Forecasting Hospital Readmissions with Machine Learning.

Area of Science:

• Healthcare Informatics
• Machine Learning Applications
• Health Services Research

Background:

• Hospital readmissions represent a significant economic burden on healthcare systems globally.
• Readmission rates are frequently utilized as a key performance indicator for healthcare institutions.
• Predicting patient readmissions enables proactive interventions and enhanced post-discharge planning, mitigating adverse events and reducing costs.

Purpose of the Study:

• To evaluate the efficacy of four distinct machine learning models in forecasting hospital readmissions.
• To identify the most accurate model for predicting patients likely to be readmitted.

Main Methods:

• Utilized a dataset of 11,172 hospitalization records from the General Hospital of Komotini "Sismanogleio".
• Employed four machine learning models: Support Vector Machines (linear kernel), Support Vector Machines (RBF kernel), Balanced Random Forests, and Weighted Random Forests.
• Included 24 independent variables encompassing administrative, medical-clinical, and operational data.

Main Results:

• The Balanced Random Forest model demonstrated superior performance compared to other models.
• Achieved a sensitivity of 0.70 and an Area Under the Curve (AUC) of 0.78.
• Indicated the potential of machine learning for effective readmission prediction.

Conclusions:

• Machine learning, particularly the Balanced Random Forest model, offers a promising approach to accurately predict hospital readmissions.
• Effective readmission forecasting can lead to optimized resource allocation and improved patient outcomes.
• Implementing predictive models can significantly contribute to reducing healthcare expenditures associated with preventable readmissions.