Prediction of the Spontaneous Breathing Test Success Using Biosignal and Biomarker in Critical Care Unit by a Machine Learning Approach

Centre Hospitalier Universitaire de Nice

Status: Enrolling

Conditions: Weaning From Mechanical Ventilation in Care Unit

Treatments: Other: Spontaneous ventilation test

Study type: Observational

Funder types: Other

Identifiers

NCT05886803

23Rea01

Details and patient eligibility

About

Context:

Several authors have been interested in applying Artificial Intelligence (AI) to medicine, using various Machine Learning (ML) techniques: managing septic shock, predicting renal failure... [1, 2] AI has an important place in decision support for clinicians [3]. The weaning period is a really important time in the management of a patient on mechanical ventilation and can take up to half of the time spent in intensive care unit. The first weaning attempt is unsuccessful in 20% of patients However, mortality can be as high as 38% in patients with the most difficult weaning [4]. Only a few studies have looked at the application of machine learning in this area, and only one has looked at the use of biosignals (cardiac rate, ECG, ventilatory parameters...) [5-7]. To improve morbidity, mortality and reduce length of stay, it is essential to be able to predict the success of the spontaneous breathing test and extubation.

Investigators propose to develop a predictive algorithm for the success of a ventilatory weaning test based on biosignal records and others features.

Methods:

It is a critical care, oligo-centric and retrospective study the investigators included biosignal variables extracted from the electronic medical record, such as respiratory (RR, minute volume...), cardiac (systolic pressure, heart rate...), ventilator parameters and other discrete variables (age, comorbidity...). Most biosignal variables are minute-by-minute records. Recording starts 48 hours before the test and stops at the start of the weaning test. The investigators extracted features from these records, combined them with other biomarkers, and applied several machine learning algorithms: Logistic Regression, Random Forest Classifier, Support Vector Classifier (SVC), XGBoost, and Light Gradient Boosting Method (LGBM)...

Enrollment: 500 estimated patients

Sex: All

Volunteers: No Healthy Volunteers

Inclusion criteria

• Computerized health report (CHR)
• Spontaneous breathing test should have been performed

Exclusion criteria

• Spontaneous breathing test has not been performed,
• Biosignal (cardiac, respiratory) are not registered in the CHR
• Patient died before the spontaneous breathing test
• Opposition to the study has been expressed.

Trial design

500 participants in 2 patient groups

Spontaneous Breathing Test

Description:

The first group will be composed only by patients admitted in intensive care/critical care for ventilation support, and who successed the spontaneous breathing test.

Treatment:

Other: Spontaneous ventilation test

Non Spontaneous Breathing Test

Description:

The second group will be composed only by patients admitted in intensive care/critical care for ventilation support, and who failed the spontaneous breathing test.

Treatment:

Other: Spontaneous ventilation test