Explainable Machine Learning-Based Risk Prediction Model for In-Hospital Mortality after Continuous Renal Replacement...

Area of Science:

• Nephrology
• Intensive Care Medicine
• Biomedical Informatics

Background:

• In-hospital mortality remains a significant concern for intensive care unit (ICU) patients requiring continuous renal replacement therapy (CRRT).
• Accurate and personalized risk prediction is crucial for timely intervention and improved patient management.

Purpose of the Study:

• To establish an explainable and personalized machine learning (ML) model for predicting in-hospital mortality risk following CRRT initiation.
• To identify key clinical factors influencing mortality risk in this patient population.

Main Methods:

• A retrospective cohort study of 2932 ICU patients receiving CRRT was conducted.
• Recursive feature elimination with 10-fold cross-validation was employed to select optimal features for ML model development.
• Explainable AI techniques, including SHapley Additive exPlanation (SHAP), were utilized for model interpretation.

Main Results:

• Extreme gradient boosting and gradient boosting machine models demonstrated high discrimination ability (AUCs 0.806 and 0.823, respectively).
• Key predictors of mortality included Acute Physiology and Chronic Health Evaluation II score, albumin level, and timing of CRRT initiation.
• SHAP analysis provided visual interpretation of individual risk predictions, highlighting the impact of various clinical features.

Conclusions:

• Explainable ML models, integrated with SHAP, can effectively predict in-hospital mortality risk in CRRT patients.
• These models offer valuable insights into critical risk factors, empowering clinicians to make informed decisions and potentially reduce mortality.