Selective classification with machine learning uncertainty estimates improves ACS prediction: A retrospective study...

Area of Science:

• Cardiology and Emergency Medicine
• Artificial Intelligence in Healthcare
• Machine Learning for Clinical Decision Support

Background:

• Timely identification of acute coronary syndrome (ACS) in prehospital settings is crucial for minimizing myocardial damage.
• Existing machine learning models demonstrate insufficient accuracy for reliably ruling in or out ACS prehospital.
• A performance gap exists in current prehospital ACS diagnostic tools.

Purpose of the Study:

• To identify an improved machine learning method for accurate prehospital ACS classification.
• To evaluate the efficacy of ensemble gradient boosted decision trees (GBDT) and selective classification (SC) for ACS diagnosis.
• To bridge the performance gap in current prehospital ACS diagnostic capabilities.

Main Methods:

• Retrospective evaluation of GBDT and SC approaches on prehospital patient data.
• Analysis included consecutive patients with chest pain transported via ambulance to the emergency department.
• Utilized 23 prehospital covariates for ACS classification task.

Main Results:

• The combined GBDT+SC model demonstrated superior performance compared to existing methods.
• GBDT+SC improved sensitivity by 8% and specificity by 23% for ACS classification.
• The fused model offers enhanced safety for ruling in and out ACS prehospital.

Conclusions:

• The GBDT+SC fusion represents a significant advancement in prehospital ACS identification.
• This novel approach provides a safer and more accurate tool for emergency medical services.
• Improved diagnostic accuracy can lead to more timely and effective ACS treatments.