In the realm of clinical decision-making, the complexity and variability inherent in patient care require advanced methods to assist healthcare professionals. This paper introduces a novel approach that leverages dynamic graph representation learning to enhance clinical decision-making assistance. By modeling EHRs data (Electronic Health Records) as discrete-time dynamic graphs and employing Graph Neural Networks (GNNs), our method captures the intricate and evolving interactions between patients and medical items. This reconceptualization of clinical decision-making as a recommendation system task aligns more closely with real-world scenarios, addressing limitations of previous methods such as limited patient coverage and delayed recommendations. Our experiments, conducted on two real-world clinical datasets, demonstrate the superior performance of our approach compared to traditional models, highlighting its practical utility and potential for future research and providing insight into the effective use of dynamic graphs in healthcare applications ( https://github.com/Nemo4110/GNNs-Driven-Clinical-Decision-Making-Assistance.git ).