Related Experiment Video
Updated: Aug 12, 2026
05:33
Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System
Published on: July 11, 2025
The Sydney Triage to Admission Risk Tool With Artificial Intelligence (START-AI) to Support Decision Making
Michael Dinh1,2, Elizabeth Corbett1,2, Thuy Truc Ngo3
1RPA Green Light Institute, Sydney Local Health District, Sydney, New South Wales, Australia.
Emergency Medicine Australasia : EMA
|July 29, 2026
Summary
The Sydney Triage to Admission Risk Tool with Artificial Intelligence (START-AI) model effectively predicts emergency department admissions. Blood tests, triage comments, and history notes significantly improve prediction accuracy, while vital signs have minimal impact.
Area of Science:
- Artificial Intelligence in Healthcare
- Clinical Decision Support Systems
- Emergency Medicine Analytics
Background:
- The Sydney Triage to Admission Risk Tool with Artificial Intelligence (START-AI) is a novel model for predicting inpatient admission from the Emergency Department (ED).
- Understanding feature importance is crucial for refining AI models and ensuring their effective clinical deployment.
Purpose of the Study:
- To evaluate the contribution of specific variables to the predictive performance of the START-AI model.
- To enhance the interpretability of the START-AI tool for clinical application.
Main Methods:
- A model explainability analysis was performed on electronic medical record data from a single ED over two years.
- The START-AI model, integrating ensemble machine learning and a transformer-based algorithm, was analyzed by sequentially adding features.
- Feature importance was quantified using feature permutation, change in Area Under the Receiver Operator Curve (AUROC), and SHapley Additive exPlanations (SHAP) analyses.
Main Results:
- The START-AI model achieved a high AUROC of 0.90, significantly improving upon the original START tool's AUROC of 0.78.
- Triage comments, ED case history notes, blood test results (especially lactate and C-reactive protein), and CT orders were key drivers of increased AUROC.
- Vital signs did not contribute to stepwise increases in AUROC, and feature importance was highest for blood test results, the START score, and C-reactive protein.
Conclusions:
- Model explainability analysis clarified the sequential and relative importance of features within the START-AI model.
- These insights are vital for the further development and clinical implementation of the START-AI tool in emergency departments.