Real-time sepsis severity prediction on knowledge graph deep learning networks for the intensive care unit

Sepsis is the third-highest mortality disease in intensive care units (ICUs). In this paper, we proposed a deep learning model for predicting the severity of sepsis patients. Most existing models based on attention mechanisms do not fully utilize knowledge graph based information for different organ systems, such that might constitute crucial features for predicting the severity of sepsis patients. Therefore, we have employed a medical knowledge graph as a reliable and robust source of side information. End-to-end neural networks that incorporate analyses of various organ systems simultaneously and intuitively were developed in the proposed model to reflect upon the condition of patients in a timely fashion. We have developed a pre-training technique in the proposed model to combine it with labeled data by multi-task learning. Experimental results on realworld clinical datasets, MIMIC-III and eIR, demonstrate that our model outperforms state-of-the-art models in predicting the severity of sepsis patients.