SuperOrder: Provider order recommendation system for outpatient clinics

Area of Science:

• Health Informatics
• Clinical Decision Support Systems
• Machine Learning in Healthcare

Background:

• Outpatient clinics face challenges in efficiently managing and predicting patient orders.
• Electronic health records (EHR) contain valuable data for improving clinical workflows.
• Accurate order prediction can streamline appointment preparation and reduce clinician workload.

Purpose of the Study:

• To develop and evaluate SuperOrder, a novel order recommendation system for outpatient clinics.
• To enhance the accuracy of predicting necessary medical orders for upcoming appointments.
• To leverage EHR data and advanced machine learning techniques for improved clinical efficiency.

Main Methods:

• A two-level prediction framework was designed, combining aggregated machine learning models at the base-level.
• Meta-level predictions were generated by integrating base-level predictions with an order co-occurrence network.
• Retrospective data from pulmonary clinics (April 2014 - March 2015) across five hospital sites were utilized for feasibility testing.

Main Results:

• The meta-level predictions demonstrated a significant improvement in recall (approximately 20%) compared to the base-level.
• A minor decrease in precision (6%) was observed at the meta-level.
• The integration of the order co-occurrence network proved effective in enhancing prediction performance.

Conclusions:

• The SuperOrder system, utilizing a two-level prediction framework with an order co-occurrence network, significantly improves the recall of outpatient order predictions.
• This approach offers a more effective and efficient method for placing outpatient orders.
• The findings suggest a promising application of AI in optimizing clinical workflows and patient care in outpatient settings.