Net (91.42%).
• The stacked ensemble model achieved an AUC of 92.14%, with Random Forest demonstrating moderate accuracy (85.5%) on the test dataset.

Conclusions:

• Random Forest demonstrated high predictive accuracy on training data, with moderate performance on test data.
• The study highlights the potential of ML in sepsis mortality prediction.
• Increased availability of open-access intensive care unit (ICU) databases is recommended to further advance ML applications in healthcare.