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

Background: Patients with heart failure (HF) are at high risk of readmission, contributing to substantial healthcare costs. This study investigated machine learning (ML) approaches to predict HF readmissions. Methods: A systematic review was conducted using several medical databases, adhering to the PRISMA guidelines, to identify studies employing ML to predict HF readmissions. Three reviewers independently screened the articles and extracted data. Results: Twenty-two studies from six countries were included in this study. Some studies examined 30-day readmissions, whereas others assessed 90-day, 180-day, or 1- to 3-year readmissions. Fourteen studies used supervised learning algorithms, with area under the curve (AUC) values ranging from 0.70 to 0.99, and unsupervised algorithms had AUCs of 0.69 to 0.72. The average age of the patients was 73 years, with approximately equal numbers of males and females. Conclusions: ML can predict HF-related hospitalization across various time frames. Supervised ML approaches and the incorporation of clinical knowledge may enhance model performance. Collaboration between providers and data scientists is needed to improve patient outcomes and reduce costs by using more accurate predictive models.