Prehospital and Emergency Department STEMI Alerts: AI-based ECG Interpretation vs Physician Accuracy Assessment
Faculty Information
Name:
Dr. Brandon Allen
Email
brandonrallen@ufl.edu
Phone
(954) 675-4321
Faculty Department/Division
Emergency Medicine
This project is primarily:
Clinical
Research Project Description:
Research Project Description
Background:
Timely and accurate identification of ST-Elevation Myocardial Infarction (STEMI) remains critical for reducing morbidity and mortality. In both the prehospital and emergency department (ED) settings, ECG interpretation drives the activation of STEMI alerts, influencing rapid transfer to the cardiac catheterization lab. Despite standardized training and protocols, variability in human ECG interpretation leads to false activations and delayed recognition, particularly in patients with subtle or atypical findings.
Recent advances in artificial intelligence (AI)–based ECG interpretation tools, such as the Queen of Hearts app, have demonstrated high diagnostic accuracy in detecting acute coronary occlusion. These technologies may complement physician interpretation and improve diagnostic precision in STEMI alerts. This project aims to evaluate the comparative accuracy of AI-based ECG analysis versus physician interpretation in prehospital and ED STEMI alert activations, providing insight into opportunities for AI-assisted decision support in acute cardiovascular care.
Hypothesis:
AI-based ECG interpretation using the Queen of Hearts algorithm will demonstrate comparable or superior accuracy to physician interpretation in identifying true STEMI cases, while reducing false activations and missed diagnoses.
Methods:
Data Collection:
Conduct a retrospective review of all prehospital and ED STEMI alert activations across the UF Health system over a defined study period.
Collect ECGs, activation details, final adjudicated diagnoses, and relevant clinical outcomes.
Each ECG will be independently reviewed by:
A board-certified cardiologist
An emergency medicine physician
The Queen of Hearts AI algorithm
Accuracy Assessment:
Compare diagnostic accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of AI and human interpretations using the adjudicated STEMI outcome as the reference standard.
Perform subgroup analyses between prehospital and ED-originating alerts.
Data Analysis:
Use statistical tests (e.g., McNemar’s test, Cohen’s kappa, ROC curves) to assess agreement and comparative accuracy.
Conduct logistic regression to explore predictors of false activations or missed STEMI cases.
Role of Medical Student:
The medical student will be an integral part of the research team, gaining experience in cardiovascular data analysis and AI evaluation. Responsibilities include:
Data Management: Assist with data extraction, de-identification, and organization of ECG and clinical data.
Literature Review: Conduct a focused review of AI applications in ECG interpretation and prior validation studies.
Analytical Skills: Participate in inter-rater reliability analysis, ROC curve construction, and comparative accuracy modeling.
Manuscript Preparation: Contribute to drafting the methods and results sections for abstract and manuscript submission.
Funding and Publications:
This project leverages institutional data infrastructure and existing collaborations with AI developers. No external funding is required for data collection or analysis. Foundational work includes Allen BR et al., “Artificial intelligence–based ECG interpretation for rapid STEMI detection” (PubMed ID: 40763602
).
Does this project have an international component or travel?
No
AI-HEART: Retrospective Evaluation of Combined Artificial Intelligence ECG Interpretation and Delta Troponin Integration for Enhanced Acute Coronary Syndrome Risk Stratification
Faculty Information
Name:
Dr. Brandon Allen
Email
brandonrallen@ufl.edu
Phone
(954) 675-4321
Background:
Accurate early risk stratification of patients presenting to the emergency department (ED) with chest pain is essential for optimizing care and resource use. The HEART score is a validated tool for assessing the likelihood of acute coronary syndrome (ACS); however, it does not fully incorporate dynamic biomarker or electrocardiographic information. Recent work at UF Health (the GatorHEART study) demonstrated that integrating delta high-sensitivity troponin (hs-cTn) and limit-of-quantification (LOQ) subtraction criteria into the HEART score safely reclassified 41% of intermediate-risk patients to low-risk, with no adjudicated major adverse cardiac events (MACE).
In parallel, artificial intelligence (AI)–based ECG interpretation tools—such as the Queen of Hearts algorithm—have shown promising accuracy in detecting myocardial injury and ischemic changes comparable to expert readers. Combining AI-driven ECG analytics with biomarker dynamics may enhance diagnostic precision and reduce unnecessary admissions and testing.
This study will retrospectively assess the performance of a combined AI-HEART model that integrates the Queen of Hearts AI ECG interpretation output with the recalibrated GatorHEART score, aiming to create a multimodal risk-stratification tool for ED chest pain evaluation.
Hypothesis:
Integration of AI-derived ECG interpretation with delta hs-cTn data (AI-HEART model) will improve diagnostic accuracy for acute myocardial injury and 30-day MACE compared with the standard HEART and GatorHEART scores alone.
Methods:
Data Collection:
- Retrospective cohort of adult patients presenting to the UF Health ED with chest pain from June 2019 – July 2024.
- Inclusion: serial hs-cTnI testing and ECG data suitable for AI analysis using the Queen of Hearts platform.
- Collected variables: demographics, risk factors, serial troponin values (baseline and delta), AI-ECG outputs (STEMI probability, ischemic likelihood), HEART and GatorHEART components, and adjudicated outcomes (AMI diagnosis, coronary intervention, 30-day MACE).
Model Development and Analysis:
- Calculate HEART, GatorHEART, and combined AI-HEART scores.
- Compare diagnostic performance (sensitivity, specificity, positive and negative predictive values, and AUC from ROC curves).
- Assess reclassification improvement (NRI) and calibration across subgroups (sex, renal function, age ≥ 75).
- Perform secondary analysis of false activations and missed events using AI versus clinician interpretation.
Role of Medical Student:
The student will be a core member of the investigative team, developing research and data analytics skills.
Responsibilities include:
- Data Management: Assist in extracting, cleaning, and merging ECG, biomarker, and outcome data from institutional databases.
- AI Integration: Work with the research mentor to import and analyze AI-generated ECG interpretations.
- Statistical Analysis: Participate in building comparative models and generating ROC and NRI metrics under supervision.
- Manuscript Development: Contribute to drafting figures, methods, and preliminary results for abstract submission (UF Research Symposium, SAEM, or ACEP).
Funding and Prior Work:
Supported by departmental resources and building upon the prior GatorHEART project (Allen BR et al., “Integrating Delta Troponin and LOQ Subtraction into the HEART Score,” UF Health, 2024). No external funding is required for retrospective data analysis.