Physician Usage and Acceptance of a Machine Learning Recommender System for Simulated Clinical Order Entry

Area of Science:

• Medical Informatics
• Clinical Decision Support Systems

Background:

• Clinical decision support (CDS) tools can optimize patient care by reducing errors and improving physician workflows.
• Physician adoption and behavioral impact of automated CDS tools remain largely unexamined.

Purpose of the Study:

• To evaluate physician acceptance and behavioral changes when using an automated clinical order recommender system.
• To assess the performance of a clinical order recommender system compared to manual search.

Main Methods:

• A randomized controlled study involving 34 physicians using a simulated clinical order entry interface.
• Comparison of order placement and system performance with and without the recommender system across five simulated emergency cases.

Main Results:

• Physicians using the recommender system placed more orders per case (17.1 vs. 15.8) with similar time spent per case (6.7 minutes).
• The recommender system showed significantly higher recall (59% vs. 41%) and precision (25% vs. 17%) than manual search.
• Physicians positively received the system, acknowledging its workflow benefits.

Conclusions:

• Automated clinical order recommender systems can enhance physician order accuracy and efficiency.
• Further research is needed to determine the clinical impact of these tools and their integration into electronic health records.