Temporal Modeling of Traumatic Patient for Early Sepsis Onset Prediction as Rare Event in ICU

Abstract

Sepsis is a hyper-inflammatory syndrome that develops in a patient's body in response to the presence of infection. It leads to severe organ dysfunction and is one of the leading causes of mortality that occur in Intensive Care Units (ICUs) worldwide. These complications can be reduced through early application of antibiotics, hence the ability to anticipate the onset of sepsis early is crucial to the survival and well-being of patients. Current machine learning algorithms deployed inside medical infrastructures have demonstrated poor performance and are insufficient for anticipating sepsis onset. In recent years, there has been a substantial effort developing various deep learning methodologies for predicting sepsis, but many fail to capture the time of onset and in a manner suitable for integration into medical facilities. In this study, we propose a temporal deep learning framework that can capture the temporal progressing pattern and predict whether sepsis onset will occur within a 24-hour window using data collected at night, when patient-provider ratios are higher due to cross-coverage resulting in limited observation to each patient. Moreover, we design a pre-training technique to alleviate the rare event problem of sepsis onset (i.e., the class imbalance problem). Our empirical study using data from a level 1 trauma center demonstrates the effectiveness of our proposed method.