EUVIMEDObjevte EUVIMED
Open access · CC BY 4.0

DMGExNet: A Dual‐Stream Transformer‐Guided Multi‐Graph Collaborative Explainable Network for Medication Recommendation

Chao Pan, Haitao Wang, Jianfeng He, Xing Chen

Engineering Reports · 2026-07 · 8 (7)

DOI: 10.1002/eng2.70899 ↗

At a glance

Journal article in Engineering Reports, published online on 2026-07-06. Written by Chao Pan and 3 co-authors. 36 references. The full text is freely available (open access, CC BY 4.0). Funded by: National Natural Science Foundation of China.

AbstraktBibliografické údajeCiteRelated studies

Abstract (original language)

Medication recommendation is a critical research direction in healthcare. It aims to provide physicians with decision support for medication prescribing by leveraging electronic health records. However, existing studies tend to focus more on the sequential modeling of longitudinal records while neglecting the potential interactions between diagnoses and procedures, which makes it challenging to obtain an appropriate representation of patients' health status. Moreover, most methods lack interpretability in the recommendation process, and the recommendation results are often associated with a high risk of drug‐drug interactions. Therefore, we propose a Dual‐Stream Transformer‐Guided Multi‐Graph Collaborative Explainable Network (DMGExNet) for medication recommendation. Specifically, we employ a dual‐stream encoder designed to concurrently capture a patient's longitudinal medical history and cross‐sectional diagnostic and procedural interactions, generating a comprehensive patient representation. Furthermore, we introduce an aspect explanation mapping method that projects patient representation onto interpretable aspect features, which improves the interpretability of medication recommendations. Finally, we adopt a piecewise threshold strategy to minimize the prediction loss, combined with an explanation loss to optimize the recommendation performance. We conducted extensive experiments on two real‐world medical datasets, MIMIC‐III and MIMIC‐IV. The results demonstrate that DMGExNet achieves state‐of‐the‐art performance compared to existing methods, delivering higher recommendation accuracy while concurrently ensuring greater medication safety. The source code is publicly available at https://github.com/pc‐star123/DMGExNet .

Funding

  • National Natural Science Foundation of China

Cite

APA

Pan, C., Wang, H., He, J., & Chen, X. (2026). DMGExNet: A Dual‐Stream Transformer‐Guided Multi‐Graph Collaborative Explainable Network for Medication Recommendation. Engineering Reports, 8(7). https://doi.org/10.1002/eng2.70899

Vancouver

Pan C, Wang H, He J, Chen X. DMGExNet: A Dual‐Stream Transformer‐Guided Multi‐Graph Collaborative Explainable Network for Medication Recommendation. Engineering Reports. 2026;8(7). doi:10.1002/eng2.70899

BibTeX
@article{Pan2026,
  author = {Pan, Chao and Wang, Haitao and He, Jianfeng and Chen, Xing},
  title = {DMGExNet: A Dual‐Stream Transformer‐Guided Multi‐Graph Collaborative Explainable Network for Medication Recommendation},
  journal = {Engineering Reports},
  year = {2026},
  volume = {8},
  number = {7},
  doi = {10.1002/eng2.70899}
}
Start for free →