Knowledge enhanced attention aggregation network for medicine recommendation.

Area of Science:

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

Background:

• Deep learning shows promise in medical applications, especially for medicine recommendations.
• Patient clinical records contain redundant information that can affect health outcomes.
• Existing models often fail to capture the impact of individual diagnoses and procedures, limiting patient representation and recommendation accuracy.

Purpose of the Study:

• To propose KEAN, a novel medicine recommendation model.
• To improve patient representation by aggregating diagnoses and procedures.
• To enhance medicine recommendation accuracy and patient safety by incorporating medical knowledge and reducing drug-drug interactions.

Main Methods:

• Developed KEAN, a model integrating an attention aggregation network and enhanced graph convolution.
• Aggregated individual diagnoses and procedures within patient visits to identify significant disease-affecting features.
• Incorporated complex medicine combination knowledge and reduced drug-drug interactions (DDIs).

Main Results:

• KEAN demonstrated superior performance compared to existing advanced methods on the MIMIC-III dataset.
• The model effectively captures significant features impacting patient diseases.
• Incorporating medical knowledge and DDIs improved the quality of medicine recommendations.

Conclusions:

• KEAN offers an effective approach for medicine recommendation by enhancing patient representation.
• The model's ability to aggregate clinical information and leverage medical knowledge is crucial for accuracy.
• This work highlights the potential of advanced deep learning techniques in personalized medicine.