Towards a Learning Health System to Reduce Emergency Department Visits at a Population Level.

Area of Science:

• Health Services Research
• Predictive Analytics
• Population Health Management

Background:

• High Emergency Department (ED) utilization is often linked to complex patient needs requiring inter-organizational care coordination.
• Identifying individuals with high ED visit frequency is crucial for implementing effective interventions.

Purpose of the Study:

• To develop and validate a predictive model for identifying high ED utilizers.
• To implement a care coordination intervention for high ED utilizers within a Learning Health System (LHS) framework.

Main Methods:

• A random forest model was developed using electronic health record data from three health systems across two Michigan counties.
• The model predicted the number of ED visits per resident within a six-month period.
• 5-fold cross-validation and time-based validation were employed to assess model performance.

Main Results:

• The random forest model achieved a root-mean-squared error (RMSE) of 0.51 and a mean absolute error (MAE) of 0.24 using 5-fold cross-validation.
• Time-based validation yielded an RMSE of 0.74 and an MAE of 0.29.
• Patients identified as high ED utilizers were enrolled in a community-wide care coordination intervention.

Conclusions:

• The study demonstrates a Learning Health System (LHS) in action through iterative cycles of predictive modeling and intervention.
• This approach shows promise for reducing ED visits by targeting high-risk patient populations.
• Effective care coordination for high ED utilizers can be facilitated by predictive analytics within an LHS framework.