ExpDrug: An explainable drug recommendation model based on space feature mapping

Xuan Lu, Yanhong Hao, Furong Peng, Zheqing Zhu, Zhanwen Cheng

Abstract: Drug recommendation uses AI technology to combine a patient’s electronic health records with medical knowledge to help doctors recommend safer and more accurate drug combinations. Time-dependence of patients’ historical records plays a crucial role in existing methods, but some of them ignore or disregard the explainability of the recommendation, and some recommendation results are accompanied by high drug-drug interactions (DDIs). Therefore, an explainable drug recommendation model(ExpDrug) with low DDIs is proposed in this article. It maps the ”black-box” features of patient diagnoses, procedures and medications to the explainable aspect features, and minimizes both rating prediction loss and explainable loss to improve recommendation performance. In addition, a controllable threshold strategy is proposed to reduce the DDI rate. Extensive experiments are conducted on the MIMIC-III dataset, and the proposed ExpDrug achieves new state-of-the-art performance across safety, explainability and four accuracy metrics. The source code is publicly available at https://github.com/hyh0606/ExpDrug.