Automatic detection of omissions in medication lists

Area of Science:

• Health Informatics
• Artificial Intelligence
• Clinical Pharmacy

Background:

• Incomplete patient medication lists are a significant concern, potentially leading to adverse patient outcomes.
• Current medication reconciliation processes can be error-prone and may miss crucial drug information.

Purpose of the Study:

• To propose and evaluate the application of collaborative filtering methods for enhancing medication reconciliation.
• To leverage patient data in electronic medical records and AI for improved accuracy in identifying missing medications.

Main Methods:

• Formulated medication omission detection as a collaborative filtering problem, drawing parallels with e-commerce recommendation systems.
• Employed machine learning approaches to predict missing drugs from observed patient medication lists.
• Evaluated the methodology using medication data from three long-term care facilities.

Main Results:

• Collaborative filtering successfully identified missing drugs within the top-10 predictions 40-50% of the time.
• The therapeutic class of missing drugs was identified with 50-65% accuracy across the studied clinics.
• Decision-theoretic extensions were proposed to integrate medical knowledge into the recommendation process.

Conclusions:

• Collaborative filtering shows promise as a valuable adjunct to existing medication reconciliation strategies.
• The effectiveness of this AI-driven approach can be enhanced by considering patient-specific contexts and the potential impact of drug omissions.