Novel criteria to classify ARDS severity using a machine learning approach.

Area of Science:

• Critical Care Medicine
• Pulmonary Medicine
• Biomedical Engineering

Background:

• Arterial oxygenation in Acute Respiratory Distress Syndrome (ARDS) often improves with increased positive end-expiratory pressure (PEEP).
• Current ARDS severity definitions have limitations in accurately assessing patient status.
• Machine learning (ML) offers novel approaches to analyze complex clinical data.

Purpose of the Study:

• To introduce and validate a new variable, P/FP_E (PaO₂/(FiO₂xPEEP)), for assessing ARDS severity.
• To address the gap in the Berlin criteria for ARDS severity classification.
• To utilize ML to predict ARDS severity over time using the novel P/FP_E ratio.

Main Methods:

• Examined P/FP_E values to define boundaries for mild, moderate, and severe ARDS.
• Applied ML (LightGBM, RF, XGBoost) to predict ARDS severity using data from MIMIC-III and eICU databases.
• Tracked disease progression over the first 3 ICU days post-ARDS onset for severity assessment.

Main Results:

• The P/FP_E ratio demonstrated superior performance over the PaO₂/FiO₂ ratio in all ML models for predicting ARDS severity.
• ML models using P/FP_E achieved higher Area Under the Curve (AUC) and Correlation (CORR) values in both MIMIC-III and eICU datasets.
• The novel ratio effectively predicted ARDS severity progression over time.

Conclusions:

• The P/FP_E ratio is a more effective tool for assessing ARDS severity over time compared to current PaO₂/FiO₂ criteria.
• Implementing the P/FP_E ratio can lead to more precise therapeutic regimens for ARDS patients.
• This novel approach enhances clinical decision-making in managing ARDS severity.