Using the Electronic Medical Record to Identify Patients at High Risk for Frequent Emergency Department Visits and...

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• A small patient subset drives a significant portion of healthcare utilization and costs.
• Proactive identification of high-cost patients can enable targeted interventions.
• Current methods may not effectively identify future high-cost patients before significant utilization occurs.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models using electronic medical record (EMR) free text to predict future high emergency department (ED) use and total healthcare costs.
• To identify patients at risk of becoming high utilizers or high cost, who are not currently classified as such.

Main Methods:

• Utilized text data from the cumulative patient profile of 43,111 patients within an EMR.
• Processed and indexed 11,905 words to develop logistic regression models.
• Created separate training and validation cohorts, assessing outcomes (≥3 ED visits or top 5% healthcare expenditures) in the 12 months post-prediction via administrative databases.

Main Results:

• The model predicting frequent ED visits (excluding prior high users) achieved an area under the receiver operating characteristic curve (AUC) of 0.71.
• The model predicting top 5% total system costs achieved an AUC of 0.76.
• The ML model demonstrated predictive capability for future healthcare costs.

Conclusions:

• Machine learning techniques are effective in analyzing unstructured EMR free text for predictive purposes.
• EMR text data is more predictive of future high healthcare costs than future high emergency department visits.
• Further research is needed to determine if these predictions can be leveraged for cost reduction through early interventions.