ACDNet: Attention-guided Collaborative Decision Network for effective medication recommendation.

Area of Science:

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

Background:

• Medication recommendation from Electronic Health Records (EHR) is complex due to intricate medical data.
• Existing personalized recommendation models often suffer from inadequate patient representation and fail to consider medication record similarity.
• A gap exists in accurately modeling longitudinal patient data for precise medication suggestions.

Purpose of the Study:

• To propose an Attention-guided Collaborative Decision Network (ACDNet) for enhanced medication recommendation from EHR.
• To improve patient representation and incorporate medication-medicine similarity for more accurate recommendations.
• To validate the effectiveness of ACDNet against state-of-the-art models using real-world medical datasets.

Main Methods:

• Developed ACDNet, integrating attention mechanisms and Transformer architecture to model historical patient visits globally and locally.
• Implemented a collaborative decision framework that leverages the similarity between medication records and medicine representations.
• Evaluated ACDNet on the MIMIC-III and MIMIC-IV datasets.

Main Results:

• ACDNet significantly outperformed existing state-of-the-art models in medication recommendation tasks, achieving superior Jaccard, PR-AUC, and F1 scores.
• Ablation experiments confirmed the significant contribution of each module within the ACDNet architecture.
• A case study demonstrated the practical applicability and value of ACDNet in real-world healthcare settings.

Conclusions:

• ACDNet offers a superior approach to medication recommendation by effectively capturing patient conditions and medication histories.
• The model's ability to consider medication record similarity enhances recommendation accuracy and clinical utility.
• ACDNet shows strong potential for integration into clinical decision support systems for personalized patient care.