Hyper-Relational Knowledge Enhanced Network for Hypertension Medication Recommendation

Hypertension is a prevalent cardiovascular disease that requires timely and precise medication management. However, previous medication recommendation studies have largely relied on analyzing electronic health records (EHR), overlooking the specialized knowledge required for hypertension treatment. Moreover, the hypertension-related knowledge contained in existing general medical knowledge graphs is overly simplistic, and the binary relation representations they employ fail to accurately represent the complex treatment logic, thus falling short of meeting medication recommendation needs. To tackle these concerns, we present a novel hyper-relational knowledge-enhanced hypertension medication recommendation model (HKRec). HKRec incorporates both professional treatment knowledge and individual characteristics of patients to provide personalized medication treatment plans. Specifically, a hyper-relational knowledge graph designed for hypertension medication treatment is first constructed. Next, we design a knowledge-driven encoder to capture the representations of hyper-relational knowledge within the graph, and develop an EHR-driven encoder to extract patient-specific features from the EHRs. By integrating medical knowledge entities and patient information, a recurrent mechanism is introduced to model the development process of patients’ hypertension conditions, thereby enabling more effective medication recommendations. Results from experiments on real-world MIMIC-III and MIMIC-IV datasets demonstrate that the HKRec model outperforms several competitive baseline methods. The approach enables physicians to create more accurate and personalized medication plans, leading to better management of hypertension and improved patient outcomes. Our code is publicly accessible at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/zk0814/HKRec</uri>.