Early sepsis mortality prediction model based on interpretable machine learning approach: development and validation...

Area of Science:

• Critical Care Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Sepsis is a life-threatening condition characterized by a dysregulated immune response to infection, leading to high mortality rates.
• Early prediction of sepsis outcomes is crucial for timely intervention and improved patient management.
• While machine learning (ML) shows promise in medical research, local validation using real-world datasets like MIMIC-IV for sepsis prognosis is limited.

Purpose of the Study:

• To develop and validate a machine learning-based prognostic model for predicting sepsis-related mortality.
• To leverage the Medical Information Mart for Intensive Care IV (MIMIC-IV) database for model development and internal validation.
• To externally validate the model's performance in a Chinese teaching hospital setting.

Main Methods:

• Utilized patient data from the MIMIC-IV database, split into training and internal validation sets.
• Developed and compared four machine learning models: logistic regression (LR), support vector machine (SVM), deep neural networks (DNN), and extreme gradient boosting (XGBoost).
• Employed Shapley additive interpretation for model interpretability and assessed predictive performance using the area under the receiver operating characteristic curve (AUROC).

Main Results:

• The study included 27,134 sepsis patients from MIMIC-IV and 487 from a Chinese hospital, with 52 selected clinical indicators.
• All developed ML models demonstrated strong discriminative ability in predicting sepsis mortality.
• XGBoost achieved the highest predictive performance, with an internal AUROC of 0.873 and external AUROC of 0.844, outperforming LR, SVM, DNN, and established clinical scoring systems.

Conclusions:

• An interpretable machine learning model was successfully developed for predicting sepsis mortality risk.
• Machine learning algorithms, especially XGBoost, significantly outperformed traditional clinical scores in forecasting sepsis-related deaths.
• The findings were validated in a Chinese teaching hospital, confirming the generalizability and effectiveness of the ML approach for sepsis outcome prediction.