Data-driven Discovery of a Sepsis Patients Severity Prediction in the ICU via Pre-training BiLSTM Networks

Sepsis is the third-highest mortality disease in intensive care units(ICU) and expensive treatment costs, but the best treatment strategy remains uncertain. In this paper, we proposed a pre-training bidirectional LSTM Networks to predict the Sepsis severity of patients in ICU. Most previous models for severity prediction rely on the multi-task recurrent neural networks. In addition, state-of-the-art neural models based on attention mechanisms do not fully utilize information of organ systems that may be the most crucial features for severity prediction. To address these issues, we propose an end-to-end recurrent neural model which incorporates simultaneously analyses different organ systems and intuitively reflect the condition of the patients in a timely fashion. Specifically, we apply a pre-training technique in our model to combines it with labeled data via multi-task learning. Experimental results on the real-world clinical dataset (MIMIC-III), one of the most popular sepsis severity prediction tasks, demonstrate that our model outperforms existing state-of-the-art models.