RTSANet: learning sequential-attentive hybrid networks for drug information-enhanced medication recommendation

Medication recommendation supports safe and personalized treatment based on patients' electronic health records(EHR). However, modeling complex dependencies in longitudinal records while ensuring medication safety remains challenging. We propose RTSANet, which integrates inter-visit drug similarity and external adverse drug reaction information. It captures medication relationships across visits, enhances drug representations with safety-related knowledge, and generates more accurate and safer medication combinations. Validation on the MIMIC-III and MIMIC-IV datasets demonstrates that RTSANet achieves 0.5360 Jaccard, 0.6895 F1 and 0.7869 PRAUC, improving recommendation accuracy and medication safety.