Machine Learning for Myocardial Infarction Compared With Guideline-Recommended Diagnostic Pathways.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Acute Coronary Syndrome (ACS) diagnosis relies on clinical assessment and cardiac troponin measurements.
• Current diagnostic pathways for myocardial infarction (MI) use fixed time points and thresholds, which may not be optimal.
• The Collaboration for the Diagnosis and Evaluation of Acute Coronary Syndrome (CoDE-ACS) is a machine learning tool for MI probability assessment.

Purpose of the Study:

• To evaluate the diagnostic performance of the CoDE-ACS tool at different serial cardiac troponin measurement time points.
• To compare CoDE-ACS performance with guideline-recommended diagnostic pathways for MI.
• To assess the effectiveness of CoDE-ACS in identifying low- and high-probability scores for type 1 MI.

Main Methods:

• A prospective study enrolled 4105 patients with suspected MI across 12 sites in 5 countries.
• Serial high-sensitivity cardiac troponin I measurements were taken at 0, 1, and 2 hours.
• Diagnostic performance was assessed for CoDE-ACS against European Society of Cardiology (ESC) and High-STEACS pathways.

Main Results:

• CoDE-ACS identified 56% of patients as low probability at presentation (NPV 99.7%, sensitivity 99.0%), ruling out more patients than ESC 0-hour and High-STEACS pathways.
• At 1 or 2 hours, CoDE-ACS identified 65%-68% as low probability (NPV 99.7%) and 18%-19% as high probability (PPV 64.9%-68.8%).
• Guideline pathways identified fewer low-risk patients (49%-71%) with varying PPVs and NPVs, leaving more patients for further observation.

Conclusions:

• CoDE-ACS demonstrates consistent performance irrespective of serial troponin measurement timing.
• The tool identifies a greater proportion of low-probability patients compared to guideline-recommended pathways.
• Prospective evaluation is needed to determine if probability-guided care improves early MI diagnosis.