Demographic Information Initialized Stacked Gated Recurrent Unit for an Early Prediction of Sepsis

Abstract

Sepsis is a life-threatening condition that occurs when the body’s response to infection causes tissue damage, organ failure, or death. Sepsis is a major public health issue responsible for significant morbidity, mortality, and healthcare expenses. Early detection and antibiotic treatment of sepsis are critical for improving sepsis outcomes, where each hour of delayed treatment has been associated with roughly an 4-8% increase in mortality. Thus, an early detection of sepsis can have significant impact on both patient outcome and reduce in medical expenses.In recent years, deep neural networks has shown significant improvement in variety of tasks. One of the approach to apply deep neural networks to sequential data is a Recurrent Neural Netowrks (RNNs). In this study, we modify gated recurrent units (GRU), RNNs with gate structure, to predict sepsis from provided Physionet Challenge 2019 dataset. In proposed model, initial value of hidden state in GRU was determined by demographic information of patients, and two-step training was performed with customized loss function.With the proposed method, we achieved normalized utility score of 0.323 on full test set (Team name: NN-MIH).