Prehospital triage of acute aortic syndrome using a machine learning algorithm

Area of Science:

• Cardiology
• Emergency Medicine
• Data Science

Background:

• Acute aortic syndrome (AAS) is a life-threatening condition requiring immediate specialist care.
• Current prehospital triage for AAS faces challenges with high rates of overtriage and undertriage.

Purpose of the Study:

• To develop and validate a predictive algorithm for prehospital identification of AAS.
• To improve the accuracy of emergency medical services in triaging patients with suspected AAS.

Main Methods:

• Prospective data collection from a regional specialist aortic network.
• Development of two prediction algorithms: logistic regression and SuperLearner (SL) machine learning.
• Evaluation of algorithms based on undertriage and overtriage rates and ROC curve analysis.

Main Results:

• The SuperLearner model demonstrated superior performance (ROC AUC 0.87) compared to logistic regression (ROC AUC 0.68).
• SL significantly reduced undertriage to 1.0% and overtriage to 30.2% compared to logistic regression (undertriage 33.7%, overtriage 7.2%).
• The study included 976 hospital admissions, with 609 confirmed AAS cases.

Conclusions:

• Machine learning, specifically the SuperLearner algorithm, offers a robust tool for prehospital AAS prediction.
• This model can enhance clinical decision-making in emergency triage, potentially improving patient outcomes.
• Accurate prehospital triage is crucial for timely intervention in acute aortic syndrome.