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

Area of Science:

• Emergency Medicine
• Cardiology
• Artificial Intelligence in Healthcare

Background:

• Accurate prehospital identification of acute coronary syndrome (ACS) is crucial for timely myocardial reperfusion therapy.
• Current machine learning models for prehospital ACS diagnosis exhibit insufficient sensitivity and specificity.
• There is a need for improved diagnostic tools to safely rule-in or rule-out ACS in emergency medical services.

Purpose of the Study:

• To evaluate the performance of an ensemble of gradient boosted decision trees (GBDT) and selective classification (SC) for prehospital ACS detection.
• To determine if the fusion of GBDT and SC (GBDT+SC) offers superior diagnostic accuracy compared to existing methods.
• To assess the safety and efficacy of the GBDT+SC approach in a prehospital chest pain patient cohort.

Main Methods:

• Retrospective analysis of consecutive patients with chest pain or anginal equivalents transported by ambulance.
• Application of GBDT and SC models using 23 prehospital covariates for ACS classification.
• Evaluation of the combined GBDT+SC model performance against individual models and established benchmarks.

Main Results:

• The fused GBDT+SC model demonstrated an 8% improvement in sensitivity and a 23% improvement in specificity for ACS classification.
• This enhanced performance surpasses previously reported metrics for prehospital ACS identification.
• The GBDT+SC approach showed a significant improvement in diagnostic accuracy for ruling in and ruling out ACS.

Conclusions:

• The GBDT+SC fusion represents a significant advancement in prehospital ACS diagnostic capabilities.
• This novel machine learning approach offers a safer and more accurate method for emergency medical services to manage patients with chest pain.
• Implementing GBDT+SC can lead to more timely and appropriate interventions, potentially reducing myocardial damage.