Medicine Package Recommendation via Dual-Level Interaction Aware Heterogeneous Graph.

Area of Science:

• Artificial Intelligence
• Medical Informatics
• Computational Medicine

Background:

• Medicine package recommendation aids clinical decision-making.
• Current methods often overlook interactions between medicine packages and other medical entities, limiting recommendation completeness.
• Existing approaches have limited access to medicine commonsense knowledge, hindering deeper insights into clinical decision processes.

Purpose of the Study:

• To propose DIAGNN, a Dual-level Interaction Aware heterogeneous Graph Neural Network for improved medicine package recommendation.
• To explicitly model interactions at both individual medicine and medicine package levels within electronic health records (EHRs).
• To integrate medication indications as commonsense knowledge to enhance recommendation accuracy.

Main Methods:

• Developed a heterogeneous graph to represent medical entities and their relationships.
• Employed a dual-level interaction aware graph convolutional network to capture semantic information.
• Incorporated medication indications into the graph as a source of commonsense knowledge.

Main Results:

• The proposed DIAGNN method demonstrated effectiveness in medicine package recommendation.
• Explicitly modeling dual-level interactions improved the completeness of recommended medicine packages.
• Integrating commonsense knowledge enhanced the model's ability to understand clinical decision-making.

Conclusions:

• DIAGNN offers a novel approach to medicine package recommendation by considering dual-level interactions and commonsense knowledge.
• The method enhances clinical decision support systems by providing more comprehensive and informed medicine package suggestions.
• Future work could explore further integration of diverse medical knowledge sources to refine recommendation algorithms.