Predicting Hospitalization among Medicaid Home- and Community-Based Services Users Using Machine Learning Methods.

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Public Health

Background:

• Home- and Community-Based Services (HCBS) users represent a vulnerable population with high healthcare utilization.
• Predicting hospitalizations is crucial for resource allocation and proactive patient management.

Purpose of the Study:

• To compare machine learning algorithms for predicting future hospitalizations among HCBS users.
• To identify key predictors of hospitalization within this population.

Main Methods:

• Utilized 2012 national Medicaid Analytic eXtract and Medicare Provider Analysis and Review data.
• Applied and compared multiple machine learning algorithms, including Random Forest and XGBoost.
• Calculated feature importance to determine significant predictors of hospitalization.

Main Results:

• Random Forest demonstrated the most robust predictive performance for hospitalizations, with XGBoost showing similar results.
• Key predictors identified include chronic conditions, previous hospitalizations, and use of services like ambulance, personal care, and durable medical equipment.
• Feature importance varied across algorithms, but consistent predictors emerged.

Conclusions:

• Machine learning models, particularly Random Forest, can effectively predict hospitalizations in HCBS users.
• Identifying high-risk individuals through these models can facilitate targeted early interventions.
• Proactive interventions can potentially improve health outcomes and reduce healthcare costs for HCBS users.