Ensemble Risk Model of Emergency Readmissions (ERMER).

Area of Science:

• Healthcare analytics
• Predictive modeling in medicine
• Health informatics

Background:

• Approximately 50% of hospital readmissions are preventable through targeted interventions.
• Identifying patients at high risk for emergency readmission is crucial for effective preventive strategies.
• Existing predictive models struggle to adapt to dynamic healthcare systems and evolving population demographics.

Purpose of the Study:

• To develop a generic ensemble Bayesian risk model for predicting emergency hospital readmissions.
• To create a decision support tool capable of continuous adaptation to healthcare system and population changes.

Main Methods:

• Utilized England's Hospital Episode Statistics inpatient database.
• Developed an optimal feature set using a novel framework.
• Created an ensemble model by combining Bayes Point Machine (BPM) models for various patient cohorts, named Ensemble Risk Model of Emergency Admissions (ERMER).
• Trained and validated the ERMER model on multiple time-frames (1999-2004, 2000-05, 2004-09).

Main Results:

• The ERMER model demonstrated high precision (71.6-73.9%) and specificity (88.3-91.7%).
• Sensitivity ranged from 42.1-49.2%, with an Area Under the Curve (AUC) of 75.9-77.1%.
• Performance was robust and stable across different time-frames and sub-populations.

Conclusions:

• The developed decision support tool significantly outperformed previous modeling approaches.
• The ERMER framework and Bayesian model enable continuous adjustment to new features, population characteristics, and system changes.
• This adaptive model holds promise for improving the identification of at-risk patients and reducing avoidable emergency readmissions.