Medication stocking recommendation, which proactively provides medication suggestions by forecasting disease evolution, is critical for optimizing preventive healthcare inventory management. To address challenges such as the long-tail distribution of medical codes in electronic health records (EHRs), insufficient multi-source relational mining, and the lack of inter-task coordination mechanisms, it is essential to enhance representation learning capabilities under complex medical associations. This paper proposed a multi-relational enhanced medication stocking recommendation method based on hyperbolic graph contrastive learning (HGCL-MR). The method first integrated diagnosis ontologies, medication ontologies, and clinical statistical patterns to construct a heterogeneous multi-source prior knowledge graph for representing complex medical associations. On this basis, a dual-view hyperbolic graph contrastive learning mechanism was designed, which utilized the Poincaré ball model to align ontological knowledge with statistical features in a unified hyperbolic space. By leveraging the negative curvature properties of hyperbolic space, the model precisely captured the hierarchical structure of medical codes and enhanced the representation capability for long-tail nodes. To explicitly model inter-task interaction logic, a learnable association matrix with soft constraints was introduced to achieve collaborative optimization between disease prediction and medication recommendation through cross-task knowledge transfer and prior guidance. Finally, a hierarchical multi-task decoding architecture was adopted, utilizing a sequential learning network to capture dynamic patient disease evolution and performing cross-space mapping from disease prediction vectors to the medication recommendation space via the association matrix. Experimental results on the MIMIC-III dataset demonstrated that HGCL-MR improved Jaccard scores by approximately 3.5% and 3.6% in the two respective tasks, and increased F1 scores by about 1.9% and 1.8%. Ablation studies validated the gains of hyperbolic geometric representation in alleviating label imbalance. The model achieved balanced performance when the weight parameters were set to and . The proposed HGCL-MR can alleviate modeling difficulties caused by medical data sparsity and improve prediction performance by explicitly characterizing task associations, thereby providing robust technical support for intelligent medical decision-making.
RAJKOMARA, ORENE, CHENK, et al. Scalable and accurate deep learning with electronic health records[J]. npj Digital Medicine, 2018, 1: 18.
[2]
CHOIE, BAHADORIM T, SONGL, et al. GRAM: Graph-based attention model for healthcare representation learning[C]//Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2017: 787-795.
[3]
NICKELM, KIELAD. Poincaré embeddings for learning hierarchical representations[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems, 2017: 6341-6350.
[4]
CHAMII, YINGR, RÉC, et al. Hyperbolic graph convolutional neural networks[C]//Advances in Neural Information Processing Systems. 2019(32): 4869-4880.
[5]
ANY, TANGH, JINB, et al. KAMPNet: Multi-source medical knowledge augmented medication prediction network with multi-level graph contrastive learning[J]. BMC Medical Informatics and Decision Making, 2023, 23(1): 243.
[6]
ANY, MAOY, ZHANGL, et al. RAHM: Relation augmented hierarchical multi-task learning framework for reasonable medication stocking[J]. Journal of Biomedical Informatics, 2020, 108: 103502.
[7]
CHOIE, BAHADORIM T, SCHUETZA, et al. Doctor AI: Predicting clinical events via recurrent neural networks[DB/OL]. (2016-09-28) [2025-10-23].
[8]
CHOIE, BAHADORIM T, SUNJ, et al. RETAIN: An interpretable predictive model for healthcare using reverse time attention mechanism[DB/OL]. (2017-02-26) [2025-10-23].
SHANWenqi, WANGBo, HUANGQingsong, et al. Disease prediction based on weighted hierarchical attention mechanism[J]. Journal of Chinese Information Processing, 2023, 37(1): 97-103. (in Chinese)
[11]
SHANGJ, XIAOC, MAT, et al. GAMENet: Graph augmented Memory networks for recommending medication combination[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2019, 33(1): 1126-1133.
[12]
YANGC, XIAOC, MAF, et al. SafeDrug: Dual molecular graph encoders for recommending effective and safe drug combinations[DB/OL]. (2022-07-17) [2025-10-23].
[13]
RASMYL, XIANGY, XIEZ, et al. Med-BERT: Pretrained contextualized embeddings on large-scale structured electronic health records for disease prediction[J]. npj Digital Medicine, 2021, 4: 86.
ZHAOZhenzhen, DONGYanru, LIUJing, et al. Medical named entity recognition incorporating word information and graph attention[J]. Computer Engineering and Applications, 2024, 60(11): 147-155. (in Chinese)
SHENXiyu, CAIXiaohong, CAOHui. Research progress of recommendation system based on medical knowledge graph[J]. Computer Engineering and Applications, 2023, 59(19): 40-51. (in Chinese)
[18]
WANGH, ZHAOM, XIEX, et al. Knowledge graph convolutional networks for recommender systems[C]//The World Wide Web Conference, 2019: 3307-3313.
[19]
SHANGJ, MAT, XIAOC, et al. Pre-training of graph augmented transformers for medication recommendation[C]//Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, 2019: 5953-5959.
[20]
VELIČKOVIĆP, FEDUSW, HAMILTONW, et al. Deep graph infomax[DB/OL]. (2018-12-21) [2025-10-23].
[21]
YOUY N, CHENT L, SUIY D, et al. Graph contrastive learning with augmentations[DB/OL]. (2021-04-03)[2025-10-23].
[22]
ZHUY, XUY, YUF, et al. Deep graph contrastive representation learning[DB/OL]. (2020-07-13) [2025-10-23].
[23]
CAOP, CHENY, LIUK, et al. HyperCore: Hyperbolic and co-graph representation for automatic ICD coding[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020: 3105-3114.
WANGQiang, JIANGHao, YIShuwen, et al. Hyperbolic representation learning for complex networks[J]. Journal of Software, 2021, 32(1): 93-117. (in Chinese)
[26]
YANGX, CHANGH, LAIZ, et al. Hyperbolic contrastive learning for cross-domain recommendation[C]//Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, 2024: 2920-2929.
[27]
JOHNSONA E W, POLLARDT J, SHENL, et al. MIMIC-III, a freely accessible critical care database[J]. Scientific Data, 2016, 3: 160035.