Using Ensemble Machine Learning Methods for Predicting Risk of Readmission for Heart Failure.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Heart failure readmission poses a significant healthcare burden.
• Machine learning (ML) shows promise in predicting patient risk.
• Ensemble methods can enhance ML model performance.

Purpose of the Study:

• To evaluate two ensemble ML schemes for heart failure readmission risk prediction.
• To compare the performance of ensemble models against single ML models.
• To assess the clinical impact of the developed models.

Main Methods:

• Utilized a real-world electronic medical record (EMR) dataset of 36,245 heart failure patients.
• Applied and compared two distinct ensemble ML techniques.
• Evaluated model performance using Area Under the Curve (AUC) and F1-score.
• Assessed clinical utility via decision curve analysis.

Main Results:

• Both ensemble schemes achieved comparable discriminative ability (AUC: 0.70, F1-score: 0.58).
• Ensemble models performed at least as well as, or better than, single ML models.
• Decision curve analysis indicated a 20% net benefit for ensemble models at a 0.40 probability threshold.

Conclusions:

• Ensemble ML schemes are effective for predicting heart failure readmission risk.
• These models offer improved predictive performance and clinical utility over single ML approaches.
• Ensemble models provide a valuable tool for optimizing patient management and reducing readmissions.