Using Machine Learning to Risk Stratify Emergency Department Patients With Chest Pain but No Acute Myocardial...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Informatics

Background:

• Emergency department (ED) patients with chest pain, after acute myocardial infarction (MI) exclusion, remain at risk for adverse cardiac events.
• Accurate risk stratification is crucial for these patients to guide further management and prevent adverse outcomes.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting 30-day major adverse cardiac events (MACE) in ED patients presenting with chest pain and no acute MI.
• To assess the model's performance using key clinical and biomarker data.

Main Methods:

• A retrospective analysis of 14,177 adult patients presenting with chest pain to 5 Texas hospitals (2021-2024) was conducted.
• A long short-term memory (LSTM) algorithm was employed to build the predictive model.
• The model utilized patient demographics, vital signs, ECG findings, and serial high-sensitivity cardiac troponin (hs-cTn) levels.

Main Results:

• The study identified 30-day MACE in 0.3% of patients (myocardial infarction in 0.1%, all-cause mortality in 0.2%).
• The LSTM model demonstrated strong predictive performance for 30-day MACE (AUC, 0.884), myocardial infarction (AUC, 0.963), and all-cause mortality (AUC, 0.849).
• High-sensitivity cardiac troponin levels above the 99th percentile were observed in 3.8% of patients.

Conclusions:

• The developed LSTM model accurately predicts 30-day MACE in emergency department patients with chest pain and no acute myocardial infarction.
• The model's ability to integrate diverse data points, including serial hs-cTn levels, offers a promising tool for risk stratification.
• This ML approach can enhance clinical decision-making for chest pain patients in the ED setting.