Predicting the Length of Mechanical Ventilation in Acute Respiratory Disease Syndrome Using Machine Learning: The...

Area of Science:

• Critical Care Medicine
• Pulmonology
• Medical Informatics

Background:

• Clinician ability to predict long mechanical ventilation (MV) duration is limited.
• Moderate-to-severe acute respiratory distress syndrome (ARDS) often requires prolonged MV.
• Early prediction of MV duration is crucial for patient management and resource allocation.

Purpose of the Study:

• To assess the value of machine learning (ML) for early prediction of MV duration > 14 days in moderate-to-severe ARDS patients.
• To develop and validate ML models using clinical data captured at various time points post-ARDS diagnosis.
• To identify key prognostic factors influencing prolonged MV duration.

Main Methods:

• Development, testing, and external validation of ML models (logistic regression, Multilayer Perceptron, SVM, Random Forest).
• Utilized data from 1173 patients with moderate-to-severe ARDS on MV ≥ 3 days.
• Features captured at ARDS diagnosis, 24h, and 72h post-diagnosis were analyzed.

Main Results:

• The best prediction model utilized data from 72 hours post-ARDS diagnosis.
• Multilayer Perceptron identified PaO2/FiO2, PaCO2, pH, and PEEP as major prognostic factors.
• Model showed modest discrimination for predicting MV > 14 days (AUC 0.71).

Conclusions:

• Early prediction of prolonged MV in moderate-to-severe ARDS remains difficult, even with advanced ML techniques.
• Further research is required to discover novel markers for predicting MV length.
• The study highlights the complexity of predicting ventilation duration in ARDS patients.