Related Experiment Video
Updated: May 5, 2026
07:31
Implementation of a Real-Time Psychosis Risk Detection and Alerting System Based on Electronic Health Records using CogStack
Published on: May 15, 2020
8.3K
Multicenter Development and Prospective Validation of eCARTv5: A Gradient-Boosted Machine-Learning Early Warning
Matthew M Churpek1,2, Kyle A Carey3, Ashley Snyder4
1Department of Medicine, University of Wisconsin-Madison, Madison, WI.
Critical Care Explorations
|March 28, 2025
Summary
A new machine learning model, eCARTv5, effectively identifies patients at risk of clinical deterioration, outperforming existing early warning scores in retrospective and prospective validation. This advance aids in timely interventions for hospitalized patients.
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 rely on logistic regression and lack rigorous subgroup validation.
- There is a need for advanced machine learning models for timely identification of at-risk patients.
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.
Main Methods:
- A gradient-boosted trees algorithm was used with predictor variables including demographics, vital signs, documentation, and laboratory values.
- The study involved a large derivation cohort (901,491 admissions) and extensive external validation cohorts (retrospective: 1,769,461; prospective: 205,946 admissions).
- Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC) and compared to MEWS, NEWS, and eCARTv2.
Main Results:
- eCARTv5 demonstrated superior performance with the highest AUROC (0.834) in retrospective validation compared to eCARTv2 (0.775), NEWS (0.766), and MEWS (0.704).
- The model maintained high performance (AUROC ≥0.80) across diverse patient demographics and clinical conditions.
- Prospective validation confirmed eCARTv5's robust predictive capability for clinical deterioration.
Conclusions:
- eCARTv5 significantly outperformed existing early warning scores, including eCARTv2, NEWS, and MEWS, in both retrospective and prospective validation.
- The model's consistent performance across subgroups supports its reliability in identifying hospitalized patients at risk of deterioration.
- These findings led to FDA clearance for eCARTv5, enabling its clinical use in monitoring ward patients.
