Adaptive Multi-Hop Deep Learning based Drug Recommendation System with Selective Coverage Mechanism

The goal of drug recommendation is to generate drug prescriptions based on patient’s electronic medical records and provide clinical decision support for doctors. Extracting temporal patterns and contextual information in electronic medical records is the key to successful drug recommendation. However, previous studies have ignored the relationship between patients. As a result, there are differences in the data volume of medical records among different patients, and it is impossible to adjust the focus of attention and the number of data reading iterations in the data reading process according to the individual conditions of different patients. To solve the above problems, this paper proposes a selective coverage mechanism and adaptive memory neural network, The drug recommendation model combined with reading. The model uses the neural memory network to store the temporal pattern encoding results corresponding to the patient’s health status and uses the coverage mechanism to perform data filtering and attention weight adjustment in the iterative reading process. At the same time, the model is based on the patient’s situation, Adaptively determining the number of readings of the neural memory network. Experimental results based on accurate clinical data show that this model can adaptively extract essential data from electronic medical records, construct an effective representation vector of the patient’s health status, and then complete drug recommendations.