Multicenter Development and Prospective Validation of eCARTv5: A Gradient Boosted Machine Learning Early Warning...

Area of Science:

• Clinical informatics
• Machine learning in healthcare
• Patient monitoring

Background:

• Early detection of clinical deterioration is crucial for improving patient outcomes.
• Existing early warning scores often use logistic regression and lack robust validation, especially in subgroups.
• There is a need for advanced machine learning models for accurate and reliable identification of patient decline.

Purpose of the Study:

• To develop and prospectively validate a gradient boosted machine model, eCARTv5, for identifying clinical deterioration in hospitalized patients.
• To compare the performance of eCARTv5 against established scores like MEWS and NEWS.
• To ensure the model's effectiveness across diverse patient demographics and clinical conditions.

Main Methods:

• A multicenter observational study involving retrospective and prospective data from adult inpatient admissions.
• Development and validation using a gradient boosted trees algorithm on demographics, vital signs, documentation, and laboratory values.
• Comparison of eCART with Modified Early Warning Score (MEWS) and National Early Warning Score (NEWS) using Area Under the Receiver Operating Characteristic Curve (AUROC).

Main Results:

• The eCART model demonstrated superior performance in retrospective validation with an AUROC of 0.835, compared to NEWS (0.766) and MEWS (0.704).
• eCART maintained high performance (AUROC ≥0.80) during prospective validation and across various patient subgroups.
• The model accurately predicted intensive care unit transfer or death within 24 hours.

Conclusions:

• The developed eCART model significantly outperformed NEWS and MEWS in both retrospective and prospective validations.
• eCART's consistent performance across diverse patient populations supports its utility in clinical settings.
• These findings paved the way for FDA clearance, enabling eCART's use in identifying deterioration among hospitalized ward patients.