Machine Learning Models Versus the National Early Warning Score System for Predicting Deterioration: Retrospective...

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Healthcare
• Patient Monitoring

Background:

• Early warning score systems, like the National Early Warning Score (NEWS), aid clinical decision-making by identifying high-risk patients.
• Limited evidence exists on the reliability of these scores in the United Arab Emirates (UAE) patient population.

Purpose of the Study:

• To develop and validate a data-driven model for predicting in-hospital patient deterioration within a UAE inpatient cohort.
• To compare the performance of machine learning models against the established NEWS system.

Main Methods:

• Retrospective cohort study utilizing a real-world dataset from a large Abu Dhabi hospital (April 2015 - August 2021).
• Data included 16,901 unique patients and routine vital sign measurements.
• Machine learning models (logistic regression, gradient-boosting, neural network) were developed and evaluated against the NEWS system using Area Under the Receiver Operating Characteristic Curve (AUROC).

Main Results:

• The NEWS system achieved an AUROC of 0.682.
• Gradient-boosting and neural network models demonstrated superior performance with AUROC values of 0.778 and 0.756, respectively.
• Temperature and respiratory rate were identified as key predictors of deterioration.

Conclusions:

• Machine learning models offer a more accurate alternative to traditional early warning scores for predicting patient deterioration.
• Development and implementation of cohort-specific machine learning models are strongly recommended, especially for external patient populations.
• This approach can enhance clinical decision-making and improve patient outcomes in diverse healthcare settings.