Developing and evaluating a machine-learning-based algorithm to predict the incidence and severity of ARDS with...

Area of Science:

• Critical Care Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Acute Respiratory Distress Syndrome (ARDS) has a high mortality rate, necessitating personalized treatment strategies.
• Early prediction of ARDS is crucial for timely intervention, including targeted drug therapy and mechanical ventilation.
• Current prediction methods may not be sufficiently timely or user-friendly for widespread clinical application.

Purpose of the Study:

• To develop and evaluate a novel dynamic prediction machine learning model for ARDS incidence and severity.
• To assess the model's performance using only continuous, non-invasive parameters obtained from standard patient monitors and ventilators.
• To establish a user-friendly and timely tool for early ARDS detection in clinical settings.

Main Methods:

• Utilized a dataset of 4738 ICU patients from 159 hospitals, with data extracted from electronic medical records.
• Developed seven machine learning models trained on 28 hourly recorded, continuous, non-invasive parameters.
• Compared model performance against methods using complete parameter sets and the traditional oxygenation saturation index method.

Main Results:

• The model using continuous non-invasive parameters achieved an Area Under the Curve (AUC) of 0.8691 for ARDS incidence and 0.7765 for severity.
• Prediction performance for both mild and severe ARDS demonstrated AUC values exceeding 0.85.
• Models utilizing only non-invasive parameters showed a marginal decrease in AUC (0.0133) compared to those using complete datasets.

Conclusions:

• A machine learning model for ARDS incidence and severity prediction was successfully developed using readily available, continuous non-invasive parameters.
• The developed method is convenient and user-friendly, suitable for integration into routine clinical practice.
• The model holds potential for application in pre-hospital settings for early ARDS warnings and interventions.