Predicting Hospital Readmission: A Joint Ensemble-Learning Model.

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Hospital readmissions present a significant challenge due to high costs and prevalence.
• Accurate prediction models are crucial for identifying high-risk patients and enabling proactive interventions.
• Existing predictive analytics for hospital readmission face challenges like data dimensionality, sparsity, and class imbalance.

Purpose of the Study:

• To develop and validate an analytical framework using data-driven approaches for predicting hospital readmissions.
• To address challenges in electronic health record (EHR) data, including high dimensionality, medical code sparsity, and class imbalance.
• To improve the accuracy and effectiveness of hospital readmission prediction models.

Main Methods:

• Utilized hospital inpatient administrative data from a nationwide healthcare dataset.
• Developed a joint ensemble-learning model combining modified weight boosting and stacking algorithms.
• Explored feature engineering methods for medical vector representation and sparsity, and applied modified weight boosting to handle class imbalance.

Main Results:

• The proposed framework, using modified weight boosting, improved overall model performance by 22.7%.
• Recall was significantly enhanced, increasing from 0.726 to a maximum of 0.891 compared to benchmark models.
• The study demonstrated the effectiveness of different misclassification costs in model training for tailored predictions.

Conclusions:

• The developed ensemble learning framework effectively addresses challenges in predicting hospital readmissions from EHR data.
• The modified weight boosting algorithm significantly enhances prediction accuracy and recall.
• The model provides actionable insights for hospital practitioners to implement cost-effective readmission interventions based on CMS policies and hospital costs.