Heart Failure Emergency Readmission Prediction Using Stacking Machine Learning Model.

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Healthcare
• Clinical Data Science

Background:

• Heart failure (HF) is a critical condition with high mortality and reduced quality of life.
• Emergency hospital readmissions are frequent in HF patients, often due to suboptimal management.
• Early identification and intervention are key to reducing readmission rates and improving patient prognosis.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting emergency readmissions in discharged heart failure patients.
• To leverage Electronic Health Record (EHR) data for predictive modeling of HF readmissions.
• To assess the efficacy of various ML models and feature selection techniques in this prediction task.

Main Methods:

• Utilized a dataset comprising 166 clinical biomarkers from 2008 heart failure patient records.
• Investigated three distinct feature selection techniques.
• Evaluated 13 classical machine learning models using five-fold cross-validation.
• Developed a stacking ML model integrating predictions from the top three performing models for final classification.

Main Results:

• The stacking ML model achieved high performance metrics: 89.41% accuracy, 90.10% precision, 89.41% recall, 87.83% specificity, 89.28% F1-score, and an AUC of 0.881.
• Demonstrated the significant effectiveness of the proposed ML approach in predicting emergency readmissions.
• Indicated that the model can reliably identify heart failure patients at high risk for readmission.

Conclusions:

• The developed stacking ML model is effective in predicting emergency readmissions for heart failure patients.
• Healthcare providers can utilize this model for proactive interventions, thereby reducing readmission risks.
• Implementing this predictive model can lead to improved patient outcomes and decreased healthcare expenditures.