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

Area of Science:

• Medical Informatics
• Clinical Decision Support Systems
• Machine Learning in Healthcare

Background:

• Early detection of clinical deterioration is crucial for improving patient outcomes.
• Existing machine learning early warning scores often lack rigorous validation and subgroup analysis.
• Traditional logistic regression models have limitations in capturing complex patient data.

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 early warning scores like MEWS and NEWS.
• To ensure the model's robustness across diverse patient populations and clinical scenarios.

Main Methods:

• Utilized a gradient-boosted trees algorithm with predictor variables including demographics, vital signs, documentation, and laboratory values.
• Developed the model (eCARTv5) using a large dataset from adult patients in inpatient medical-surgical wards (2006-2022).
• Externally validated the model retrospectively (2009-2023) and prospectively (2023-2024) across multiple health systems.

Main Results:

• eCARTv5 demonstrated superior performance with the highest Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.834 in retrospective validation.
• Outperformed eCARTv2 (0.775), NEWS (0.766), and MEWS (0.704) in retrospective validation.
• Maintained high performance (AUROC ≥0.80) across various patient demographics, clinical conditions, and in prospective validation.

Conclusions:

• The developed eCARTv5 model offers improved accuracy in identifying clinical deterioration compared to existing scores.
• eCARTv5's validated performance across diverse subgroups and prospective testing supports its clinical utility.
• The study's findings provided the basis for FDA clearance, enabling eCARTv5's use in clinical settings for hospitalized ward patients.