Predicting Emergency Department Visits.

Area of Science:

• Public Health
• Health Services Research
• Data Science

Background:

• Emergency departments (EDs) face high utilization, with a significant portion of visits for non-emergency conditions.
• Identifying patients with high ED utilization is crucial for resource allocation and care management.

Purpose of the Study:

• To prospectively identify high utilizers of emergency departments.
• To develop predictive models for repeat ED visits within one, three, and six months.

Main Methods:

• Utilized routinely recorded registration data from the Indiana Public Health Emergency Surveillance System.
• Trained and tested several predictive models, including Random Forest, to forecast revisits.
• Evaluated model performance using the area under the receiver operating characteristic curve (AUROC).

Main Results:

• Random Forest models demonstrated strong performance and calibration across all outcome periods (1, 3, 6 months).
• Achieved an AUROC of at least 0.96, indicating high predictive accuracy.
• Non-linear interactions among variables significantly contributed to the models' high performance.

Conclusions:

• Predictive modeling can accurately identify patients likely to revisit the ED.
• This capability enables targeted interventions to ensure appropriate care access outside the ED.
• Proactive identification can optimize resource utilization and improve patient outcomes.