Performance of AI models in ECG interpretation and risk stratification for major adverse cardiac events in the emergency department

BackgroundArtificial intelligence (AI) may support electrocardiogram (ECG) interpretation and risk stratification in emergency departments, but its clinical utility remains uncertain.ObjectiveTo evaluate the performance of ChatGPT and Gemini in ECG interpretation and risk stratification for major adverse cardiac events (MACE).MethodsIn 89 prospective ED patients presenting with chest pain or palpitations, ECGs evaluated by AI models and emergency specialists were compared against cardiologist consensus. Cohen's kappa assessed agreement; diagnostic metrics were calculated for MACE risk stratification, and McNemar's test compared paired positive-call frequencies between the models.ResultsEmergency specialists showed high agreement with cardiologists (κ = 0.689-0.874), whereas ChatGPT (κ = 0.386) and Gemini (κ = 0.224) demonstrated lower agreement. McNemar testing revealed Gemini had a significantly higher positive-call tendency for ischemic features (ST elevation, ST depression, T-wave inversion; p < 0.05) than ChatGPT. For 30-day MACE risk stratification (n = 27, 30.3%), ChatGPT (96.3% sensitivity, 85.4% accuracy) and Gemini (92.6% sensitivity, 79.8% accuracy) showed numerically higher performance for ChatGPT; the proportion of positive MACE risk calls did not differ significantly between models (McNemar's exact test, p = 0.55).ConclusionIn this single-center pilot cohort (n = 89), both AI models showed high sensitivity in an exploratory MACE risk-stratification analysis, with no significant difference in positive-call frequency; equivalence cannot be inferred. Both showed lower agreement with cardiologists than emergency specialists. These preliminary findings do not support rule-out use and, at most, suggest a supplementary decision-support role, requiring further multi-center validation.

Authors

Institutions

Publication Details

Journal
Technology and Health Care
Published
2026-11-18
DOI
https://doi.org/10.1177/09287329261509123
Primary Topic
Artificial Intelligence in Healthcare and Education
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Performance of AI models in ECG interpretation and risk stratification for major adverse cardiac events in the emergency department

Faruk Aydınyılmaz, Halil İbrahim Doru, Şenol Arslan, Nazım Onur Can et al.
7 links
Technology and Health Care
Artificial Intelligence in Healthcare and Education
article

Performance of AI models in ECG interpretation and risk stratification for major adverse cardiac events in the emergency department

Faruk Aydınyılmaz, Halil İbrahim Doru, Şenol Arslan, Nazım Onur Can, Kürşat Arslan, Abdulkadir ODACI, Abdullah Onur Oflaz, Ahmet Şenol
article en

Abstract

BackgroundArtificial intelligence (AI) may support electrocardiogram (ECG) interpretation and risk stratification in emergency departments, but its clinical utility remains uncertain.ObjectiveTo evaluate the performance of ChatGPT and Gemini in ECG interpretation and risk stratification for major adverse cardiac events (MACE).MethodsIn 89 prospective ED patients presenting with chest pain or palpitations, ECGs evaluated by AI models and emergency specialists were compared against cardiologist consensus. Cohen's kappa assessed agreement; diagnostic metrics were calculated for MACE risk stratification, and McNemar's test compared paired positive-call frequencies between the models.ResultsEmergency specialists showed high agreement with cardiologists (κ = 0.689-0.874), whereas ChatGPT (κ = 0.386) and Gemini (κ = 0.224) demonstrated lower agreement. McNemar testing revealed Gemini had a significantly higher positive-call tendency for ischemic features (ST elevation, ST depression, T-wave inversion; p < 0.05) than ChatGPT. For 30-day MACE risk stratification (n = 27, 30.3%), ChatGPT (96.3% sensitivity, 85.4% accuracy) and Gemini (92.6% sensitivity, 79.8% accuracy) showed numerically higher performance for ChatGPT; the proportion of positive MACE risk calls did not differ significantly between models (McNemar's exact test, p = 0.55).ConclusionIn this single-center pilot cohort (n = 89), both AI models showed high sensitivity in an exploratory MACE risk-stratification analysis, with no significant difference in positive-call frequency; equivalence cannot be inferred. Both showed lower agreement with cardiologists than emergency specialists. These preliminary findings do not support rule-out use and, at most, suggest a supplementary decision-support role, requiring further multi-center validation.

Technology and Health Care
Ministry of Health (TR), Dokuz Eylül University (TR), Erzurum Regional Training and Research Hospital (TR)
Good health and well-being
Artificial Intelligence in Healthcare and Education
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon