PDF
Citation
Print
Accurate differential diagnosis of acute chest pain in the emergency department (ED) is crucial. By identifying patients with acute coronary syndrome (ACS), timely emergent or urgent interventions can be determined, while excluding non-cardiogenic causes can prevent unnecessary invasive testing. Electrocardiography (ECG) is widely used as an initial test for acute chest pain due to its rapid and relatively simple assessment, providing baseline clinical information for early ACS evaluation. Recently, artificial intelligence (AI) has been applied to ECG analysis, extending beyond conventional rule-based interpretation to develop models that predict various cardiovascular diseases. Previously, we have also developed an AI-based quantitative ECG analysis (QCG system) and demonstrated its favorable performance in diagnosing coronary artery disease (CAD), including ST-segment elevation myocardial infarction (STEMI) and obstructive CAD.123456 Prospective studies and trials are expected to determine if such AI-based ECG analysis can genuinely improve patient management and outcomes. As evidence builds, this will support the integration of AI-based ECG into real-world clinical practice.7
In this context, before initiating a large-scale prospective trial, this pilot study aimed to evaluate whether the QCG system, an AI-based ECG analysis tool, could enhance the appropriateness of urgent coronary angiography (CAG) decisions for patients presenting with acute chest pain in the ED.
We reviewed consecutive patients aged ≥ 20 years who visited the ED at Seoul National University Bundang Hospital in 2021 for chest pain. A total of 6,224 patients were identified, and those who underwent CAG or cardiac computed tomography angiography (CCTA) to rule out ACS were selected. Patients with a history of myocardial infarction, coronary revascularization, or heart transplantation were excluded, as were those with significant bradyarrhythmia (heart rate < 40 bpm), tachyarrhythmia (heart rate > 150 bpm), or pacing support, resulting in 1,096 eligible patients. Two experts, one in cardiology (Cho Y) and one in emergency medicine (Kim J), performed a detailed medical review, categorizing patients into three groups: Group 1 included non-coronary chest pain patients with less than 50% diameter stenosis (DS) on CAG or CCTA and normal cardiac enzymes (n = 596); Group 2 included ACS patients with DS ≥ 50% but no occlusion (n = 243); and Group 3 included ACS patients with near-total or total obstruction (n = 99). From these, a final study population of 300 patients was randomly sampled: 150 from Group 1, and 75 each from Groups 2 and 3. Subsequently, the patients were exclusively categorized into three case subsets of 100 each, maintaining the group proportions.
The QCG system, a convolutional neural network-based analyzer, is approved by the Korean Ministry of Food and Drug Safety (MFDS) for AI-ECG applications.12 In brief, the system analyzes ECG images as input and provides risk scores for ACS (QCGACS) and STEMI (QCGSTEMI) on a normalized scale (0 to 100), with higher scores indicating greater likelihood of these conditions based on the analyzed ECG data. To examine the QCG system’s impact on clinical decision-making regarding urgent CAG for patients with acute chest pain, clinicians were recruited from the cardiology and emergency medicine departments. An open call led to the enlistment of six clinical fellows—three from cardiology and three from emergency medicine—actively involved in acute chest pain management in the ED. None of the participating clinicians had prior experience with QCG scores; thus, a brief introduction to QCG was given at the start, including the MFDS-approved cutoff values (19.7 for ACS and 6.6 for STEMI). Each physician was randomly assigned to one of the three case subsets, ensuring that both cardiology and emergency medicine departments evaluated each subset.
We employed a two-stage decision-making process. In the first stage, clinicians received baseline information, including demographics (age, sex, blood pressure), clinical risk factors (hypertension, diabetes, dyslipidemia, family history of cardiovascular disease, smoking, previous CAD, and recurrent angina), and initial ECG data (Step 1: clinical judgment). Based on this information, clinicians decided whether urgent CAG was necessary, within 24 hours of initial admission for acute chest pain. In the second stage, the QCGACS and QCGSTEMI scores were additionally provided (Step 2: clinical judgment + QCG), after which clinicians made a final decision on urgent CAG necessity. Appropriate clinical decisions were defined as withholding urgent CAG for Group 1 and proceeding with it for Groups 2 and 3. We utilized a generalized estimating equation model to assess significant changes in the proportion of appropriate decisions from Step 1 to Step 2, adjusting for baseline clinical data.8 Appropriate decisions for urgent CAG were also determined based solely on QCG scores (QCG-alone), deemed necessary if QCGACS or QCGSTEMI exceeded their respective cutoffs of 19.7 and 6.6. Subgroup analyses were performed for each specialty. Finally, diagnostic performance for ACS was compared across Step 1, Step 2, and the QCG-alone approach, defining an urgent CAG decision in ACS (Groups 2 and 3) as a correct ACS diagnosis. All statistical analyses were conducted using R software (version 4.1.1; R Foundation for Statistical Computing, Vienna, Austria), with statistical significance set at P < 0.05.
The detailed baseline characteristics of the study population are summarized in Supplementary Table 1 (median age 63 years; male 67%). The proportion of appropriate clinical decisions significantly increased from Step 1 to Step 2 in Group 2 (36.0% vs. 45.3%, P = 0.003) and Group 3 (85.3% vs. 90.0%, P = 0.017) but showed no significant difference in Group 1 (92.7% vs. 95.0%, P = 0.125). Notably, QCG-alone approach, solely based on QCG scores, yielded the highest proportion of appropriate decisions, outperforming Step 2 results (97.3% vs. 95.0%, P = 0.005 for Group 1; 58.7% vs. 45.3%, P < 0.001 for Group 2; and 97.3% vs. 90.0%, P < 0.001 for Group 3). When evaluating diagnostic performance for ACS, Step 1 achieved an area under the receiver operating characteristic curve (AUC) of 0.767, overall accuracy of 0.77, and F1 score of 0.72 (Table 1). Step 2 demonstrated improvements across all performance measures, achieving an AUC of 0.813, accuracy of 0.81, and F1 score of 0.78. The QCG-alone approach showed the highest diagnostic performance, reaching an AUC of 0.877, accuracy of 0.88, and F1 score of 0.86. Subgroup analyses by specialty consistently demonstrated increased appropriate decision-making from Step 1 to Step 2 in Group 2 and Group 3 (Fig. 1). Additionally, subgroup analysis stratified by individual clinicians yielded consistent results, with the QCG-assisted approach improving appropriate decision-making across both groups (Supplementary Fig. 1).
Previous studies have demonstrated the robust diagnostic capabilities of QCG for ACS and obstructive CAD123456; the current findings further indicate that a QCG-assisted approach could enhance clinical decision-making for urgent CAG in patients with acute chest pain. Notably, diagnostic performance was highest when decisions were based solely on QCG scores, highlighting the potential of QCG scores as valuable digital markers for rapid and accurate decision-making. Given that ECG is typically the initial test for acute chest pain, the QCG system may enhance diagnostic accuracy at this early stage, supporting clinicians in rapidly identifying ACS patients and aiding their clinical assessment. This improvement could, in turn, lead to timely downstream testing and treatment for ACS patients.
However, it should be acknowledged that the findings of the current study are limited by the retrospective nature of the data, which were derived from a single-center study. Additionally, the number of clinicians participating in the study was relatively small. Therefore, to expand the clinical utility and generalizability of the QCG-assisted approach, further validation in more diverse clinical settings and with a larger number of clinicians are warranted. In clinical practice, ECG changes in ACS are often dynamic and may evolve with the patient's clinical course. Therefore, future studies assessing whether the QCG-assisted approach can effectively support clinical decision-making when serial ECG data are incorporated would be valuable. Lastly, prospective trials are essential to validate the impact of the QCG-assisted approach on clinical decision-making and patient outcomes in real-world settings. Encouragingly, our institution has integrated the QCG system into the electronic medical record system, allowing QCG scores to be automatically included with ECG tests in the ED (Supplementary Fig. 2). Building on this environment, we are planning a prospective study and anticipate sharing future results that will further support the practical advantages of QCG-assisted clinical approaches.
Ethics statement
This study was approved by the Institutional Review Board (IRB) of Seoul National University Bundang Hospital (IRB No. B-2309-855-302). All participating clinicians provided written informed consent, but the patient’s consent was waived due to retrospective collection of de-identified clinical records.
XML Download