MGRN: toward robust drug recommendation via multi-view gating retrieval network

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems

Background:

• Drug recommendation systems aim to identify safe and effective drug combinations using electronic health records.
• Existing systems struggle with comprehensive patient health status representation and patient drug history modeling.

Purpose of the Study:

• To address limitations in current drug recommendation approaches.
• To develop a robust model for improved drug combination suggestions in clinical practice.

Main Methods:

• Proposed a multi-view gating retrieval network (MGRN).
• Incorporated visit-, sequence-, and token-level views for comprehensive drug representation.
• Developed a gating drug retrieval module to leverage historical drug information.

Main Results:

• MGRN achieved new benchmark performance on MIMIC-III and MIMIC-IV datasets.
• Demonstrated significant improvements in Jaccard, PRAUC, and F1-score metrics.
• Outperformed state-of-the-art models in drug recommendation accuracy.

Conclusions:

• The MGRN offers a more robust and comprehensive approach to drug recommendation.
• The model effectively captures patient health status and drug history for better clinical decision support.
• The proposed method sets a new standard for electronic health record-based drug recommendation.