Explainable machine learning for predicting 30-day readmission in acute heart failure patients

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Acute heart failure (AHF) is a leading cause of hospital readmissions.
• Predicting 30-day readmission risk is crucial for effective patient management and resource allocation.

Purpose of the Study:

• To develop and validate a machine learning-based predictive model for identifying patients at high risk of 30-day unplanned readmission after hospitalization for AHF.

Main Methods:

• A cohort of 2232 AHF patients was analyzed.
• Clinical variables were selected using variance inflation factor and 5-fold cross-validation.
• Five machine learning algorithms were applied, with performance evaluated using sensitivity, specificity, and AUC.
• SHapley Additive exPlanations (SHAP) were used for result interpretation.

Main Results:

• The XGBoost model achieved the highest performance with an AUC of 0.763 (0.703-0.824).
• The model demonstrated a sensitivity of 0.660 and an accuracy of 0.709.
• The XGBoost model outperformed the traditional logistic regression (LR) model in predicting readmission risk.

Conclusions:

• An optimal XGBoost model was developed for predicting 30-day unplanned readmission risk in AHF patients.
• This machine learning approach offers a significant improvement over conventional methods for risk stratification.