First-line risk stratification with machine learning models facilitates rapid triage for non-ST-elevation myocardial...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Diagnosing non-ST-elevation myocardial infarction (NSTEMI) promptly is challenging due to reliance on serial troponin tests, which can cause delays and emergency department overcrowding.
• Current diagnostic protocols for NSTEMI often involve repeated high-sensitivity cardiac troponin (hs-cTn) testing, leading to extended decision-making times and increased patient volume in emergency settings.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for early risk stratification of NSTEMI using initial routine laboratory tests.
• To create a clinically actionable algorithm based on ML-derived risk scores to support rapid NSTEMI diagnosis in emergency departments.
• To evaluate the ML model's performance against hs-cTn testing alone and in combination with existing early diagnostic protocols.

Main Methods:

• Retrospective analysis of 54,636 patients undergoing hs-cTn testing in emergency departments across Taiwan (May 2016-Dec 2021).
• Development of an ML model using demographics and 23 routine lab tests from the initial blood draw on 15,096 eligible patients.
• Internal and external validation of the ML model and assessment of a decision support algorithm with defined risk thresholds.

Main Results:

• The ML model demonstrated superior predictive performance compared to hs-cTn alone, evidenced by a higher area under the receiver-operating characteristic curve.
• The decision algorithm identified low-risk (<1.8) and high-risk (≥38.5) NSTEMI patient groups with a negative predictive value (NPV) of 98.8% and positive predictive value (PPV) of 78.1%.
• Combining the ML model with the 0 h/1 h algorithm enabled safe rule-in/out for 85.3% of patients within 1 hour, achieving an NPV of 100% and PPV of 84.9%.

Conclusions:

• The developed ML-based approach provides accurate NSTEMI prediction and an actionable tool for rapid, safe triage in emergency care settings.
• This novel strategy can significantly improve early decision-making for NSTEMI, potentially reducing emergency department wait times and improving patient outcomes.
• The integration of ML with routine lab tests offers a promising alternative or adjunct to current serial hs-cTn testing protocols for NSTEMI diagnosis.