RAHM: Relation augmented hierarchical multi-task learning framework for reasonable medication stocking.

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Preventive Medicine

Background:

• Active medication stocking is crucial for secondary preventive healthcare.
• Predicting patient medication needs is complex and existing models often fail to capture disease-medication relationships.
• Current approaches typically focus on single tasks like disease prediction or medication recommendation.

Purpose of the Study:

• To develop an advanced framework for accurate medication stocking in preventive healthcare.
• To address the limitations of existing models by incorporating hierarchical and relational information.
• To improve patient-specific medication prediction using multi-level patient representations.

Main Methods:

• Proposed a relation augmented hierarchical multi-task learning framework (RAHM).
• Utilized Electronic Health Record (EHR) data to learn patient visit representations.
• Employed a regular LSTM for disease-level representation and a relation-aware LSTM (R-LSTM) for medication-level representation.
• Introduced pseudo residual structures to enhance learning and preserve relational information.

Main Results:

• RAHM demonstrated consistent superiority over baseline methods in suggesting appropriate stock medications.
• The framework effectively learned multi-level, relation-aware patient representations.
• Experimental validation on a real-world clinical dataset confirmed the method's effectiveness.

Conclusions:

• The proposed RAHM framework offers a significant advancement in medication stocking for preventive healthcare.
• RAHM's ability to model complex relationships between diseases and medications improves prediction accuracy.
• This approach holds promise for optimizing digital preventive healthcare management.