A Solution Approach with Ensemble-Based Learning Technology for Predicting Early Readmission Among Patients with Heart Failure (HF) Diagnosis Using Electronic Health Records (EHR)

Abstract

Heart failure is a major health issue affecting millions globally, placing a heavy burden on patients and healthcare systems. A critical challenge in managing heart failure is the high rate of early hospital readmissions. Many patients, after discharge, are readmitted within a short period, leading to further physical and emotional distress. These frequent readmissions also result in significant financial strain, both for patients and healthcare providers. Despite efforts by hospitals and government agencies, early readmission rates remain high, causing continued patient suffering and economic hardship. Health information technology (IT) provides a promising solution by utilizing data-driven approaches to predict and mitigate readmission risks. Advanced tools integrated with electronic health records (EHR) can help identify patients at higher risk of early readmission, enabling timely interventions. This approach has the potential to improve patient outcomes while alleviating the financial and logistical challenges associated with repeated hospital stays. This study explores the implementation of a Health IT solution leveraging ensemble learning, specifically an extreme gradient boost (XGBoost) algorithm, to predict early readmission risk in heart failure patients. By analyzing data from EHR, the model aims to accurately identify high-risk patients, allowing healthcare providers to take preventive measures. The findings emphasize the potential of machine learning tools to enhance healthcare efficiency and transform the management of heart failure readmissions, benefiting patients and healthcare systems alike. The XGBoost model achieved an AUC of 0.78, with a recall of 0.76 for predicting early readmissions. However, the model demonstrated high overall accuracy but struggled with lower precision (0.23) for minority class predictions due to class imbalance. The study further used SHAP to explain feature importance.

Received: 31 December 2025 | Revised: 23 March 2025 | Accepted: 19 May 2025

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support the findings of this study are available (upon request) in Physionet website at Hospitalized patients with heart failure: integrating electronic healthcare records and external outcome data. The GitHub code repository link for the implementation can be found at https://github.com/adibML007/XGBoost_early_readmission_paper.

Author Contribution Statement

Muhammad Adib Uz Zaman: Conceptualization, Methodology, Software, Validation, Resources, Writing – original draft, Writing – review & editing, Supervision, Project administration, Funding acquisition. Odunayo Gabriel Adepoju: Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Visualization.