Towards early sepsis detection from measurements at the general ward through deep learning (Preprint)

Abstract

Sepsis is one of the leading causes of death in the hospital. Several warning scores have been developed to categorize patients’ degrees of illness, with the purpose of recognizing sepsis development at an early stage and consequently reducing time before starting treatment. The most accurate classification method, known as the SOFA score, is developed for use in the intensive care unit (ICU).

Sepsis is not exclusively developing in the ICU and may occur in any hospitalized patient. Therefore, a method for sepsis recognition outside the ICU is of major importance.

Recently, the use of computational methods has been proposed for early sepsis prediction. Multiple sepsis classifiers have been devised using machine learning methods. We validated the linear classification model devised by Calvert et al. and improved upon it using a deep neural network trained on data from the MIMIC-III database.

The reference model based on Calvert et al. approach yielded an AUROC of 0.81 for a 3-hour prediction time. The deep neural network outperformed the linear model, reaching an AUROC of 0.85 for a 3-hour prediction time.

Our results are comparable to the high-resolution model derived by Nemati et al. yet using only 8 simple and commonly performed measurements, instead of the complex set of 65 measurements leveraged by Nemati et al. Therefore, sepsis prediction may also be viable in less monitored environments in the hospital, such as the general ward and the emergency room.