Attention mechanism and graph augmented network for medication recommendation

In order to assist doctors in making better medical decisions, the model recommends accurate and safe drug combinations for patients by analyzing diagnosis, surgery, historical medication records and drug interaction knowledge base of a large number of patients. This paper uses a double attention mechanism to capture the influence of multiple internal members of diagnosis and surgery in the stage of patient feature representation, and considers the similarity of patients' personal history medication information in drug recommendation. In order to reduce the adverse reactions of the recommended drugs, the algorithm uses a graph neural network to model the fusion of the historical drug combination and drug interaction knowledge base and calculates the correlation with patients on this basis. The method has been extensively tested on public electronic medical record data. Experimental results of the proposed model outperformed all baselines in predicting accurate and safe drug recommendations, with a 1.02 percent improvement in accuracy and a 27 percent reduction in the proportion of drug interactions compared to several advanced methods. The drug recommendation method does not take into account the dynamic change of drug use over time of disease development. The drug recommendation model combining attention mechanism and graph neural network is effective and safe, and can perceive the influence of patient characteristics on drug use, so as to provide more accurate reference for doctors to recommend clinical drug use.