Predicting Need for Vasopressors in the Intensive Care Unit Using an Attention Based Deep Learning Model

Abstract BackgroundPrevious models on prediction of shock mostly focused on septic shock and often required laboratory results in their models. Database analyses often report only limited data pre-processing which may introduce significant bias into machine classifiers. The purpose of this study was to use deep learning approaches to predict vasopressor requirement for critically ill patients within 24 hours of ICU admission using only vital signs. Performance bias from relative missing data and inadequate matching was assessed. MethodsWe used data from the Medical Information Mart for Intensive Care III database and the eICU Collaborative Research Database to develop a vasopressor prediction model. We performed systematic data pre-processing using matching of cohorts, oversampling and imputation to control for bias, class imbalance and missing data. After pre-processing we used bidirectional long short-term memory (Bi-LSTM), a multivariate time series model to predict the need for vasopressor therapy using serial physiological data collected 21 hours prior to prediction time. ResultsUsing data from 10,941 critically ill patients from 209 ICUs, our Bi-LSTM model achieved an initial area under the curve (AUC) of 0.96 (95%CI 0.96-0.96) to predict the need for vasopressor therapy in 2 hours within the first day of ICU admission. After matching to control class imbalance, the Bi-LSTM model had AUC of 0.83 (95%CI 0.82-0.83). Heart rate, respiratory rate and mean arterial pressure contributed most to the model amongst other serial physiological variables of systolic blood pressure, diastolic blood pressure, pulse oximetry and temperature. ConclusionsWe used Bi-LSTM to develop a model to predict the need for vasopressor for critically ill patients for the first 24 hours of ICU admission. With attention mechanism, respiratory rate, mean arterial pressure and heart rate were identified as key sequential determinants of vasopressor requirements. Although rigorous data pre-processing such as missing value analysis and class matching reduced predictive performance, it minimized bias in data and should be performed for database studies..

Medienart:	Preprint

Erscheinungsjahr:	2020
Erschienen:	2020

Enthalten in:	ResearchSquare.com - (2020) vom: 22. Juli Zur Gesamtaufnahme - year:2020

Sprache:	Englisch

Beteiligte Personen:	Kwak, Gloria Hyunjung [VerfasserIn] Ling, Lowell [VerfasserIn] Hui, Pan [VerfasserIn]

Links:	Volltext [kostenfrei]

doi:	10.21203/rs.3.rs-45876/v1

funding:
Förderinstitution / Projekttitel:

PPN (Katalog-ID):	XRA034197214