Related Experiment Video

Updated: Jul 16, 2025

Experimental Model to Evaluate Resolution of Pneumonia
09:49

Experimental Model to Evaluate Resolution of Pneumonia

Published on: February 17, 2023

1.3K

Long short-term memory model identifies ARDS and in-hospital mortality in both non-COVID-19 and COVID-19 cohort

Jen-Ting Chen1,2, Rahil Mehrizi3, Boudewijn Aasman4

  • 1Department of Medicine, Division of Pulmonary and Critical Care Medicine, UCSF, San Francisco, California, USA tina.chen@ucsf.edu.

BMJ Health & Care Informatics
|September 14, 2023
PubMed
Abstract

Insights

This study developed a Long Short-Term Memory (LSTM) framework to predict acute respiratory distress syndrome (ARDS) and in-hospital mortality in mechanically ventilated patients. The LSTM model offers timely risk identification for improved patient management.

Area of Science:

  • Critical Care Medicine
  • Artificial Intelligence in Healthcare
  • Respiratory Medicine

Background:

  • Mechanically ventilated (MV) patients face risks of acute respiratory distress syndrome (ARDS) and in-hospital mortality.
  • Early identification of these risks is crucial for timely intervention and improved patient outcomes.
  • Distinguishing risk factors and prediction models between COVID-19 and non-COVID-19 patient cohorts is essential.

Purpose of the Study:

  • To develop and validate a Long Short-Term Memory (LSTM) framework for predicting ARDS and in-hospital mortality.
  • To assess the performance of the LSTM model in both a non-COVID-19 mechanically ventilated cohort and a COVID-19 cohort.
  • To evaluate the timeliness of risk prediction for ARDS and mortality.

Main Methods:

  • Trained an LSTM model using 132 structured features on a large cohort of MV non-COVID-19 patients (n=3905).
  • Validated the LSTM model on a separate MV non-COVID-19 cohort and a distinct COVID-19 cohort (n=5672).
  • Employed active learning to enrich the non-COVID-19 patient dataset from 2016-2019.

Main Results:

  • The LSTM model achieved a score of 0.9 on the non-COVID-19 validation cohort with 86% sensitivity and 57% specificity for ARDS risk.
  • The model predicted ARDS risk 10 hours before onset and mortality 9.4 days prior in the non-COVID-19 cohort.
  • Performance on the COVID-19 cohort showed lower sensitivity (70%) and higher specificity (84%), predicting mortality 2.4 days before death.

Conclusions:

  • The developed LSTM algorithm accurately and promptly identifies the risk of ARDS or death in both non-COVID-19 and COVID-19 mechanically ventilated patients.
  • Timely risk alerts from the LSTM model can enhance the implementation of evidence-based ARDS management strategies.
  • Facilitating early goals-of-care discussions for high-risk patients is a key benefit of this predictive tool.
Keywords:
COVID-19Critical Care OutcomesDecision Support Systems, ClinicalMedical Informatics

More Related Videos

An Educational Video Demonstration of How to Prone a Critically Ill Intubated Patient07:16

An Educational Video Demonstration of How to Prone a Critically Ill Intubated Patient

Published on: November 30, 2022

3.3K
A Model of Self-limited Acute Lung Injury by Unilateral Intra-bronchial Acid Instillation00:07

A Model of Self-limited Acute Lung Injury by Unilateral Intra-bronchial Acid Instillation

Published on: August 30, 2019

9.0K

Related Concept Videos

Acute Respiratory Failure-II01:21

Acute Respiratory Failure-II

Type I Respiratory Failure, or hypoxemic respiratory failure, occurs when the partial pressure of oxygen (PaO2) in arterial blood falls below 60 mmHg while breathing room air without a corresponding increase in arterial carbon dioxide levels (PaCO2). This condition highlights a significant impairment in the lungs' capacity to oxygenate the blood.
The underlying physiological abnormalities that contribute to hypoxemic respiratory failure include:
269
Acute Respiratory Failure-I01:21

Acute Respiratory Failure-I

Acute respiratory failure is a condition characterized by the inability of the lungs to perform their primary function: gas exchange. This failure leads to insufficient oxygen levels (hypoxemia) in the blood, elevated carbon dioxide levels (hypercapnia), or both, causing critical impairment in organ function.
Definition: It is defined by specific criteria based on blood gas measurements. Hypoxemia happens when the partial pressure of oxygen (PaO2) falls below 60 mmHg. At the same time,...
237
Acute Respiratory Failure-IV01:23

Acute Respiratory Failure-IV

Respiratory failure can manifest suddenly or gradually, characterized by a rapid decline in PaO2 and a rapid rise in PaCO2. This situation indicates a severe respiratory problem that may quickly become a life-threatening emergency. One of the early signs of hypoxemic Acute Respiratory Failure (ARF) is a change in mental status due to the brain's sensitivity to oxygen levels and changes in acid-base balance. Symptoms such as restlessness, confusion, and agitation suggest inadequate oxygen...
176
Acute Respiratory Failure-III01:30

Acute Respiratory Failure-III

Hypercapnic respiratory failure, also known as Type 2 or ventilatory respiratory failure, is a severe condition characterized by the body's inability to effectively remove carbon dioxide (CO2) from the bloodstream. It leads to an arterial CO2 pressure (PaCO2) exceeding 45 mmHg and a blood pH above 7.35. This situation indicates that the body's ventilatory demand, or the ventilation needed to maintain normal PaCO2 levels, surpasses its supply or the maximum gas flow achievable without...
213
Acute Respiratory Failure-V01:29

Acute Respiratory Failure-V

The treatment for acute respiratory failure varies based on factors like the underlying cause, overall health, and severity. A collaborative healthcare team is essential for early detection, often through arterial blood gas analysis. Identifying the cause is the primary goal, with treatment strategies adjusted for ventilation/perfusion (V/Q) mismatch, shunting, or diffusion impairment.
Ensure that patients are monitored continuously for their response to therapy, including changes in...
159