Continuous and automatic mortality risk prediction using vital signs in the intensive care unit: a hybrid neural network approach

Stephanie Baker,Ian Atkinson,Wei Xiang

doi:10.1038/s41598-020-78184-7

Stephanie Baker, Ian Atkinson + Show 1 more

Open Access

https://doi.org/10.1038/s41598-020-78184-7

Copy DOI

Journal: Scientific Reports	Publication Date: Dec 1, 2020
Citations: 23	License type: open-access

Affiliation: James Cook University

Abstract

Mortality risk prediction can greatly improve the utilization of resources in intensive care units (ICUs). Existing schemes in ICUs today require laborious manual input of many complex parameters. In this work, we present a scheme that uses variations in vital signs over a 24-h period to make mortality risk assessments for 3-day, 7-day, and 14-day windows. We develop a hybrid neural network model that combines convolutional (CNN) layers with bidirectional long short-term memory (BiLSTM) to predict mortality from statistics describing the variation of heart rate, blood pressure, respiratory rate, blood oxygen levels, and temperature. Our scheme performs strongly compared to state-of-the-art schemes in the literature for mortality prediction, with our highest-performing model achieving an area under the receiver-operator curve of 0.884. We conclude that the use of a hybrid CNN-BiLSTM network is highly effective in determining mortality risk for the 3, 7, and 14 day windows from vital signs. As vital signs are routinely recorded, in many cases automatically, our scheme could be implemented such that highly accurate mortality risk could be predicted continuously and automatically, reducing the burden on healthcare providers and improving patient outcomes.

Highlights

The authors considered using vital signs in combination with other parameters, achieving a much higher area under the receiver-operator curve (AUROC) of 0.84 when vitals data was combined with the Glasgow Coma Score (GCS), Simplified Acute Physiology Score (SAPS)-II score, patient demographics and information obtained about the patient during their hospital stay prior to intensive care units (ICUs) admission
Artificial Intelligence Mortality Score (AIMS) scheme, a mortality risk classifier based on a hybridized CNNLSTM network that uses only age, gender, and statistical parameters derived from a 24-h window of vital sign measurements as features
This cross-validation confirms that the results obtained from the AIMS-3, AIMS-7, and AIMS-14 models are a realistic representation of how the network would perform in reality

Summary

Introduction

The authors considered using vital signs in combination with other parameters, achieving a much higher AUROC of 0.84 when vitals data was combined with the GCS, SAPS-II score, patient demographics and information obtained about the patient during their hospital stay prior to ICU admission. While there are advantages of early mortality risk prediction, this method is inflexible and does not consider how a patient might respond to treatments after ICU admission. This was attempted by Alvis et al.[15], with their scheme achieving an AUROC of 0.836 when predicting ICU mortality from a 48-h window of features including vital signs and laboratory values.

Results

Conclusion