Early Prediction of Sepsis in the Intensive Care Unit Using the GRU-D-MGP-TCN Model

Abstract

Sepsis is a life-threatening condition with significant risk to individuals, most prevalent in intensive care units (ICUs). Early diagnosis and prompt treatment are crucial to reducing sepsis-related mortality. Research on developing an artificial intelligence-based clinical decision support system for the early prediction of sepsis is underway. However, a state-of-the-art model has not yet been developed. In this study, we developed a predictive model for the early detection of sepsis by leveraging advanced machine learning techniques, specifically the Gated Recurrent Unit (GRU-D) and Multitask Gaussian Process-Temporal Convolutional Network (MGP-TCN) models. This newly developed model demonstrated improved performance compared to existing results, with an area under the precision-recall curve of 0.965 (0.710) from 0.689 (0.432) and an area under the receiver operating characteristic curve of 0.994 (0.924) from 0.915 (0.828). This study demonstrates the potential of the GRU-D-MGP-TCN model to improve sepsis prediction in the ICU setting, thereby contributing to a significant reduction in sepsis-related mortality through timely intervention.

Keywords

• GRU-D
• intensive care unit
• MGP-TCN
• sepsis