Predicting Duration of Mechanical Ventilation in Acute Respiratory Distress Syndrome Using Supervised Machine...

Area of Science:

• Critical Care Medicine
• Data Science
• Pulmonology

Background:

• Acute respiratory distress syndrome (ARDS) is a severe inflammatory lung condition.
• Mechanical ventilation (MV) is critical for most ARDS patients.
• Predicting MV duration is crucial for resource management but remains challenging.

Purpose of the Study:

• To develop and validate machine learning (ML) models for early prediction of MV duration in ARDS patients.
• To identify the optimal time point within the first two intensive care unit (ICU) days for predicting MV duration.

Main Methods:

• Utilized supervised ML techniques (LightGBM, RF, XGBoost) on data from the MIMIC-III database.
• Extracted patient data from the first three ICU days post-ARDS diagnosis for model training.
• Externally validated models using the eICU database.

Main Results:

• The LightGBM model demonstrated the best performance in predicting MV duration.
• Early prediction using data from the second ICU day yielded the most accurate results.
• The model achieved a root mean square error (RMSE) of 6.10-6.41 days in MIMIC-III and 5.87-6.08 days in eICU.

Conclusions:

• Supervised ML models enable early and accurate prediction of MV duration in ARDS patients.
• These models can significantly improve ICU resource allocation and reduce healthcare costs.
• Early prediction facilitates proactive patient management and resource planning.