A Knowledge-Guided Method for Disease Prediction Based on Attention Mechanism

Abstract

Disease prediction, aimed at predicting possible future diseases of patients, is a fundamental research problem in medical informatics. Many studies have proposed the introduction of external knowledge to enhance existing models with some effect, but since most of these studies only consider entities directly related to the patient, they fail to take full advantage of the correlation between entities in the knowledge graph. To this end, we propose a new approach, which uses medical knowledge graphs for multi-hop reasoning to guide the self-attention based transformer model for disease prediction. Specifically, our approach design a reinforcement learning algorithm to perform path reasoning in the knowledge graph to obtain explicit disease progression paths. Since there is a semantic gap between the Electronic Health Records (EHR) data and the knowledge path data, we feed them into two separate transformer encoders to obtain the embedding representation. In order to measure the importance of the different knowledge information in relation to the patient information, an attention module is introduced to obtain a global attention representation. Experimental results on the real-world medical dataset MIMIC-III show the superiority of the proposed approach compared to a series of state-of-the-art baselines. At the same time, multi-hop knowledge paths bring stronger interpretability for disease prediction.