This article requires a subscription to view the full text. If you have a subscription you may use the login form below to view the article. Access to this article can also be purchased.
Abstract
Background Noninvasive respiratory support is increasingly used to support patients with acute respiratory failure. However, noninvasive support failure may worsen outcomes compared to primary support with invasive mechanical ventilation. Therefore, there is a need to identify patients where noninvasive respiratory support is failing so that treatment can be reassessed and adjusted. The objective of this study was to develop and evaluate three recurrent neural network models to predict noninvasive respiratory support failure.
Methods This is a cross-sectional observational study to evaluate the ability of deep recurrent neural network models (long short-term memory, gated recurrent unit, and gated recurrent unit with trainable decay) to predict failure of noninvasive respiratory support. Data were extracted from electronic health records from all adult (≥ 18 years) patient records requiring any type of oxygen therapy or mechanical ventilation between November 1, 2013, and September 30, 2020 across 46 ICUs in the Southwest United States in a single healthcare network. Input variables for each model included serum chloride, creatinine, albumin, respiratory rate, heart rate, pulse oximetry oxygen saturation (SpO2), fraction of inspired oxygen (FiO2), oxygen saturation (SaO2), and two measurements each (point-of-care and laboratory measurement) of partial pressure of carbon dioxide and partial pressure of arterial oxygen from an arterial blood gas.
Results Time series data from electronic health records were available for 22,075 patients. The highest accuracy and area under the receiver operating characteristic curve were for the long short-term memory model (94.04% and 0.9636, respectively). Accurate predictions were made 12 hours after ICU admission and performance remained high well in advance of noninvasive respiratory support failure.
Conclusion Recurrent neural network models using routinely collected time-series data can accurately predict noninvasive respiratory support failure well before intubation. This lead time may provide an opportunity to intervene to optimize patient outcomes.
Footnotes
- Corresponding Author: Jarrod M. Mosier, MD FCCM, Department of Emergency Medicine, 1501 N., Campbell Ave., AHSL 4171D, PO Box 245057, Tucson, AZ 85724–5057, Phone: 520–626-2038, jmosier{at}aemrc.arizona.edu
- Received June 25, 2022.
- Accepted October 8, 2022.
- Copyright © 2022 by Daedalus Enterprises
Pay Per Article - You may access this article (from the computer you are currently using) for 1 day for US$30.00
Regain Access - You can regain access to a recent Pay per Article purchase if your access period has not yet expired.