Developing an Interpretable Machine Learning Model to Predict in-Hospital Mortality in Sepsis Patients: A...

Area of Science:

• Critical Care Medicine
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Sepsis management requires accurate risk stratification for heterogeneous patient populations.
• Current methods struggle to adequately assess mortality risk in critically ill sepsis patients.
• Developing advanced predictive models is crucial for improving sepsis care.

Purpose of the Study:

• To develop and validate a machine learning model for predicting in-hospital mortality in critically ill sepsis patients.
• To compare the performance of the developed model against existing clinical scores and logistic regression.
• To identify key predictors of survival in sepsis patients within the first 24 hours of ICU admission.

Main Methods:

• Adult patients meeting Sepsis-3 criteria were analyzed from a tertiary medical center.
• Clinical features within the first 24 hours in the ICU were extracted and used for model development.
• Extreme Gradient Boosting (XGBoost) was trained and validated against logistic regression (LR) and established severity scores (SOFA, SAPS-II, LODS).

Main Results:

• The XGBoost model utilizing all features achieved an AUROC of 0.85 in temporal validation, outperforming LR (0.82) and other scores (SOFA: 0.63, SAPS-II: 0.73, LODS: 0.74).
• The model demonstrated robust performance across various subgroups.
• Key predictors of improved survival included increased urine output and supplemental oxygen therapy.

Conclusions:

• A novel XGBoost-based model was successfully developed and validated for predicting sepsis mortality.
• The model significantly outperforms logistic regression and traditional scoring systems in predicting in-hospital mortality for sepsis patients.
• Early clinical features within 24 hours of ICU admission are valuable for accurate sepsis risk stratification.