Predicting 28-Day Mortality in Critically Ill Patients Receiving Continuous Renal Replacement Therapy: A Novel...

Area of Science:

• Critical Care Medicine
• Machine Learning Applications
• Renal Replacement Therapy

Background:

• Critically ill patients undergoing continuous renal replacement therapy (CRRT) have high mortality rates.
• Early identification of high-risk patients is crucial for timely intervention and improved outcomes.
• Existing risk stratification tools may lack interpretability or generalizability.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) model for predicting 28-day all-cause mortality in CRRT patients.
• To facilitate early risk stratification and enhance clinical decision-making.
• To leverage interpretable ML for transparent risk assessment in critical care.

Main Methods:

• Analysis of data from 1362 CRRT patients (1224 training, 138 external validation).
• Feature selection using LASSO, SVM-RFE, and Boruta algorithms.
• Construction and comparison of nine ML models, including Gaussian Process (GP), with performance assessed by AUC and other metrics.
• Interpretation of ML models using SHapley Additive exPlanations (SHAP).

Main Results:

• The GP model demonstrated consistent predictive performance across training, internal, and external validation cohorts (AUCs ranging from 0.780 to 0.841).
• Key predictors identified include red cell distribution width, age, lactate, septic shock, and vasoactive drug use.
• SHAP analysis provided transparent insights into the contribution of each feature.

Conclusions:

• The developed GP-based ML model accurately predicts 28-day mortality in CRRT patients with strong generalizability.
• The integration of SHAP explanations offers an interpretable tool for clinicians.
• Early identification of high-risk patients using this interpretable model has the potential to improve clinical outcomes.