Predictive Performance of Machine Learning Models for Heart Failure Readmission: A Systematic Review

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Heart failure (HF) patients face high readmission rates, increasing healthcare expenditures.
• Predictive modeling for HF readmissions is crucial for cost containment and improved patient management.

Purpose of the Study:

• To systematically review machine learning (ML) applications for predicting heart failure readmissions.
• To assess the performance and characteristics of ML models used in HF readmission prediction.

Main Methods:

• A systematic review following PRISMA guidelines was performed across multiple medical databases.
• Data extraction and screening were conducted independently by three reviewers.
• Studies utilizing machine learning algorithms for heart failure readmission prediction were included.

Main Results:

• Twenty-two studies from six countries were analyzed, examining various readmission time frames (30-day to 3-year).
• Supervised learning algorithms demonstrated high predictive accuracy, with Area Under the Curve (AUC) values ranging from 0.70 to 0.99.
• Unsupervised algorithms achieved AUCs between 0.69 and 0.72.

Conclusions:

• Machine learning models can accurately predict heart failure hospitalizations across diverse time intervals.
• Supervised ML approaches, enhanced by clinical knowledge, show promise for improving predictive model performance.
• Enhanced collaboration between healthcare providers and data scientists is essential for developing superior predictive models to improve patient outcomes and reduce costs.